Information overload ("IO") is a common term to most managers, but few of us have ever stopped to consider what the term really means and what impact it may have on our decision-making processes. Professor Peter Gordon Roetzel has studied IO extensively (and especially in the context of decision-support systems) and defines it thus:

Information overload is a state in which a decision-maker faces a set of information (i.e., an information load with informational characteristics such as an amount, a complexity, and a level of redundancy, contradiction and inconsistency) comprising the accumulation of individual informational cues of differing size and complexity that inhibit the decision maker's ability to optimally determine the best possible decision. The suboptimal use of information is caused by the limitation of scarce individual resources. A scarce resource can be limited individual characteristics (such as serial processing ability, limited short-term memory) or limited task-related equipment (e.g., time to make a decision, budget).

There is a lot to unpack in Roetzel's statement, so let's take it one step at a time.

The first part of his definition posits that IO arises when the information that a given individual receives crosses certain thresholds that make it difficult, if not impossible, to process it correctly. These thresholds, in his view, are as follows:

• Quantity

• Complexity

• Redundancy

• Contradiction

• Inconsistency

Roetzel does not claim the list to be exhaustive, and we can add at least three other factors that could trigger IO:

• Velocity, i.e., the rate of obsolescence of the information makes it difficult or even impossible to process

• Accessibility, i.e., the information can only be processed through the acquisition of specialized skills

• Bias, i.e., information is consciously (or unconsciously) presented to drive a specific outcome

Whether considering technical innovations like distributed ledgers or social movements for environmental justice, business leaders are surrounded by issues that, unless addressed with care and intelligence, will trigger IO within even the most sophisticated executive.

One might be tempted to assume that IO is simply a "fact of life" and can be ignored, but that's hardly the case for a reason we can understand intuitively, as illustrated in Figure 1.

Figure 1: Decision-making performance at and beyond the IO threshold. (Source: Peter Roetzel)

The figure illustrates a critical point: that as information loads approach the point at which they can no longer be processed effectively (for whatever reason), decision-making performance begins to deteriorate. This change in performance occurs because the decision-maker is able to process less and less of the total information available to help make good decisions. In other words, imagine a scenario in which making the right choice requires an executive to process ten pieces of information, the maximum amount manageable by this particular leader. Suddenly, because of a competitor’s major innovation, the necessary information load increases from ten to twenty pieces. In this case, the decision-maker who was maxed out at ten analytical tasks (100%) is forced to shift and make a decision at only 50% capacity. As the required information count climbs higher, the processing capacity percentage drops, thus generating a greater probability of making a bad choice.

Decision-makers in the situation described above experience what the American executive Herbert Simon described as "a wealth of information [that] creates a poverty of attention." This attention-deficit creates "a need to allocate that attention efficiently among the overabundance of information sources that might consume it." 

There is no senior executive who has not faced this dilemma. Perhaps an issue was manageable for years until a disruptive competitor suddenly created the need for more information to properly manage it. Or an entirely new social or environmental phenomenon appears and, because of its novelty, executives must scramble to gather and process the information needed to respect this new force when making decisions. In either case, managers are confronted with a high-stakes (and often exasperating) challenge to "get up to speed" quickly and, even more critically, correctly.

Addressing the challenge inadequately is all too easy, of course. Executives can rely on poor information sources, or they can waste months or even years of effort analyzing the wrong information. Or they can also process brand new content with out-of-date skill sets, only to find out later that their analyses missed critical elements and so drew incorrect conclusions. Roetzl notes that in these situations, managers can harm not only themselves but others:

Users seem to ignore possible side effects of information overload up to a very high level before retreating from these channels or platforms. From a bird's eye perspective, this situation might be compared with the spread of a disease. Thus, people often act irrationally by infecting others (i.e., sending more messages, likes, news to other members of their network) instead of sparing themselves (i.e., making a rest/recovery from their overloaded status).

The biological metaphor is not unwarranted. Indeed, Edward Hallowell, a psychiatrist and expert on attention-deficit disorders, has observed what he calls an "attention deficit trait" in managers that presents similarly to the medical condition often seen in children and adults. Author Linda Stone, who coined the term "continuous partial attention," has noted that the inability to process something as simple as an e-mail inbox can lead to what she calls e-mail apnea: "The unconscious suspension of regular and steady breathing when people tackle their e-mail."

While the sheer volume of information is the main driver of IO, there are other factors at play as well. Researchers have noted that the trend over the last few decades to flatten organizations has increased the number of direct reports executives are forced to manage. Indeed, there are CEOs who now manage over a dozen individual executives directly, each overseeing a complex function. In turn, each direct report creates yet an additional information flow to process, compounding the organizational information overload facing any executive arising from her own functional responsibilities.

IO Under Negative Escalation

Intriguingly, there may be one other IO-related phenomenon at play in the examples above, and it is related to how IO affects decision-makers when their situation deteriorates. Roetzl and his colleagues Pedell Burkhard and Daniel Groninger recently looked at this question. They found that when someone's course of action does not yield the desired results, IO has an increasingly negative effect, which in turn leads to further bad decisions. As their paper notes:

The finding of a significant interaction between the type of feedback and the information load extends our knowledge about the role of information processing in decision-making in escalation situations. Furthermore, we find that the type of feedback affects self-justification, and we find a negative and significant interaction between information load and self-justification in negative-feedback cases. 

To understand the implication of this finding, consider a retailer back in the 2000s reacting to the rise of Amazon. Sensing that Amazon is a threat, the incumbent CEO begins to react in the ways he knows: discounting, coupons, more advertising, etc. All the responses fail, and now the competitor finds himself trying to understand not just Amazon's innovations but the reasons his responses failed. His information load has increased even more than when he was not responding. This increased level of IO can progress to a point where recovery is impossible, and the CEO becomes destined to fail completely. Much like a pilot who, in a crisis, becomes overwhelmed by cockpit alarms when corrective maneuvers fail, it's precisely when strategies do not go as planned that we are most vulnerable to the negative, or even catastrophic, impacts of IO.

Conclusions

Human beings enjoy an unlimited capacity for creating new combinations of words and ideas and long, it seems, to let them out into the world. Consider that the total amount of data created, captured, copied, and consumed each year in the world is forecast to increase to somewhere around 150 zettabytes by the year 2024. Not too long ago, “one gig” was a wealth of information. Today, it is but a tiny fraction (one trillionth) of the total information produced in one year.

Figure 2: Volume of data created, captured, copied, and consumed worldwide from 2010 to 2024 (in zettabytes) (Source: Statista)

Every day, ideas are born and released into the world – most to fade into obscurity, and a few to change society forever. For a business leader, most of that new information is easily ignored. On the other hand, some of that information is critical to innovation, market leadership, or even just a company’s survival. An executive’s challenge is, of course, how to tell one from the other. How do we know whether the information we encounter signals a sea-change in our environment, is a critical imperative, or just a rehash of tired notions gift-wrapped in new terms? How do we, as leaders, decipher the information that shouts to us every day from within books, articles, lectures, webinars, magazines, peers, consultants, conferences, white papers, and websites? Is that even possible in today’s world?

While it is tempting to say “it’s not,” and move on, this position is not an option for today’s global business leaders. As Roetzl's research illustrates, ignoring IO is a dangerous game. Indeed, I contacted Dr. Roetzl after reading his papers and asked when his interest in this topic originated. He informed me that it began during his time as an Air Force officer. It was in that role where he saw first-hand how generals struggled to make sense of the torrent of information they were expected to process and the negative consequences of failing at that task. Today he focuses a lot of his work on helping software companies design user interfaces that alleviate the IO issues raised by new management information systems that too often hurt business decision-making more than they help it.

I hope Roetzl and others will continue to shed light on IO and how it shapes decision-making in our ever-expanding world of information. Executives, for their part, should seek to understand how increasing information processing loads impact their decision-making and that of their most important leaders.

The Research

Roetzel, P.G. Information overload in the information age: a review of the literature from business administration, business psychology, and related disciplines with a bibliometric approach and framework development. Business Research 12,479–522 (2019). https://doi.org/10.1007/s40685-018-0069-z

Competition with Chinese companies — in global markets as well as in China — is one of the defining facts of life for many companies around the world. In a variety of markets and industries, Chinese brands continue to evolve upwards and outwards, often meeting or surpassing the performance of Western incumbents. As an MIT Tech Review article noted about competition within China:

In China’s ice cream market, Unilever and Nestlé S.A. had won market shares of only 7% and 5%, respectively, by 2013 — despite decades of investment. The market is dominated by two companies that most people outside of China have probably never heard of: China Mengniu Dairy Co. Ltd., with a 14% market share, and Inner Mongolia Yili Industrial Group Co. Ltd., with 19%. Meanwhile, in the Chinese market for laundry detergent, Procter & Gamble was the leading foreign brand, with an 11% share in 2013, but it was overshadowed by two China-based companies: Nice Group Co. Ltd., with more than 16% of the market, and Guangzhou Liby Enterprise Group Co. Ltd., with 15%. The home appliance market is similarly structured. Chinese companies dominate the market, with Haier Group at 29%, followed by Midea Group (12%) and Guangdong Galanz Group Co., Ltd. (4%). The two top multinational competitors, Germany’s Robert Bosch GmbH and Japan’s Sanyo Electric Co. Ltd., have only niche positions (each with less than 4%). Competition with Chinese companies — in global markets as well as in China — is one of the defining facts of life for many companies around the world. In a variety of markets and industries, Chinese brands continue to evolve upwards and outwards, often meeting or surpassing the performance of Western incumbents.

Indeed, Chinese companies such as Huawei (telecommunications equipment and smartphones), Lenovo (PCs and servers), Haier (home appliances), Galanz (microwave ovens), DJI (commercial drones), BGI (gene sequencing), CRRC Corporation (high-speed rail), Pearl River (pianos), and ZPMC (port machinery) have even managed to outperform their multinational rivals not just in China but also in external markets.

Despite the success stories, the evolution of Chinese corporate competitiveness is not evenly distributed. In some industries, Chinese companies have done very well; yet in other sectors, they have struggled against foreign competitors at home and abroad. Given the variety in outcomes, an important strategic question is what factors determine whether a Chinese brand is more or less likely to equal or surpass foreign incumbents. A paper from Peter J. Williamson (Cambridge Judge), Bin Guo (Zhejiang), and Eden Yin (Cambridge Judge) provides a useful model for considering this question.

The Two Ladder Model 

At the very start of their paper, the authors suggest that two strategic factors determine the likelihood of Chinese competitiveness. The first factor is the degree to which a sector’s local demand is fragmented, i.e., is the Chinese version of the sector contiguous, forming a “smooth path” from low to high end, or is it uneven as a rocky path climbed in leaps of various lengths. The second factor is the capability progression required to move up the sector’s value chain. Again, is there a clearly defined capability path for a Chinese company to ascend, or are there big technology leaps required along the way? The authors see these two dimensions as “ladders” — market and capability, respectively — that Chinese companies must climb in order to attain global competitiveness. Understanding these ladders, the authors argue, is a useful way to assess whether a Western incumbent faces a higher or lower risk of being disrupted by a Chinese competitor.

Ladder 1: Demand

For the authors, the market ladder is characterized by three dimensions: “relative segment size, increments between relative segments, and the continuity between adjacent segments.” 

Segment size is important because Chinese companies typically need a “wide first step” in order to start their climb. The lowest-end segment, the authors note, “often acts as an important stepping stone Chinese companies can use to establish themselves in a market initially.” Because the product functionality in this segment is relatively basic, “the quality and depth of capabilities required to compete in this segment are limited.” Competitive rivalry tends to “be waged on price, enabling local firms, often from rural areas, to leverage their access to very low-cost labor.”

Increment size is important because Chinese companies find it easier to move up in a market where demand has many levels, i.e., it is easier to “step up” from one demand class to another if there are narrow segments. As the authors note, “a ladder with closely spaced rungs will enable Chinese competitors that enter at the low end to gradually climb up to higher-end segments without the need to understand a radically different set of customers or buying behaviors.” Rather than having to serve complex customers from the start, “they can succeed in moving up-market by serving successively more demanding customers who, while they have higher expectations of the product or service, do not have fundamentally different buying criteria.” 

A ladder that has widely spaced rungs is much harder to climb. A case in point is the Chinese business software industry. The authors explain that local players Yonyou and Kingdee “have been able to dominate the segment made up of small- and medium-sized enterprises, which require localized features, simple installation, and low prices.” But the huge gap between this segment and large enterprises “means that foreign companies such as Oracle and SAP remain the leading players in the top-end enterprise resource planning software.”

As for segment continuity, markets with no large gaps are easier to ascend. When one or more segments are missing, progress becomes more difficult because any company wishing to move from the low to high end “has to leap across the missing segment.” The domestic motorcycle sector is a good example of this phenomenon:

Some 95% of the market was in the lower end segment of motorcycles with between 50 and 150 cubic centimeters (cc) of engine displacement (as a measure of power). There was a small but valuable upper-end segment of powerful motorcycles with greater than 250 cc engines that accounted for around 3% of the market. This was dominated by Japanese firms such as Honda and Yamaha. But the mid-market segment of 150 cc motorcycles, the largest segment in most countries, was almost entirely missing in China, accounting for just 1.8% of the total market...As a result of this segment discontinuity, Chinese motorcycle manufacturers had neither the incentive nor the opportunity to gain experience in upgrading their products to serve the tiny mid-market segment. At the same time, the leap of understanding and capability required to compete with their Japanese rivals presented an almost impossible challenge. As a result, Chinese companies failed to catch up with their international competitors in the motorcycle market. They have still not done so today.

Naturally, the three dimensions are not static. Forces such as changing company ownership, strategic direction, and technological shifts such as e-commerce are continuously reshaping the market ladder. Indeed, in today’s China, “two of the most important factors reshaping China’s market ladders are the rise of a huge middle class and the boom in e-commerce, particularly the use of smartphones.”

Ladder 2: Capability

The second ladder refers to the fact that “companies usually catch up by first developing or accessing simpler technologies that allow them to compete at the lower end of the market.” As they enhance their skills and abilities, they attempt to serve more complex markets. As with market ladders, three factors shape capability ladders:

• Length: “the existence of simpler technologies that are relatively easy to master and are suitable to meet the needs of lower-end market segments (parallel to the existence of sizeable demand for low-end products and services in the market ladder);”

• Increments: “the evaluation of the difficulty and complexity between different rungs of the capability ladder;” and

• Continuity: “which looks at whether or not there are gaps or discontinuities in the capability ladder.”

Relative to the first dimension, the authors note that the longer a ladder is — the more it spans a wide range of technologies, from simple solutions to complex, sophisticated ones — the easier it is for Chinese companies to climb. For example, in the Chinese market for “digital direct radiography machines, the availability of simpler line-scanning technologies that were easy to master and suitable for basic applications such as chest X-rays allowed Chinese competitors such as Zhongxing Medical to build an initial base of capabilities.” Over time, the company eventually mastered “the flat-panel X-ray imaging technologies pioneered by Koninklijke Philips N.V (Philips) and General Electric, which are capable of creating videos of a patient’s beating heart.” On the other hand, in the passenger jet industry’s truncated ladder, “Chinese companies such as the Commercial Aircraft Corporation of China (Comac) could not enter by adopting a simpler technology and gradually climbing the capability ladder.” Instead, “Comac was forced to embark on the decade-long, difficult process of mastering state-of-the-art technologies before it could enter the market.”

The jet case also illustrates the important role played by “the size of the increments in difficulty and complexity between different rungs of the capability ladder.” Comac’s affiliate, Harbin Aircraft Industry Group, has been selling turboprop aircraft to customers since 1985. However, the large gap between turboprop technology and the technology needed for passenger jet aircraft meant that its C919 passenger jet entered service for the first time in 2021.

As with the market ladder, continuity refers to a ladder made up of smooth or disjointed increments. For example, note the authors, mobile phone sophistication is defined by “a series of more powerful chipsets, better lenses and cameras, antennas, microphones, speakers, and more powerful software.” This smooth ladder “has enabled many Chinese firms ranging from Huawei, Lenovo, and Xiaomi through to Shanzai players such as SciPhone — to upgrade the quality and features of their products rapidly through a fast-paced series of incremental improvements to match world standards and sometimes achieve leadership.” In contrast, the passenger elevator business “exhibits significant discontinuities between, for example, the high-speed hydraulic elevator technology delivering speeds of 2.5 meters per second used in taller buildings and the traction technology that can be used in less demanding, low-rise applications.” Consequently, it has been difficult for Chinese competitors to climb the capability ladder and the high-end segments in the Chinese elevator market are still dominated by foreign companies, “including Otis, Mitsubishi Electric, Schindler, and Kone, who together enjoy over 70% share despite China being the world’s largest market and with local companies in the market for several decades.”

Other factors

In addition to ladder characteristics, the researchers note, certain other factors impact the success a Chinese company has moving up the capability ladder. For example, Chinese companies ascend more rapidly when there is wider availability of external knowledge about a sector. The government’s regulatory position is also important, though sometimes not in the ways intended. For example, note the authors, while government policies requiring joint ventures helped Chinese train manufacturers to become globally competitive, they had the opposite effect in the car industry where VW and Toyota remain the dominant players. 

Another key factor is the structure of the global value chain itself. In industries such as car manufacturing, Chinese companies entered as relatively low-level suppliers and then moved up the global value chain. For example, Wanxiang Group, today a tier-one supplier to the global automobile industry, first entered the automotive value chain as a supplier of universal joints. However, “Wanxiang was gradually able to understand higher value-added technologies and the components in which they were embedded.” As it climbed the capability ladder, “Wanxiang expanded its product range to include complete steering systems, driveshafts, and braking systems.” In contrast, “flow” industries, such as petrochemicals have been difficult for Chinese companies because these sectors require “a chain of highly systemic interactions between research, development, and clinical testing teams working together to use their tacit knowledge of interrelated, often proprietary processes.”

Figure 1: Determinants of when Chinese competitors can catch up (Source: Authors)

As expected, acquisition opportunities represent another way “through which Chinese companies can access external technological know-how to help them climb the ladder.” Wanxiang, in fact, used acquisitions to climb the technological ladder. It made “its first overseas acquisition in 2001 with the purchase of a controlling interest in Universal Automotive Industries (UAI), a Nasdaq-listed U.S. supplier of braking systems.” Over the next 15 years, it acquired more than 30 companies, and “each of these [acquisitions] brought access to new technologies, helping Wanxiang climb the capability ladder.”

Competitive Responses

In the final sections of their paper, the authors summarize their model graphically. The authors explain that “by analyzing the characteristics of the respective market and capability ladders for their industry, companies faced with growing Chinese competition can first assess their level of vulnerability and then develop appropriate strategies to counter this competitive threat.” Depending on the nature of the market and capability ladders, four scenarios are possible as depicted in Figure 2.

Figure 2: Assessing vulnerability to Chinese competition (Source: Authors)

The authors explain the figure thus:

Scenarios 1 and 3 are straightforward: market and capability ladders with characteristics that make it relatively easy for Chinese competitors to climb, increase vulnerability, and characteristics that impede Chinese competitors trying to climb the ladder reduce vulnerability. Given the nature of their market and capability ladders, some of the industries prone to Scenario 1 include personal computers, smartphones, and other consumer electronics as well as home appliances...Scenario 3 includes aerospace, banking, branded luxury goods, and possibly semiconductors.

Scenario 2 is more complex. It is relatively easy for Chinese competitors to climb the capability ladder, but they face impediments in climbing the market ladder and, thus, they are likely to try to win share by disrupting existing market segments. A common strategy to achieve this involves rapidly deploying new technologies into the mass market, gaining scale advantages that undermine the price premium incumbents enjoy in up-market segments. Industries that have begun to see this scenario play out include medical equipment, machinery, and renewable energy equipment such as wind turbines and solar panels.

In Scenario 4, it is relatively easy to climb the market ladder, but not to match the quality of high-end technology. The risk here is that Chinese competitors reverse-engineer the offering to eliminate features most customers do not often use and deploy adequate but cheaper and simpler technologies to offer improved value-for-money, commoditizing the market and forcing rivals to compete on cost. A classic example is the revolution Haier initiated in the market for specialist wine-storage refrigerators as well as those used in commercial bars and restaurants.

Figure 3: Strategies to respond to Chinese competition (Source: Authors)

As shown in Figure 3, the authors also provide recommended strategies that can be applied to each of these scenarios. Space does not allow a summary of them all, but one of the more interesting ones is “polarization.” In this scenario, a Western incumbent disrupted by a Chinese product that is inferior on some dimensions but delivers the key attributes that most customers value, responds by emphasizing the non-technical attributes of its products. Tesla, note the authors, “recently embarked on this path in China, emphasizing its Californian roots, promoting its styling and driving experience, and enhancing the comfort of its rear seats to appeal to wealthy Chinese who typically employ a chauffeur.”

Conclusions

In their final discussion, the authors rightly note that “predicting the industries in which fast-rising Chinese companies will be able to catch up with their incumbent multinational competitors and when is not an exact science.” But the framework they present for Western executives “can provide an important first step toward a reliable guide.” As noted above, the “guide” function comes down to three strategic assessments of a market:

1. The relative size of different segments from low end to high end;

2. The size of the quality increments between these segments; and

3. Continuity, which looks at whether or not there are gaps in the segments of this market ladder that could trip Chinese companies up.

Likewise, Western executives should understand the capability ladder that Chinese companies are climbing, determining whenever possible:

1. The length of the capability ladder;

2. The size of the increments in capabilities necessary to move from one technology to the next; and

3. Whether or not there are missing links between technologies that are difficult to leapfrog.

Wherever the results of these assessments ultimately land, the researchers advise, “it is important to remember that whatever else Chinese companies are doing, they are certainly not standing still.” This is a challenging statement, of course. Still, the model presented in this paper offers a novel and helpful framework for Western strategists to understand the progression of Chinese competitors in their markets and develop the appropriate strategic responses.

The Research

Peter J. Williamson, Bin Guo, Eden Yin. When can Chinese competitors catch up? Market and capability ladders and their implications for multinationals. Business Horizons, Volume 64, Issue 2, 2021, Pages 223-237, ISSN 0007-6813. https://doi.org/10.1016/j.bushor.2020.11.007.

Author Interview

The DEI Monthly

A conversation about how Chinese companies move up the global value chain

Listen now (15 min) | DEI Monthly presents brief conversations with the authors of the most-viewed posts featured in the newsletter. In this episode, I speak with Peter Williamson, Ph.D. (Honorary Professor of International Management, Cambridge Judge Business School at the University of Cambridge) about his research into strategic competition in China. The research summary …

Listen now

4 years ago · 2 likes · Carlos Alvarenga

Leadership changes in big companies are an especially alluring subject for researchers, and there is an extensive literature that looks at how external factors such as the economic downturns or technology disruptions can lead a board to fire a CEO. A driver of this interest is that the frequency of CEO dismissal has significantly increased in recent years. Indeed, studies report that CEO dismissal now accounts for 24% of succession events in S&P 500 companies, reducing the average CEO tenure in the U.S to 9 years in 2017 compared to 14 years before the 1990s.

As noted, researchers tend to focus on company performance or external factors when analyzing CEO changes. Far less attention has been paid to the actions, or even words, of the CEO themselves. We should ask to what extent things CEOs tell internal and external constituents affect their tenure. After all, most CEOs are in a continuous dialogue with their employees and boards, as well as with important external agents such as financial analysts, journalists, activists, and even professors. It stands to reason that what they communicate about themselves in these relationships should affect how others gauge their performance. Furthermore, it is likely that in many cases, external narratives impact a CEO's tenure as much as internal performance. But how exactly does that work and how do a CEO’s own words shape how she is perceived by others? These questions are the subject of research from Sun Hyun Park (Seoul National University), Sung Hun (Brian) Chung (Rice), and Nandini Rajagopalan (USC).

The authors base their research on the hypotheses that (1) a CEO’s narrative about her own performance shapes how others see her and that (2) this effect creates beliefs about a CEO’s connection to her company’s performance that can produce unintended consequences. To test their hypotheses, the authors focus specifically on what they call performance attributions, a term that refers to the links a CEO attempts to establish between her leadership and her company’s results. A CEO makes an internal performance attribution when she takes credit for the company's success and an external attribution when she does the opposite (e.g., blaming industry conditions for bad performance).

Attribution strategy is important, the authors note, because "studies of CEO career dynamics suggest that throughout a CEO's career, a firm’s constituents develop heuristics for evaluating leadership efficacy." Thus, when a CEO makes internal performance attributions, constituents may "develop enhanced expectations for continued favorable performance outcomes and the CEO's strategic leadership, causing the self-serving rhetoric to backfire when these expectations are not met." Indeed, studies show that internal performance attributions “allow romanticized conceptions of a CEO's leadership, where the firm’s constituents overestimate the leader's role in the success of firm performance outcomes, which are essentially causally indeterminate and ambiguous.”

Knowing that performance attributions affect how stakeholders judge a CEO's performance, the authors ask a simple question: Can a CEO's internal performance attributions impact not just how they are seen but how they are treated in difficult times? In other words, can a CEO’s own words lead to her termination when her company does not perform as expected?

Methodology

To test the impact of the claims of fired CEOs over their careers, the authors start by examining performance attributions made by CEOs in quarterly earnings announcements. These public discussions are not only great stages for a CEO to communicate to outsiders about her firm, they also act as a regular trigger for analysts, journalists, and boards to check in on a CEO’s performance. Analysts are a special focus for the authors, because they believe that the conclusions of financial analysts about a firm's performance are an important mediating variable between the CEO's claims and tenure. Moreover, analysts often play the role of "societal arbiters," as they shape a collective (internal and external) perspective on how well, or poorly, a CEO is performing.

As the foundation for their analysis, the authors looked at the tenure of 378 CEOs who left their job in 234 companies from 2000 to 2011. The companies were drawn from the list of S&P 1500 energy, chemical, pharmaceutical, and oil sectors. The sampling strategy focused on a handful of related industries, allowing the authors to "minimize potential confounding factors stemming from cross-industry variance, especially with respect to the heterogeneity in a firm's external environment and managerial discretion."

The team analyzed statements made by CEOs, analysts, and journalists commenting on each firm over a given CEO's leadership tenure. Press sources included national business publications (e.g., the Wall Street Journal), national non-business publications (e.g., The New York Times), news-wire agencies that consolidate other media outlets (e.g., Reuters), and local media in the market where each firm has its corporate headquarters (e.g., the Chicago Tribune).

The authors were careful to exclude turnover that occurred when CEOs died suddenly (five cases), when CEOs took a similar position at another firm (five cases), or when the focal firm was acquired (three cases). The authors also considered CEO age and continued board membership as additional criteria to identify a CEO dismissal case. CEOs older than 64 leaving a leadership position were excluded, as the change was attributed to retirement. Alternatively, CEOs under the age of 64 who did not remain on the board after the succession were classified as dismissals. In the end, the researchers focused their conclusions on 94 CEO dismissals cases that met all the criteria for analysis.

The Findings

Consistent with prior studies, the authors found that CEOs who internally attributed their firm's positive performance — especially when they reported better than unexpected results — created strong links between CEOs and their firms in the minds of the analysts who covered their firms. Of course, this phenomenon can be a useful (and very lucrative) outcome when the firm does well. However, the authors found that it quickly becomes a potentially serious problem when a firm underperforms.

Indeed, the study results indicate that a standard deviation increase in CEO internal attributions for a positive earnings surprise “is associated with an approximately 51% increase in analyst internal attributions for a negative earnings surprise, when other variables are held constant at their means." Furthermore, when analysts made negative connections, they often had a direct impact on a CEO’s termination. As the authors note, a standard deviation increase in analyst internal attributions of a CEO for a negative earnings surprise is associated with an approximately “45% higher likelihood of CEO dismissal when other variables are held constant at their means.”

These findings are even more surprising when the authors point out that their examination of records preceding a CEO’s firing reveals that unfavorable firm performance alone is often not enough to get a top leader fired. Indeed, research has shown that most CEOs are very good at installing defense mechanisms to protect their jobs. Moreover, the link between negative firm performance and a CEO’s responsibility is often disrupted by other forces, such as boardroom politics or economic downturns. The causal linkage is further “weakened in CEO-dominated boards when the leader has appointed demographically similar board members or outside directors whose corporate elite status is lower than that of the CEO.” The linkage may be also decoupled when the board interprets the performance downfall as temporary, opting to stick with a positive belief in the CEO's ability to turn around the unfavorable situation.

In short, given all the factors that keep boards from firing CEOs when firms underperform, it is notable to find that CEOs who routinely take the credit for good outcomes often have to take the blame when things go bad. 

Conclusions

In social science research, the Matthew Effect occurs when eminent team members receive credit for great work at the expense of their less-eminent collaborators. Said differently, a well-known researcher is often given more credit than less-known collaborators simply because she is famous. It is the scientific equivalent of the “rich getting richer.” The term has been extended into other areas of research, and this paper suggests that there may be a version of the Matthew effect at work in the C-suite. In other words, the stronger the link a CEO makes between herself and her firm’s positive performance, the more other people are willing to reinforce it. Over time, that high status can function as a buffer against potential failure, for, as the authors note, a CEO's internal performance attributions function as a “cognitive anchor or a source of confirmation bias” for external evaluations of the CEO's leadership throughout their career.

I suspect that what is true for CEOs also applies to other senior leaders in business, politics, and even sports. After all, it is not just CEOs who claim credit when the seas are calm and the ship speeds along in fair winds.

The real-world implications of this particular study are not difficult to comprehend. Its empirical analysis suggests that while CEOs’ internal attributions of positive firm performance "can be an active behavioral strategy of self-presentation," those very attributions can also be the reason boards fire them when things don't turn out as planned. Consequently, CEOs should think twice before they take personal credit for great firm performance. As the authors wisely advise:

Given the difficulty of predicting firm performance results over an extended period of time, CEOs should be wary of the possibility that when they take credit for positive firm performance, they may subsequently invite blame from firm constituents that is anchored around their self-serving attribution accounts. A more modest presentation of one's own abilities may be the best strategy when there exists the potential that an individual's impression management can backfire in the future.

A final note: in 2011, Swedish researcher Tobias Fredberg concluded that a CEO’s tendency to stress me over we was related primarily to whether or not she was dealing with a turnaround situation. Fredberg found that the best turnaround CEOs connected problems to themselves and success to their teams. In other words, they were more inclined to establish an internal performance attribution of negative performance, perhaps signaling to stakeholders that they were neither wholly to blame for problems nor fully to praise for success. This is an interesting, if risky, strategy for any senior leader, but it is one whose value is reinforced by the findings of this team’s insightful study.

The Research

Park, S. H., Chung, S. H. (B.), & Rajagopalan, N. (2021). Be careful what you wish for: CEO and analyst firm performance attributions and CEO dismissal. Strategic Management Journal, 1– 29. https://doi.org/10.1002/smj.3312

As Artificial Intelligence (AI) systems become commonplace in business and society, they are engendering complex debates concerning the ethical policies and rules that should govern their design, deployment, and power. Many big tech companies such as Google and Microsoft have articulated general principles that should guide their AI teams, but these efforts have, to date, mainly provided aspirational guidance. Moreover, even companies as rich and talented as Google have struggled when moving from general statements of intent—e.g., AI systems should not discriminate on the basis of race—to specific rules and development strategies.

Because the newness and complexity of AI systems are posing serious hurdles to technologists and ethicists, some regulators have decided they must move forward and create regulatory answers to AI's tough questions. In places as diverse as the European Union and China, the past few years have seen the proposal of new AI regulations that — in the case of the EU at least—would significantly constrain what AI developers and their customers can and cannot do with these new technologies. The pressing challenge for both private and public stakeholders in the AI ecosystem is to accelerate past today's general statements of intent to specific rules and guidelines that are consistent with social norms, goals, and established legal principles.

Seeking to support this effort is a proposal from the Atlantic Council's Geotech Center written by John Basl (Northeastern), Ronald Sandler (Northeastern), and Steven Tiell (Accenture). Their paper provides organizations working with AI systems a framework for developing high-level statements of intent into specifics. As shown in Figure 1 below, their approach is built on the concept of normative content. As the authors note, norms come in two types. One is descriptive, i.e., describe "what is." The other form is prescriptive, i.e., they describe "what should be." One category of prescriptive norm is ethics, and ethical norms are the foundation of their model. In the context of this paper, ethical norms should serve as the foundation of "well-justified standards, principles, and practices for what individuals, groups, and organizations should do, rather than merely describing what they currently do."

Figure 1: Framework for cascading ethical norms into specific attributes of AI systems (Source: Authors)

The authors illustrate their approach with an example from healthcare, specifically the idea that any patient who participates in a medical trial should have given informed consent. For the authors, informed consent— as a specific practice—flows down from a higher-order norm they identify as respect for the individual patient. In other words, the reason informed consent exists is that research institutions are committed to enabling and respecting a patient's individual right to be the ultimate arbiter of his own medical care. As shown in Figure 2 below, the high-order norm of respect for the individual leads to accepting the ethical concept of autonomy, which then leads to the commitment to require informed consent for any clinical trial.

Figure 2. Example: Ethical framework for respecting autonomy in medical research (Source: Authors)

For the authors, informed consent in bioethics has several lessons for moving from general ethical concepts to practical guidance in AI and data ethics: 

• To move from a general ethical concept (e.g., justice, explanation, privacy) to practical guidance, it is first necessary to specify the normative content (i.e., to specify the general principle and provide context-specific operationalization of it). 

• Specifying the normative content often involves clarifying the foundational values it is meant to protect or promote. 

• What a general principle requires in practice can differ significantly by context. 

• It will often take collaborative expertise—technical, ethical, and context-specific—to operationalize a general ethical concept or principle. 

• Because novel, unexpected, contextual, and confounding considerations often arise, there need to be ongoing efforts, supported by organizational structures and practices, to monitor, assess, and improve operationalization of the principles and implementation in practice. It is not possible to operationalize them once and then move on. 

With these lessons in mind, the authors look at two ethical norms common to AI policies today and apply their framework for their development.

Issue 1: Justice

AI ethics frameworks generally subscribe to the idea that these new technologies should, at the very least, not make the world more unjust. Ideally, the adoption of AI should make the work more just in some way. But what, the authors ask, does justice mean in the world of AI? The complexity of the concept suggests to them that "to determine what justice in AI and data use requires in a particular context— for example, in deciding on loan applications, social service access, or healthcare prioritization—it is necessary to clarify the normative content and underlying values." For the authors, justice as an abstract idea seems to be of little value. It is only when the norm of justice is placed in a specific context that it is "possible to specify what is required in specific cases, and in turn how (or to what extent) justice can be operationalized in technical and techno-social systems."

As shown in Figure 3 below, they explain their approach by looking at a hypothetical AI application in the financial services industry.

Figure 3: Example: Ethical framework for algorithmic lending (Source: Authors)

In this case, the firm—whether for internal reasons or regulatory requirements—commits to nondiscrimination, equal treatment, and equal access in deploying, for example, an AI-based mortgage approval algorithm. To ensure that its high-level norm is met, the firm also adopts a series of commitments and system specifications designed to align with the overall norm and the specific challenge of justly approving the right mortgage applicants. In discussing their hypothetical model, the authors note that the specifics presented may not be exhaustive. For example, "if there has been a prior history of unfair or discriminatory practices toward particular groups, then reparative justice might be a relevant justice-oriented principle as well." Or if a firm has a social mission to promote equality or social mobility, then "benefiting the worst-off might also be a relevant justice-oriented principle."

For all cases, however, the authors highlight two key points:

First, there is no singular, general justice-oriented constraint, optimization, or utility function; nor is there a set of hierarchically ordered ones. Instead, context-specific expertise and ethical sensitivity are crucial to determining what justice requires. A system that is designed and implemented in a justice-sensitive way for one context may not be fully justice-sensitive in another context, since the aspects of justice that are most salient can be different. 

Second, there will often not be a strictly algorithmic way to fully incorporate justice into decision-making, even once the relevant justice considerations have been identified. For example, there can be data constraints, such as the necessary data might not be available (and it might not be the sort of data that could be feasibly or ethically collected). There can be technical constraints, such as the relevant types of justice considerations not being mathematically (or statistically) representable. There can be procedural constraints, such as justice-oriented considerations that require people to be responsible for decisions. 

Issue 2: Transparency

A significant concern for AI regulators is transparency, i.e., the ability for the creators and users of AI to understand how the systems operate and reach decisions. As with justice, however, transparency is a term that can signify different things to different people. Thus, translating calls for transparency into guidance for the design and use of algorithmic decision-making systems "requires clarifying why transparency is important in that context, for whom there is an obligation to be transparent, and what forms transparency might take to meet those obligations."

The authors discuss several principles that could be used to translate the high-level idea of transparency into commitments and specifications. These include "explainability," i.e., an AI maker commits to not making a system whose inner workings cannot be explained to outsiders. Another principle might be "auditability," i.e., creating the ability for an internal or external review of an AI's operating and decision history. A third and critical issue is "interpretability," the commitment not to create AI whose capabilities are beyond the understanding of even its creators. This phenomenon is not a theoretical problem; it is already here, as a 2017 MIT Tech Review article about an AI-controlled autonomous car reported:

Getting a car to drive this way was an impressive feat. But it’s also a bit unsettling, since it isn’t completely clear how the car makes its decisions. Information from the vehicle’s sensors goes straight into a huge network of artificial neurons that process the data and then deliver the commands required to operate the steering wheel, the brakes, and other systems. The result seems to match the responses you’d expect from a human driver. But what if one day it did something unexpected—crashed into a tree, or sat at a green light? As things stand now, it might be difficult to find out why. The system is so complicated that even the engineers who designed it may struggle to isolate the reason for any single action. And you can’t ask it: there is no obvious way to design such a system so that it could always explain why it did what it did.

As an example of how to address issues such as those noted above, Figure 4 below illustrates how the authors’ framework cascades a first-order value—understanding—through a drug discovery AI system so that its final specifications align with and support the original goal.

Figure 4: Example: Ethical framework for a drug discovery algorithm (Source: Authors)

A similar approach, the authors suggest, can be used to expand notions of audibility and explainability from value statements into specific system attributes at the operational level.

Conclusions

As shown in Figure 5 below, institutional commitments to high-level AI principles are becoming commonplace. However, these right-sounding statements all too often lack specifics about the policies and technical decisions that must be adopted in order to make the high-level goals a reality. The authors conclude their paper by noting that "if these concepts and principles are to have an impact—if they are to be more than aspirations and platitudes—then organizations must move from loose talk of values and commitments to clarifying how normative concepts embody foundational values, what principles embody those concepts, what specific commitments to those principles are for a particular use case, and ultimately to some specification of how they will evaluate whether they have lived up to their commitments." 

Figure 5: Examples of AI principles, codes, and value statements (Source: Authors)

It is common to hear organizations claim to favor justice and transparency, the authors note, but "the hard work is clarifying what a commitment to them means and then realizing them in practice." This paper provides a solid point from which to start the "hard work" noted above. Some issues deserve attention in future work from this team. For example, the authors repeatedly stress the importance of "context," but making ethics contextual opens up the risk that AI ethics slide down the slippery slope of moral relativism. It would be useful to understand where and how to draw the line when granting contextual exceptions to global AI policies. This challenge is especially true of technologies such as facial recognition, which are widely adopted in countries such as China but are the subject of intense criticism in the West. In addition, the authors did not include symmetrical identification in their discussion of transparency. The idea that AI should make itself known to those who interact with it is embedded in the new EU AI regulations, so it was surprising not to see the topic included in their discussion on transparency.

Another issue the authors do not discuss relates to the ethical implications of AI's inevitable unintended consequences. As with all great technological shifts, AI will generate effects that no one expected. In fact, the recent case of Airbnb’s price optimization algorithm is a case in point. The tool worked as intended; however, because Black hosts were 41% less likely than White hosts to adopt the algorithm, it increased the racial revenue gap at the macro level. In other words, even an AI that did what it was supposed to do ended up making the overall platform less just. It would be useful to know if the authors think AI ethical systems should define the responsibility for such outcomes beforehand? If so, how?

As noted above, this paper does not address some critical issues in the current AI ethics debate. However, those omissions are likely a result of the project's limited scope, which was to provide a framework for development and not the answers themselves. As such, the paper succeeds in its aim. It provides a framework that both executives and technologists can use to have serious discussions about the ethical implications of their AI creations. As the authors note, "the challenge of translating general commitments to substantive action is fundamentally a techno-social one that requires multi-level system solutions incorporating diverse groups and expertise." This brief but thoughtful analysis is a solid starting point for business and social leaders wishing to take on this challenge.

The Research

John Basl, Ronald Sandler, and Steven Tiell. Getting from Commitment to Content in AI and Data Ethics: Justice and Explainability. Atlantic Council Report, Aug 2021, Available at https://www.atlanticcouncil.org/in-depth-research-reports/report/specifying-normative-content/

The Interview

The DEI Monthly

A conversation about AI ethics: moving from aspirations to specifics

Listen now (29 min) | DEI Monthly presents conversations with the authors of research featured in the newsletter. In this episode, I speak with Steven Tiell about his research into the ethics of AI. The original research summary is linked below…

Listen now

2 years ago · 4 likes · Carlos Alvarenga

One of the widely-reported consequences of the pandemic has been the accelerated migration of consumer purchasing activity to the internet. Indeed, for many consumers around the world e-commerce is not a complement to traditional retail activity, it is a convenient substitute that has replaced traditional in-person buying. The ease of buying online is undeniable; however, this form of commerce has consequences that many consumers and even businesses do not fully comprehend. One such consequence is the set of markers that we create as we browse and shop online. Commonly referred to as our "digital footprint," the specific features of our digital presence, and what they say about us, are increasingly important topics for researchers and innovators. 

A paper from Tobias Berg (Frankfurt), Valentin Burg (Home24), Ana Gombovic (Deloitte), and Manju Puri (Duke) provides a fascinating view into this emerging line of research. Their paper set out to understand whether a consumer's digital footprint can predict the creditworthiness of an online shopper as well as traditional credit-report databases. Their findings highlight just how much our digital footprint says about us and suggest that in the future everyday online activity may supersede traditional credit reporting with both positive and negative consequences for consumers.

As the authors note at the start of their paper:

Understanding the informativeness of digital footprints for consumer lending is significantly important. A key reason for the existence of financial intermediaries is their superior ability to access and process information relevant for screening and monitoring of borrowers. If digital footprints yield significant information about predicting defaults, then FinTech firms—with their superior ability to access and process digital footprints—can threaten the information advantage of financial intermediaries and thereby challenge financial intermediaries’ business models.

The Study

The authors based their research on data gathered from a German home furnishings retailer (similar to “Wayfair” in the United States) between October 2015 and December 2016. For any purchase over €100, the seller requires customers to create personal profiles before purchasing an item, and this profile is used to determine whether the buyer is allowed to buy “on invoice,” i.e., the merchandise is sent immediately and the customer has 14 days to pay the balance. In effect, the seller is giving creditworthy customers a short-term loan, the availability of which is determined by data drawn from two sources—two traditional credit reports and the buyer’s digital footprint. The first credit report provides basic information, such as whether the customer exists and whether the customer is currently in or has been recently in bankruptcy.  The second credit report score draws on credit history data from various banks, sociodemographic data, and past payment behavior. For the purposes of their study, the authors labeled all customers for whom both reports existed as “scorable.”

The buyer’s digital footprint contains a variety of information collected from the buyer’s internet characteristics. Examples of information contained in the digital footprint include:

• The device type (desktop, tablet, mobile) and operating system (e.g., Windows, iOS, Android)

• The consumer’s e-mail provider (e.g., Gmx, Web, T-Online, Gmail, Yahoo, or Hotmail)

• The channel through which the customer has visited the homepage of the seller, e.g, paid clicks (mainly through paid ads on Google), direct (a customer directly entering the URL of the E-commerce company in her browser), affiliate sites (customers coming from an affiliate site that links to the seller’s web page), and organic (a customer coming via the nonpaid results list of a search engine)

• The hour of the day at which the purchase was made

• Browsing time onsite

• Other technical details

The authors collected data from approximately 270,399 purchases made by customers with low to very high creditworthiness (but excluding those with “very low” scores). These selection criteria, the authors note, have "the benefit of making our data set more comparable to a typical credit card, bank loan or peer-to-peer lending data set.” They also imply that “the discriminatory power of the variables in our data set is likely to be larger in a sample of the whole population compared to a sample that is selected based on creditworthiness.”

In the sample with credit bureau scores, the average purchase volume is €318 (approximately $350), and the mean customer age is 45.06 years. On average, 0.9% of customers default on their payment. Importantly, the authors’ data set is largely representative of the geographic distribution of the German population overall.

The Findings

Having collected the purchase data, the authors make some preliminary observations that inform their overall conclusions. The distinct features of the most commonly used e-mail providers in Germany allow the authors to infer information about a customer’s economic status,e.g., “T-online is a large internet service provider and is known to serve a more affluent clientele, given that it offers internet, telephone, and television plans and in-person customer support.” A customer obtains a T-online e-mail address only if she purchased a T-online package. Yahoo and Hotmail, in contrast, “are fully free and mostly outdated services.” Moreover, other research has shown that “owning an iOS device is one of the best predictors for being in the top quartile of the income distribution.” Thus, “based on these simple variables, the digital footprint provides easily accessible proxies of a person’s economic status absent of private information and difficult-to-collect income data.”

Interestingly, the authors believe that the digital footprint also provides information about a person’s character. “Her self-control,” for example, “is also reasonably assumed to be revealed by the time of day at which the customer makes the purchase (for instance, we find that customers purchasing between noon and 6 p.m. are approximately half as likely to default as customers purchasing from midnight to 6 a.m.).” Even the choice of e-mail addresses contains risk information. Eponymous customers—those who include their first and/or last names in their e-mail address—are less likely to default than those who include numbers. Even the way information is provided online is useful: typing errors or even lack of capitalization in names and addresses are associated with a higher credit risk level.

Looking at the data itself, some findings are as expected. The credit report information is a useful indicator of creditworthiness: “the default rate in the lowest credit score quintile is 2.12%, more than twice the average default rate of 0.94% and 5 times the default rate in the highest credit score quintile (0.39%).” Of more interest is that digital footprint variables also prove to be useful predictors of future payment behavior:

For example, orders from mobile phones (default rate 2.14%) are 3 times as likely to default as orders from desktops (default rate 0.74%) and two-and-a-half times as likely to default as orders from tablets (default rate 0.91%). Orders from the Android operating systems (default rate 1.79%) are almost twice as likely to default as orders from iOS systems (1.07%), consistent with the idea that consumers purchasing an iPhone are usually more affluent than consumers purchasing other smartphones. As expected, customers from a premium internet service (T-online, a service that mainly sells to affluent customers at higher prices but with better service) are significantly less likely to default (0.51% vs. the unconditional average of 0.94%). Customers from shrinking platforms like Hotmail (an old Microsoft service) and Yahoo exhibit default rates of 1.45% and 1.96%, almost twice the unconditional average.

As the authors expected, information about online behavior is also significantly related to default rates. Customers arriving on the homepage through paid ads, for example, “exhibit the largest default rate (1.11%),” perhaps because particular ads that are shown multiple times on various websites to a customer, “seduce customers to buy products they potentially cannot afford.” Customers being targeted via affiliate links, price comparison sites, and customers directly entering the URL of the seller, on the other hand, exhibit lower-than-average default rates (0.64% and 0.84%). Finally, “customers ordering during the night have a default rate of 1.97%, approximately twice the unconditional average.”

A few more findings are also worth noting. The first is that very few customers make typographical errors while writing their e-mail addresses (roughly 1% of all orders), but those who do are much more likely to default (5.09% vs. the unconditional mean of 0.94%). The second is that “customers with numbers in their e-mail addresses default more frequently, which is plausible given that fraud cases also have a higher incidence of numbers in their e-mail address.” Furthermore, customers who use only lowercase letters in their names and shipping addresses are more than twice as likely to default as those writing names and addresses with first capital letters.

As one would expect (and as illustrated in Figure 1 below), the value of the digital footprint signals increase as they are connected:

When combining information from both variables (Operating system and E-mail host), default rates are even more dispersed. We observe the lowest default rate for Mac-users with a T-online e-mail address. The default rate for this combination is 0.36%, which is lower than the average default rate in the 1st decile of credit bureau scores. On the other extreme, Android users with a Yahoo e-mail address have an average default rate of 4.30%, significantly higher than the 2.69% default rate in the highest decile of credit bureau scores. These results suggest that even two simple variables from the digital footprint allow categorizing customers into default bins that match or exceed the variation in default rates from credit bureau deciles.

Figure 1: This figure shows default rates for combinations of the variables Operating system and E-mail host for all combinations that contain at least 1,000 observations. The x-axis shows default rates, and the y-axis illustrates whether the respective dot comes from a single digital footprint variable (e.g., “Android users”) or whether it comes from a combination of digital footprint variables (e.g., “Android + Hotmail”). (Source: Authors)

All in all, when compared to credit reports, a consumer’s digital footprint is both economically and statistically a better indicator of creditworthiness—not just at the time of purchase but with respect to recovery rates post-default. The authors are careful to point out that, in their opinion, digital footprint data is a complement, and not a replacement for, credit report data. Unfortunately, the authors do not expand on this conclusion sufficiently, given that they also claim that “even simple, easily accessible variables from the digital footprint are important for default prediction over and above [Italics mine] the information content of credit bureau scores.”

In their appendix, the authors provide additional anecdotal information that supports the conclusion that the findings of this study are not unique to this one seller. Taken together with the primary findings, all the evidence the authors present strongly suggests that digital footprints are a useful indicator of consumer behavior and may even indicate creditworthiness changes before they appear on credit reports. As illustrated in Figure 2 below, the addition of digital footprint data by the seller (which occurred in 2015) into credit decisions has had a material positive impact on the business: “introduction of the digital footprint decreases defaults by roughly one-third, yielding a decrease in default rates of approximately 0.8 percentage point or around €50,000 defaulted loans per month, equivalent to losses of €35,000 per month/0.6 percentage point with a loss given default of 70%.” Assuming a 5% operating margin, this change would be “an improvement in the operating margin of more than 10% that is attributable to the introduction of the digital footprint.”

Figure 2: This figure illustrates the development of default rates and number of observations around the introduction of the digital footprint. The vertical line indicates October 19, 2015, that is, the date of the introduction of digital footprints. (Source: Authors)

Conclusions

In a paper with many provocative points, perhaps the most arresting is the authors’ analysis of the usefulness of digital footprint in assessing the creditworthiness of consumers who are “unscorable” because they have little or no credit history. The authors conclude that “the discriminatory power for unscorable customers matches the discriminatory power for scorable customers.” In other words, digital footprint data can be used to analyze correctly the creditworthiness of consumers with little or no credit history. If this conclusion is correct, it has an important implication:

Given the widespread adaption of smartphones and corresponding digital footprints, the use of digital footprints thus has the potential to boost access to credit for some of the currently 2 billion working-age adults worldwide who lack access to services in the formal financial sector, thereby fostering financial inclusion and lowering inequality.

The authors note that their conclusion is an innovation path that some companies in the FinTech space are already following. These startups, the authors note, “have the vision to give billions of unbanked people access to credit when credit bureaus scores do not exist, thereby fostering financial inclusion and lowering inequality.” This paper’s findings clearly support that overall vision, for they suggest a deep well of information value may indeed lie untapped.

In closing their paper, the authors refer to something known as the Lucas critique, which argues that individual actors consider potential policy changes in their behavior, i.e., that the relationship between people and policies is dynamic, and that one cannot analyze the impact on the relationships between people and policy without first understanding the forces that shape people's daily behavior. This critique is relevant, for if digital footprints were to become widely used indicators of creditworthiness then people might alter their online behavior to leave a better footprint—something much easier to do than to alter the payment behavior that generally shapes credit report scores. Thus, the authors note that the digital footprint “might evolve as the digital equivalent of the expensive suit that people wore before visiting a bank.” Of course, this kind of change is easier said than done, and it may turn out to be difficult to change one’s digital footprint than one might imagine. Moreover, should digital footprints become more important in credit decisions, it is likely that regulators would take a greater interest in them, which could also alter their value.

Reflecting on this paper, we recalled a conversation with the Chief Innovation Officer of a global technology services firm a few years ago. The subject of our discussion was data privacy, and he explained to us that the internal debate his firm was having not just about defining what could be done with the data they collected then but what might be possible in the future. His comments are worth quoting:

Imagine that in 2018, as part of a cellphone warranty registration process, we ask for your favorite color. Without much concern, you casually answer “orange.” Now imagine that in 2023, our analytics team figures out that people who like orange are significantly more likely to submit fraudulent warranty claims, so we decide you can’t buy an extended warranty on your new phone in 2024. Now imagine that law enforcement agencies conclude that people who like orange are more likely to commit other crimes as well and ask for our list of orange-loving customers. What worries me is not what we can do with all your personal information today but what we—or others—might be able to do with it years from now. That’s something we can neither predict nor write into any consumer agreement at the moment.

Considering all the information about us that digital footprints may contain, it strikes us that so many digital markers we leave behind us may seem inconsequential and ephemeral. The reality, however, may be very different. On the one hand, one can imagine digital footprints, as the authors suggest, expanding banking access to young people, immigrants, the working poor, and other populations traditionally left out of the formal banking system. On the other hand, one can imagine digital footprints being used as invisible credit reports: decision support systems about which consumers have no power or even knowledge. Either, or both, futures may soon be possible.

The internet never forgets, someone once said. As this paper shows, the reality is much more complex than that. The internet not only never forgets, it is also constantly learning. As it does, it continuously finds new value in old data. What was of little value yesterday may turn out to be priceless tomorrow. Moreover, as this paper well illustrates, we exist in both physical and digital forms. As researchers and innovators continue their relentless push forward, it is the digital form that may prove to be the more powerful.

The Research

Tobias Berg, Valentin Burg, Ana Gombović, Manju Puri, On the Rise of FinTechs: Credit Scoring Using Digital Footprints, The Review of Financial Studies, Volume 33, Issue 7, July 2020, Pages 2845–2897, https://doi.org/10.1093/rfs/hhz099

Over the past two decades, the number of women in corporate top management teams has markedly increased. As shown in Figure 1 below, a 2020 study noted that women held 26.8% of senior leadership positions in the S&P 500. As more and more women take senior roles, researchers are providing valuable updates on the impact women have when they reach senior positions. A paper from Qiang Wu (Hong Kong Polytechnic), Wassim Dbouk (AUB), Iftekhar Hasan (Fordham), Nada Kobeissi (Long Island University), and Li Zheng (General Electric) follows this line of inquiry in a novel way. Their research looks at the impact women have on one of the least studied C-suite positions: the Chief Technology Officer (CTO).

Figure 1: Catalyst, Pyramid: Women in S&P 500 Companies (January 15, 2020).

The CTO is a relatively new role. It is an evolution of the Research & Development leader that has a long history, and it arose more or less in tandem with the rise of the Internet and the wave of innovation it unleashed. As the authors note, the growing clout of CTOs "has been mainly driven by the increasing strategic importance of technology and innovation for corporate survival and competitive advantage.” CTOs’ responsibilities generally include monitoring new technologies, evaluating their fit for commercialization, assessing potential new products, as well as overseeing research new projects to ensure that they meet the innovation needs of the company. Especially because of this last responsibility, CTOs play an important role in shaping overall innovation strategy.

In framing their study, the authors observe that research has shown that male leaders tend to favor a "transactional" leadership style that is "top-down, command and control, and task-oriented." Women, however, tend to favor a "transformational" leadership style that is more democratic and stresses communication, collaboration, and cooperation. In simple terms, the "transformational" leadership style inspires and motivates followers to go beyond self-interest to work for the good of the organization.  

Other research has shown that the transformational leadership style has four interrelated components: idealized influence (charismatic role-modeling), inspirational motivation (articulating an appealing and/or evocative vision), intellectual stimulation (promoting creativity and innovation), and individualized consideration (coaching and mentoring). As noted earlier, studies have shown that female executives tend to favor this approach, as well as seeing themselves as less hierarchical, more collaborative, and more democratic. Importantly, the authors note, women are also more likely than men to combine feminine and masculine leadership styles in an “androgynous” style that is predominantly transformational.

Given both the rise in the number of female CTOs and the different approaches they bring to the role, the authors set out to understand their impact on innovation output. In other words, with all else being equal, do female CTOs make companies more innovative? If so, how?

The Study

To assess the impact of female CTOs, the authors looked at a composite model of innovation in a set of American companies from 1991 to 2010. Senior leader demographics were drawn from the BoardEx database, which provides top executives' and boards of directors’ personal background information, including gender, age, and education. Patent data, a common foundation of innovation research, came from the United States Patent and Trademark Office. Compustat provided firms' accounting information. In all, the authors compiled 5,408 firm-year observations within the nineteen-year span noted above.

As Figure 2 below illustrates, the number of female CTOs increased since 1991 and peaked at 33 in 2007. Moreover, female CTOs were concentrated in high-tech industries, with chemicals and allied products having the largest female CTO representation, followed by business services, computers, industrial machinery and equipment, and electronic and other electric equipment. Female CTOs’ ages varied from 42 to 75, with most in their mid-50s. Female CTO tenure also varied, with a minimum of two years and a maximum of 13 years.

Figure 2: Distribution of female CTOs by year, industry, age, and tenure.

With the female CTO population defined, the authors looked at innovation output from the companies in the study. Output was measured by looking at the absolute number of patents granted and various related aspects of patent creation, e.g., team composition and the number of internal and external citations received in subsequent years on all the patents filed by a firm each year.

Any study of this nature requires a comprehensive set of controls to enable accurate comparisons, and this research is no exception. The authors controlled for an extensive set of factors, including company size, financial strength, industry competitiveness, firm capital expenditure levels, and a firm's culture of innovation. 

The Findings

Consistent with their principal hypothesis, the authors found that firms with female CTOs are, by and large, more innovative. These firms not only produce more patents, but their patents are also more often cited by others inside and outside the firm. As the authors note: "Our result indicates that female CTOs are expected to have a rate 1.826 times higher for patent counts compared to male CTOs, holding the other variables constant in the model." With respect to patent citations, the data indicates that "female CTOs are expected to have a rate 1.741 times higher for patent citations compared to male CTOs, holding the other variables constant in the model.”

There are interesting nuances that expand on the paper’s overall findings. For example, as is probably the case with men, female CTOs are most effective when there is a female CEO as well. As the authors note:

We find that the coefficients on Female CTO and Female CEO are both positive and significant, suggesting that both female CTOs and female CEOs have independent and positive effects on innovation. More importantly, we find that the coefficient on the interaction term, Female CTO*Female CEO, is positive and significant, suggesting that the effect of female CTOs on innovation is more pronounced for firms with female CEOs.

Crucially, having a female CEO or CTO are independent drivers of higher innovation that are amplified when both leaders are female, the authors suggest.

As one would expect, female CTOs are most effective in firms with an innovative corporate culture, in firms where the CTO has more power, and in firms with strong competitive positioning. The same goes for CTOs with a technical education, which, I suspect, is the same for male CTOs, given the nature of the role.

Perhaps the most interesting set of findings are those related to how innovation happens at firms with female CTOs. The authors hypothesize that "if female CTOs encourage teamwork and collaboration, we expect that their patent applications are more teamwork-based with more collaborations among researchers." This is indeed what the data indicate. A firm with a female CTO has, on average, 0.747 more inventors than a firm with a male CTO. This result suggests to the authors that more collaboration and greater teamwork occur when innovation activities are led by a female CTO.

Considering the typically high levels of uncertainty throughout the innovation process and the risks associated with such investments, the authors wonder whether it is possible that female CTOs are less likely to choose "risky" innovation efforts. This is an important question given the belief that female leaders are sometimes less risk-averse than men — the so-called female tolerance of failure theory. To assess this dimension, the authors looked at whether patents granted to teams led by female CTOs "exploited" existing innovations (less risky) or "explored" new innovations (more risky). Perhaps surprisingly, the data showed that "that female CTOs are not more risk-averse than male CTOs when they lead high-risk innovation activities." In other words, there is no significant difference in terms of their risk preference between male and female CTOs. Indeed, the authors found that there is a positive relationship between female CTOs and exploratory innovation in manufacturing, which is to be expected since that sector leads all others in exploratory innovation projects. In sum, the data show that rather than being risk-averse, female CTOS are better than male CTOs not only for exploitative but also for explorative innovation efforts.

In a footnote, the authors expand on the finding noted above:

Our finding of a positive relationship between female CTOs and innovation does not support the tolerance-of-failure theory for female executives. One explanation is that theory of female risk aversion is based on the female population in general. Since few women hold top management positions, female managers are unlikely to be representative of the whole female population, but are instead more likely to represent a special group of competitive women who choose to pursue careers in male-dominated professional management jobs. Several studies argue that in a predominantly male environment, women in top management positions think and behave like men, and gender differences disappear. The economics literature even rejects the theory of female risk aversion in general. Therefore, it is not surprising that we find no supportive evidence for the female CTO tolerance-of-failure argument.

Reflecting on their findings, the authors note that other research has shown that transformational management per se is more conducive to innovation in large organizations. Perhaps because female CTOs adopt that approach more often than their male counterparts, they positively impact corporate innovation to a higher degree. Put simply, their affinity for transformational leadership means leads female CTOs to be more innovative than male CTOs. (If this is indeed a major driver of the findings, then it stands to reason that male CTOs who also favor that approach could yield similar results — a topic the researchers do not address)

Conclusions

In closing their paper, the authors note a study showing that the average tenure of a company on the S&P 500 narrowed from 33 years in 1964 to 24 years in 2016, and is expected to shrink to 12 years by 2027. At the current turnover rate, about 50% of the companies in the S&P 500 today will be replaced in the next decade. As is widely accepted today, innovation is one of the key factors that determines whether or not a firm stays on that list.

Given the importance of innovation to a corporation's survival and given the increasing number of women in senior leadership positions — including technical ones such as CTO — understating the way in which women leaders impact their firms is critical. This is the first study to look at the impact of female CTOs on innovation, which means that it is too early to reach definitive conclusions. Still, it does suggest that a natural affinity in female leaders for a more collaborative, less ego-centric, management style means that female CTOs may represent a "valuable and rare internal resource that can enhance corporate collaborative culture and provide distinctive insights throughout the innovation process."

If the above conclusion is correct, then firms have a vested interest in ensuring that newly selected female CTOs have the necessary technology-related background, power, and support structure in place to succeed. Unfortunately, as the authors note, such is more often the case in European firms than in U.S. companies, but perhaps that situation will change as more female CTOs make their presence felt. Indeed, given that women control the majority of household spending worldwide, perhaps their growing C-suite impact will lead to new innovations to fit a set of needs that may have gone unserved in the past. 

In closing their paper, the authors highlight that other researchers have found that "female executives engage less in value-decreasing acquisitions, use less debt, comply more with tax rules, and promote better financial reporting." As more women reach executive positions, this kind of research may help their employers maximize the value female leaders bring to their companies and society at large. This team's work is a strong and novel addition to that enhanced understanding.

The Research

Qiang Wu, Wassim Dbouk, Iftekhar Hasan, Nada Kobeissi, Li Zheng, Does gender affect innovation? Evidence from female chief technology officers, Research Policy, Volume 50, Issue 9, 2021, 104327, ISSN 0048-7333, https://doi.org/10.1016/j.respol.2021.104327.

The world’s headlines often include accounts of hacking brought about through something known as “phishing.” In this technique, fake emails are typically sent to a company's employees who then, through error or inattention, provide login information to the hackers. Of all the ways hackers break into computer systems, phishing is one the most common and yet, in theory at least, the easiest to prevent. As with passwords given away on a phone call, or laptops left unsecured, phishing illustrates how so often a psychological phenomenon leads to a major cybersecurity failure. Understanding how to ensure cybersecurity compliance, therefore, is a subject of much research and debate. Given the prevalence of the problem, however, there remains much to be discovered and improved in this all too human aspect of information security.

An extensive analysis from Yan Chen (FIU), Dennis Galletta (Pitt), Paul Benjamin Lowry (VaTech), Xin Luo (UNM), Gregory D. Moody (UNLV), and Robert L. Willison (Xi’an Jiaotong-Liverpool) brings a novel perspective to this issue by looking at information security through the lens of healthcare. Their shift in thinking is both instructive and illuminating.

The authors start their paper by noting what they believe are three shortcomings of much previous research on this topic. First of all, they argue, most analyses have focused on compliance and treated non-compliance as the lack of the former. This is too simplistic a view, the authors think, and non-compliance should be studied and dealt with as something with its own unique characteristics and dynamics. As they write:

Nearly twice as many studies focus on compliance versus non-compliance, and over twice as many studies focus on adaptive versus maladaptive outcomes. Notably, this is more than a mere research gap; the problem is the explicit or implicit assumption that the motivations and reasons for non-compliance are merely the opposite of compliance, which is not true. 

The second flaw in a lot of the previous work, in their opinion, is that it has looked at compliance through a largely cognitive lens and omitted any discussion of emotion from the subject. In other words, researchers studied whether employees did or did not comply with a security practice but did not dig deeper into the emotional and psychological dimensions that are reflected in, or even caused, the outcomes the researchers observed. The authors argue that "substantial research indicates that employees in security settings operate with bounded, not perfect, rationality, and that researchers therefore must consider emotions, not just cognitions, to understand how employees make security decisions."

Lastly, mainly due to the first critique noted above, past research has not looked at the "tipping point" at which an employee moves from complying with security protocols and not complying, a moment whose dynamics, the authors believe, should be examined and understood. As they note, "a theoretical and empirical account should explain how and why these outcomes are different, identify the point at which a normally compliant employee may suddenly choose to become noncompliant, and determine why such choices are made."

Having presented their critiques of past attempts to understand why employees often fail to comply with information security policies (ISPs), the authors introduce a very different approach for considering the problem at hand. They suggest adopting something called the Extended Parallel Processing Model, a construct developed by communications researcher Kim Witte in the 1990s to understand how people deal with healthcare threats. EPPM suggests that when faced with a threat to their physical well-being, people think about it along two dimensions. The first dimension is rational and is based on what power, if any, the person has in response to the threat. The second dimension is emotional and is based on the fear generated by the external threat. The person’s reaction to the threat, EPPM predicts, will be based on a combination of both rational and emotional reactions:

When perceptions of a threat are strong and perceived levels of efficacy are high, the model predicts self-protective behavior. When perceptions of a threat are strong, but perceived levels of efficacy are low, the model predicts maladaptive denial or rejection of protective behaviors. By asking questions like the ones above, people in an intended audience can be classified as having either high or low levels of perceived efficacy and either high or low levels of perceived threat.

Figure 1: EPPM dimensions and responses (Source: HCCC).

The authors base their research on the idea that just as EPPM helps predict reactions in healthcare, it can also do the same in cybersecurity. Consider two possible reactions to a CEO video sent to employees in which they are told about the disastrous hacking of a competitor. Employee A is convinced by the CEO that (a) the threat is real (and thus generates a high degree of fear) and that (b) her individual actions really help keep the company safe. EPPM predicts she will follow ISPs closely and thus pose a low cybersecurity threat to the company. Employee B, however, concludes that the problem is real (high fear) but what he does really does not matter since hackers “can always figure out to to get into our systems” (very low efficacy). EPPM predicts this employee will be a high cyber risk to the organization. He will, if EPPM is right, decide that nothing he can do will make any difference in the firm’s cybersecurity posture.

To test whether their ideas about EPPM in cybersecurity have merit, the authors conducted two studies:

Study 1 examined ISP compliance/violations of 411 employees using self-report surveys. Study 2 required 405 employees to respond to manipulated scenarios (or vignettes) of ISP violations, asking whether they would do the same in their workplace (i.e., be noncompliant) or not (i.e., comply). It was important to examine these differences, because several behavioral security researchers have claimed that hypothetical vignettes are superior to self-report surveys in organizational security compliance and non-compliance contexts.

By the way, it’s important to remember that “fear” in this context is clearly not the same fear in the original EPPM application. It refers, in the authors’ work, to a sense that a cybersecurity failure could endanger not just the health of the firm but the employee’s own income, position, or even career.

Their main findings are summarized below.

Finding: The optimal condition in employees is one of high efficacy combined with high fear

For the fear dimension to trigger high levels of compliance, "security risks must be perceived by employees as personally relevant, and those employees should receive appropriate training toward that end." This elevated fear state is recommended because "when such employees encounter a meaningful level of ISP-related threat at work, they then internally weigh their efficacy to determine whether they can comply with ISP requirements to thwart the security threat." Employees who see hacks and information thefts as personally dangerous to their income and careers, and who know what specific ISP actions to take, are the ones most likely to do what the company hopes they will. They are, note the authors, "more likely to recognize and report suspicious emails to IT, encrypt attachments, recognize attempts to use social engineering to deceive them, use and remember more complex passwords, report suspicious coworker computer behavior, make use of a VPN, log off when leaving one's workstation, and pay attention and respond to system or browser warning messages.”

Finding: Increasing the fear level in settings with very low efficacy levels is ineffective or even counter-productive

When companies increase the fear level in employees but fail to raise the efficacy level, EPPM predicts that they will, at best, pursue average compliance measures or, at worst, become especially non-compliant. Like a heavy drinker who decides alcoholism is encoded in his genes and so accepts the disease as his destiny, an employee with high fear and very low/zero efficacy can easily become a major internal cyber risk:

A common example is when employees know about the severity and vulnerability of security threats — often through sensational news stories—but in fact have no training or efficacy in responding to these threats. Thus, they feel overwhelmed and ignore the threat. For instance, employees may have attended training that raised their awareness, but they came away confused, stressed, or annoyed because they did not understand the highly technical material and were too embarrassed to ask questions (a common issue for nontechnical employees). A more serious reaction can occur when an employee views non-compliance as an excellent opportunity to "get back" at their boss or company for perceived mistreatment.

Finding: the tipping point between compliance and non-compliance is variable and dependent on the efficacy/emotion conditions

According to EPPM, when faced with a health threat, patients adjust their response continuously as new information about the threat level (and their own power) is obtained. For example, someone may initially take measures not to catch COVID but then, after seeing case numbers rise day after day, become convinced that nothing he can do will prevent infection, at which point he stops taking any safety precautions. The opposite is also true. Someone may not take any precautions against COVID until she reads about a study showing that wearing a mask significantly reduces infection risk at which moment she suddenly becomes compliant with mask mandates. Put simply, as long as a threat's efficacy level is higher than its fear level, people are more likely to take action (danger control response) to lower the risk than minimize it psychologically (fear control response). Likewise, in cybersecurity, note the authors, "the extent of the response elicited by perceived threat or by perceived efficacy determines the tipping point, which identifies which control process is dominant, and the success or failure of threat coping." 

The tipping point issue is as nuanced as it is critical for cybersecurity, the authors argue, because "studies have confirmed that for ISP compliance, a self-regulatory approach relying on shared security values and beliefs is more effective than a command-and-control approach." Furthermore, "when employees struggle with competing goals (sometimes caused by role conflict introduced by different management systems) and values, they are more likely to appeal to problematic emotions and react with non-compliance." In this situation, employees "may downplay the value of the prescribed ISP, resulting in a perception of low response efficacy and more non-compliance."

So far, the research findings support what EPPM would predict, but there were some areas where the researchers found outcomes that were different in cybersecurity than in healthcare settings. For example, EPPM suggests that when someone has high fear but little efficacy, they try to control or even deny their fear. For example, heavy smokers discount warning labels as "overblown" and continue to smoke. In their study, however, the authors found that when threats produced high levels of fear, even if efficacy levels were not very high, employees did try to increase their compliance efforts. Unlike in healthcare settings, then, the higher levels of fear did not "overwhelm" the response decisions. The authors surmise that “fear can simply be more overwhelming in a potentially life-threatening personal health context than in an organizational security setting, even one in which employees care about their work and their organization." That's the good news for cybersecurity managers. 

The bad news is that the study showed a greater tendency for non-compliance in some situations, even when the efficacy level was high. This outcome suggests to the researchers that situational context might impact the EPPM model more in cybersecurity settings (with their higher dependence on group/company characteristics) than in healthcare (with their higher dependence on individual/personal characteristics).

Based on this first-of-a-kind study, the authors suggest that it may be straightforward to increase compliance through Security Education and Training Awareness (SETA) programs that focus on (a) making employees aware of threats they were not previously aware of and (b) increasing coping efficacy to deal with threat-generated fear by explaining systematically how to deal with specific threats. 

In addition, the authors suggest that cybersecurity managers should evolve their training programs to decrease unwanted responses to the various EPPM scenarios. To do so, "they may have to further account for other related contextual factors such as job roles, organizational commitment, type of security threats, organizational differences, industrial differences, and individual differences, to reduce maladaptive coping behaviors."

Furthermore, given the critical role that fear plays in shaping compliance and non-compliance, "it would also be interesting to determine whether other contexts, such as cyberphysical systems (e.g., drones and autonomous vehicles), elicit stronger fear due to the enhanced possibility of physical harm." At the same time, the authors emphasize the need to set their research into the correct cultural context, since, for example, "studies have shown that, compared to Western users, Chinese users are much more prone to uncertainty avoidance, perceive greater power distance from their managers, and adhere more strongly to collective norms." Thus cyber-EPPM [my term] might operate differently in different countries and cultures than it does in its traditional healthcare application.

As noted earlier, this research makes interesting conceptual, methodological, and even practical contributions to how managers should think about cyber compliance and non-compliance. Their novel use of EPPM has expanded the context in which ISP design and efficacy can be considered. As such, one hopes the work will get a wide reading and serious consideration by cybersecurity leaders and their business colleagues across all the many settings where information security continues to be a critical part of operational risk identification and management.

The Research

Yan Chen, Dennis F. Galletta, Paul Benjamin Lowry, Xin (Robert) Luo, Gregory D. Moody, Robert Willison. Understanding Inconsistent Employee Compliance with Information Security Policies Through the Lens of the Extended Parallel Process Model. Information Systems Research 0 (0) https://doi.org/10.1287/isre.2021.1014

The Interview

The DEI Monthly

Issue 1.1: A conversation about fear, power and passwords

Listen now (20 min) | DEI Monthly presents brief conversations with the authors of the most-viewed posts featured in the newsletter. In this episode, I speak with Dennis Galletta, Ph.D. and Greg Moody, Ph.D. about their research on the psychology of cybersecurity. The research summary is linked below…

Listen now

2 years ago · 2 likes · Carlos Alvarenga

One of the defining characteristics of almost every corporate organization is the existence of a hierarchy. For all the discussions about “flat” organizations since the idea first appeared in 1950, the reality is that most companies operate with some sort of hierarchical structure. Depending on the company, the hierarchy may be a positive structure that allows the company to fulfill its mission or it may be a negative structure that enables bureaucracy and disables innovation.

Given the impact that hierarchies have on company performance, this topic has received extensive study and analysis from researchers, consultants and executives. Generally speaking, that body of work has looked at the operational mechanics and implications of specific hierarchical forms. The emphasis has mostly been on understanding how specific hierarchic structures impact workers and companies for better or worse. A paper from Siyu Yu (Rice), Lindred L. Greer (Michigan), Nir Halevy (Stanford),  and Lisanne van Bunderen (Amsterdam) adds to this extensive body of work with a novel analysis of a topic previously ignored: the impact that hierarchy visualization has on its members. The critical question: does a hierarchy’s perceived “shape” affect employee behavior and overall organizational performance?

For their study, the authors focused primarily on two hierarchy shapes: ladders and pyramids. The authors note that individuals commonly use ladder shapes to “mentally represent hierarchical distributions that come with (a) salient vertical categorical distinctions between members in different ranks, (b) a narrow base and a relatively equally narrow top, and (c) in its most extreme form, a group structure in which each group member occupies a distinct rank in the hierarchy.” In contrast, individuals commonly use pyramid shapes “to mentally represent hierarchical distributions in which (a) few group members control most of the group’s resources, (b) the lower ranks form a wider base and the higher ranks form a narrower top, and (c) in its extreme form, a group structure in which a single individual occupies the highest and all the other group members occupy the lowest rank in the hierarchy.”

These two shapes are important because they illustrate important features of hierarchies gleaned from past research: stratification and centralization. A ladder represents a hierarchy that is most effective when each member of the group has a distinct rank relative to other members. A pyramid represents a hierarchy whose maximum value is reached when one member has the highest rank and most other members share a lower position. The shapes are also important because of a concept known as the “principle of construal,” which stipulates that people’s interpretation of the circumstances around them has a direct and material impact on how they think and behave. In other words, we adjust our thoughts and actions to fit the world around us as we imagine it; therefore, how we imagine our organization’s hierarchy works affects how we think and act inside of it. 

The Studies

The authors conducted a series of studies to discover the impact that hierarchy shape visualization has on its members. In the first study, 345 working adults in the U.S. were recruited and randomly assigned to one of three groups (ladder, pyramid, or control). In all three groups, participants were asked to choose which of four shapes captured how they think about the concept of “group hierarchy.” The options were: “pyramid,” “ladder,” “circle,” and “square.” The circle and square options were added, the authors note, "based on previous research that utilized them to visually represent equality and a highly steep hierarchy respectively."

All participants learned about an organizational model in which employees can potentially occupy one of five different ranks. In the ladder group, each employee occupied a different rank. In the pyramid group, one person occupied the top rank and the remaining four occupied the bottom rank. Participants in both the ladder and pyramid conditions then chose which of the four shapes best captured the kind of hierarchy presented to them.

As expected, most participants chose either “pyramid” (69.44%) or “ladder” (26.39%), and only a few chose “circle” (2.78%,) or “square” (1.39%), suggesting that pyramid and ladder shapes "are indeed the two most common mental representations people have of group hierarchy." 

In the second study, 380 working adults saw two figures (reproduced in Figure 1 below), one representing a ladder and the other representing a pyramid. They then indicated how similar the structure of their own work environment was to each of the two shapes using scales ranging from 1 (“not at all similar”) to 7 (“very similar”). The participants then completed a survey that looked at a variety of characteristics that define the quality of social relationships at work.

Figure 1. Illustration of ladder and pyramid used in Study 2. (Source: Authors)

From the survey data, the authors found that employees whose work hierarchy resembled a ladder experienced more conflict at work and had lower quality social relationships overall. As the authors explained in a summary of their paper:

Subjects who perceived their working group as a ladder, the researchers found, were more likely to compare their rank and station with others. Their relationships were also weaker: when asked whether they trusted their team members, most subjects disagreed or strongly disagreed. When asked whether they thought about if they were better or worse than their colleagues, they agreed and strongly agreed. These comparisons and lack of trust indirectly correlated with lower performance levels, the research showed.

Surprisingly, the authors did not find these negative effects for employees who perceived their hierarchy as similar to a pyramid. The authors hypothesize that one possibility that explains this difference is that, "because hierarchies can have both positive and negative effects, these opposite effects may have counteracted one another in pyramids, resulting in an overall null association between pyramids (but not in ladders, in which the negative effects plausibly outweigh the positive effects).

Building on the findings of Studies 1 and 2, Study 3 explored how mental representations of pyramids or ladders affect social comparisons with other employees and, again, relationship quality. In this study, the authors looked at 221 work groups from two Dutch financial services firms (1,717 employees in total). Relationship and social comparisons were reported directly by group members. Group performance was rated by managers. The authors themselves defined the shape of each group’s hierarchy based on rank distribution within each group.1

In Study 3, the authors found that "the more work groups’ hierarchical structure was shaped like a ladder, the worse the relationships employees experienced—the less group members trusted each other and the more they compared themselves with one another. Moreover, such ladder-shaped hierarchies "diminished team performance indirectly, via heightened intragroup social comparisons." Unlike Study 2, which did not find a positive effect of pyramids, "Study 3 found that pyramids positively predicted relationship quality and negatively predicted team members’ propensity to engage in social comparisons."

The general findings of the first three studies were supported by two additional studies in which participants joined one of two fictional startups, each with a different hierarchical model. Studies 4 and 5 "demonstrated a causal effect, whereby a ladder enhances group members’ propensity to engage in intragroup social comparisons, thereby undermining the relationship quality within the group." 

Conclusions

Across all the studies, the authors found that the perceived shape of the hierarchy in which employees operated had a significant impact on how they behaved at work and on overall organizational performance. As the authors note: "Our findings consistently show that, relative to pyramids, hierarchies that individuals perceive to be shaped like ladders stir social comparisons within groups, thereby undermining relationship quality and group performance."

In their discussion of why ladders and pyramids have such different outcomes, the authors reach the following conclusion:

Compared with pyramids, the rank differentiation in ladders highlights rank differences between members within the group, thereby promoting intragroup social comparison propensity, and degrading intragroup relationship quality and group performance. 

This conclusion suggests that the functions and dysfunctions of hierarchy may depend on the hierarchy’s actual and perceived shape. Therefore, paying attention to how workers perceive and experience their hierarchical environment can shed light on why and how hierarchy impacts group processes and organizational outcomes. This impact is present whether the visualizations reflect an actual company or someone’s perception of that structure. “It can be created by both perception and actual rank, for example, job titles,” one paper author said in an interview. “So, as a practical implication, companies should think about ways to reduce the ladder system, such as with a promotion system that seems more like a pyramid, or by creating the mutual belief that upward mobility within the company is not a ladder or zero-sum.”

As noted earlier, this is the first study to look at hierarchy form effect on employee behavior, and the authors are careful to note the implications of their preliminary analysis. For example, "although we found that ladders negatively influence group processes and outcomes, it is possible that certain features of individuals and groups may lessen or even reverse the harmful effects of ladders." Relatedly, "the negative effects of ladders may be attenuated depending on where one is in the hierarchy, with those at the very top or bottom ranks experiencing the most intense negative effects of hierarchical shapes." Additionally, it is not clear what happens when different group members mentally represent the hierarchy similarly vs. differently on group processes and performance. Lastly, this paper does not consider how shape perception, or its impact, might vary across countries and cultures. For example, "in interdependent cultures where harmonious relationships are emphasized, the negative effects of ladders on relationship quality might be reduced."

Despite its limitations, this paper suggests strongly that managers, consultants, and researchers should pay especially close attention to group dynamics in groups whose members see their organizations as ladders more than pyramids. The authors note that this advice does not mean that ladders should be avoided at all costs; rather, "we encourage practitioners to consciously weigh the extent to which the benefits that come from a stratified incentive structure outweigh the potential costs that accrue when group members perceive their hierarchy as a ladder."

At the highest level, this set of studies reinforces our understanding of the power that employees' mental representations of their environment have on their thoughts and actions. It makes intuitive sense to think that the way in which we see the world around us can profoundly impact how we behave in it. Decades of research in psychology, economics, and neuroscience support that view. This paper reminds us that we often use shapes and patterns to make sense of our environment. In a world where most of us work in some sort of hierarchy, the shape we assign to it mentally may prove to be a powerful determinant of our individual behavior and collective experience. 

This week’s JBS attack saw the latest case of corporate cyber attack take over headlines all over the world, and it culminates a year that has seen digital ransomware activity rise to new heights during the pandemic. This recent spike, however, is but a continuation of a recent trend. Indeed, in both 2018 and 2019 ransomware insurance claims increased by over 100%. These attacks are not only more frequent, they are also more audacious in their demands. Back in 2019, the typical ransom demand averaged less than $10,000. This figure has been increasing alongside the success of the actors, and it culminated with CWT’s $4.5M payout in 2020 (followed closely by Colonial Pipeline's rumored $4.4M payout to restore its operations last month).

Figure 1: Propensity to be hit by ransomware across different industries (Source: Sophos)

Retail and education organizations are still the most common targets, notes one recent report, but ransomware agents, according to another study, are becoming more selective about their targets. The pandemic saw a rise in healthcare sector attacks, for example, and manufacturing and government are increasing favorites given their sensitivity to downtime and public impact, respectively. Ransomware is also beginning to shift to blended “extortion-and-ransomware” attacks, notes the first report:

Here, in addition to encrypting local files, the ransomware steals copies of sensitive files and the gang threatens to make the documents public unless the ransom is paid. When the ransom has not been paid, some firms have seen their data auctioned on the dark web with prices ranging from $5,000 to over $20 million. According to IBM, the ransomware gangs are targeting the ransomware amounts to the specific firm. Known ransoms ranges from 0.08 percent of annual revenues to as high as 9.1 percent.

As common as ransomware is, the exact nature of the threat is often not fully understood outside the technical community. Yet understanding both the nature of ransomware attacks and their constituent elements is a priority for leaders outside of the cybersecurity function. Unfortunately, most of the general press discussions of ransomware are, understandably, superficial, and most of the academic literature is too complex for non-technical audiences. However, a recent survey of the subject by Masoudeh Keshavarzi and Hamid Reza Ghaffary (Islamic Azad University) provides a thorough and helpful overview of what ransomware is and how it works.

Ransomware taxonomy

Ransomware, as its name suggests, is a species of malicious software (malware) that takes a user's resources hostage and demands some sort of payment in exchange for the release of the resources. It's generally accepted that the first ransomware attack was the AIDS Trojan virus distributed via diskettes in the late 1980s. Its malicious code encrypted user files and then demanded $189 for their release. AIDS Trojan was simple, technically, and flawed in many ways. In the decades since its appearance, the ransomware ecosystem and technologies have evolved tremendously. Indeed, it may be the most lucrative form of cybercrime today.

Figure 2: The authors’ proposed taxonomy of the extortion-based attacks (Source: Authors)

Ransomware itself is a type of scareware, which is malware whose basic mechanism is the creation of fear in the affected party in order to elicit a desired action. The authors divide scareware into three categories:

Class 1: Rogue security software

Rogue security software is one of the most known and common scareware types. Rogue software typically infects a machine and pretends to have access to all of its data or to be able to destroy it in some way. This type of ransomware is a kind of "cyber-bluff" [my term] that hopes a user will pay up for fear of data loss or data discovery by outside parties. As the authors note, the main difference "between rogue utility and classical ransomware is that rogue software does not typically deny access to the resources and damage the victim device."

Class 2: Ransomware

True ransomware is distinguished by the fact that it does actually take control of a target machine or network and prevents users from accessing them until the ransom is paid. Generally speaking, ransomware comes in two forms.

The first form is denial of data resources (DoDR), in which the ransomware blocks access to the data resources and requires victims to pay a ransom to regain access to their data. Crypto-ransomware is the most common form of ransomware, and modern forms do effectively what the AIDS Trojan virus did back in the 1980s: they stop you from accessing your machine and its contents until you pay the ransom. Other examples of DoDR ransomware include WannaCry, Petya, PetWrap, NotPetya, AnonPop, Ordinypt, and MBR-ONI. 

The second type of ransomware locks out resources but leaves the user data intact. Data may seem inaccessible at first glance, note the authors, "but the main difference with the previous group (i.e., DoDR) is that the data will not be tampered with or destroyed." Therefore, "it makes DoNR less effective at extorting victims compared with its counterpart." 

The most well-known version of this second class of ransomware is locker-ransomware that generally infects a mobile, IoT or cloud device and extorts a payment to unlock it. Once infected, these attacks can compromise operating systems, applications, services, user interfaces, and other utilities. For example, note the authors, the Trojan.Ransomlock.G virus "locks the user’s screen and displays a full-screen ransom note that covers the entire desktop."

Class 3: Leakware

The third type of ransomware is known as leakware or doxware. This class is an "evolution" of cyber blackmail since the threat is the release of sensitive or secret information into the public domain. Doxware "utilizes the mechanisms of spyware and info stealer, and in some cases, it combines them with ransomware methodologies, such as cryptography or locking." Of course, doxware is unique in that rather than making data unreachable, it makes the data reachable to anyone unless the victim pays up. Leakware may be the most dangerous of all the three classes, note the authors:

...Leakware is more dangerous than all breeds of ransomware, as backup strategies cannot mitigate the damage caused by it. What makes the Leakware worse and more persistent than other extortion-based attacks is that even if the victim pays the ransom, she/he may still be threatened because a copy of the data is in the hands of adversaries.

Ransomware progression

After outlining the three classes of ransomware, the researchers move to the most useful section of their paper: the sequential analysis of a ransomware attack.

Figure 3: The authors’ proposed attack chain for all types of ransomware (Source: Authors)

Phase 1: Infection

All ransomware attacks start the moment the bad code enters a victim's machine or network. Although there are many ways to get the code into a victim's machine, most ransomware attacks employ spam phishing or what are known as exploit kits (basically a set of pre-determined software actions). The spam emails "contain either malicious attachments (e.g., a Microsoft Office document inclusive of macro, a PDF file with JavaScript) or a link to a compromised website." Some methods need a user to click to download the ransomware code, but some recent attacks do not.

As noted, the first known ransomware, AIDS Trojan, used infected diskettes to deliver its malware. But with "the extension of botnets and their prosperity in sending massive amounts of spam emails, the rate of adopting spam emails has been raised for distributing malware to as many users as possible." Moreover, new variants such as "cryptoworms" have emerged that act more like true viruses: they self-propagate without the need for any user activity and spread with amazing efficiency. As one 2016 ransomware report noted:

When Locky, which used infected Word files to spread ransomware, was brand new, there were reportedly 100,000 new infections per day; at one point there were between one to five new endpoint infections per second. If only one-fourth of the daily 100,000 victims paid the ransom of .5 bitcoins, which is about $213 today, then the cyberthugs were pulling in over $5 million per day.

Phase 2: Installation

"After the malcode is delivered to the victim machine through aforementioned vectors, ransomware enters the installation phase." At this point, ransomware must "install itself on the system and take control of the device, without attracting the attention of security software." In order to do this, ransomware has many techniques. For example, "process hollowing" and "process doppelganging" both allow malware to hide. In the former technique malicious code is injected into code, say a normal Windows routine. The latter is "a novel and very sneaky technique" for cross-process injection that inserts bad code into a small section of good code. The good code hides the bad code from security software, and so it runs undetected.

Phase 3: Communication

Once the installation of ransomware is completed, it starts to collect the victim’s specifics. This information, "which is later exfiltrated, includes the victim’s IP address, location, operating system, version of browser and its plugins, security tools installed on the device, and so on." Some ransomware codes need to communicate with their command and control (C&C or C2) server in this phase. This communication "is accomplished mostly with the intention of exchanging encryption key and receiving required instructions to keep on attacking."

Phase 4: Execution

In Crypto-Ransomware and Wiper attacks, an additional step is required: locating the most important files that should be attacked. The search process "can be as simple as seeking files with specific extensions to more complex procedures that consider the last accessed files or the entropy of the files." After scanning all directories, the malicious operation begins. Typically, "only files that are matched with predefined extensions (or conditions) are encrypted or manipulated." While all this is going on, many ransomware tools can also steal credential/login information. For example, "HDDCryptor leverages a utility tool for extracting credentials from the last session to reach the previously accessed network drives, which are not currently mounted." Once execution is complete, many ransomware packages install their own boot software and display the dreaded ransom note.

Phase 5: Extortion

Unlike many other types of malware, ransomware must alert the victim of its presence and provide further instructions. True ransomware does not seek to steal data and sell it to a third party, so it must collect compensation from the victim. Thus, in Phase 5 "a ransom note is displayed typically in the language based on the geolocation of the victim’s machine IP address (in the forms of background image, HTML file, text file, and so on)." In many ransomware attacks, the content of the ransom note is hardcoded in the malware binary itself, but other types may download it from the C&C server during Phase 3.

As with all fear-based attacks, social engineering techniques are commonly used in Phase 5 to persuade victims to pay. To do this, note the authors, "many Crypto-Ransomware and Wiper-Ransomware families specify a deadline that if the payment is not made by that time, the private key (necessary key for decryption) will be permanently destroyed or the requested amount will be doubled."

It seems quaint now that AIDS Trojan asked for payment via international money order or cashier’s check sent to a P.O. Box in Panama. Today, of course, attackers prefer an anonymous payment method. Indeed, the arrival of cryptocurrencies is thought to be a main driver in the increase in ransomware, since they eliminated one of the riskiest steps in the old business model. Indeed, notes consultancy Marsh, “in the first half of 2020, average ransomware payments increased by 60%, with bitcoin used for most payments.”

As efficient as collecting payment has become, money is no longer the only motivator of ransomware attacks: 

Many ransomware attacks are emerging for political purposes, spying, sabotaging, or camouflaging other types of malware. For instance, Unit 42 research group from Palo Alto Networks has spotted a new strain of ransomware called RanRan with a political motive instead of a monetary payment. It targets Middle Eastern organizations and extorts them by forcing victims to post a seditious political statement against a Middle Eastern political leader (the victim’s country leader). Rensenware is another example of Crypto-Ransomware with non-monetary incentive that looks more like a joke. In order to decrypt files, it asks the victim to get needed score in the TH12 ∼ Undefined Fantastic Object, a shooting game specified in the ransom note.

Phase 6: Emancipation

The authors note wryly that even in the ransomware game, survival depends on keeping your promises, so once payment is made the cyber hostages must be released. Some recovery methods are straightforward. For example, in "the Crypto-Ransomware attacks, after paying the ransom, a link to a victim-specific decryption tool is sent to the infected user." Other recovery methods are more complex. While it may be hard to believe, some ransomware pros happily provide online chat support to victims post-payment to ensure that the latter can recover as promised. 

Ransomware enablers

Throughout all six phases, several elements of today's technology landscape made ransomware possible and the authors thus highlight some of the most relevant items on this list.

Cryptography

Ransomware employs an assortment of techniques to block users’ access to their resource. The most common and interesting techniques is cryptography. Traditionally, cryptography is a serviceable technology to information security on the fly. It is defensive in nature, and provides privacy, authentication, and security to users. But this technology can be misused against security.

Social engineering

Social engineering is employed in ransomware attacks by stimulating user’s emotions, such as curiosity, fear, urgency, and so on to perform an action. Given the progress and complexity of security measures, it is possible to say that social engineering techniques are the easiest way to propagate malware. As mentioned before, one of the major vectors of pollution which requires user interaction is spam email. Most ransomware attacks are initiated by enticing victims into visiting a malicious webpage or opening an infected attachment in the phishing email through social engineering techniques.

Many ransom notes contain countdown timer, which means that criminal groups will increase the amount of ransom exponentially or, in some cases, will eliminate a number of files forever after the expiry date. In addition, the specified deadline will lead to discouraging victims from seeking therapeutic solutions and making mistakes in their decision-making. These menacing tricks will be more effective in companies and organizations such as hospitals, where ‘‘time’’ plays a crucial role, so that most victims consider themselves forced to pay ransom. WannaCry, SamSam, Defray, and BitPaymer are examples of this case.

Botnets

Spam botnets are one of the main pillars of cybercrime attacks on a large scale. Given that spam phishing email is one of the major infection vectors, the issue of spam sending botnets must be considered as one of the key actors involved in distributing ransomware. In brief, botnets are a network of hundreds to millions of compromised machines so-called zombie under the command of a botmaster. Botnet-based spam campaigns by programming a large number of distributed bots are able to transmit tens of thousands of spam emails to many users in a short time interval. Botnets have always played a primary role in many cyberattacks, including DDoS, banking Trojans, money mule spamming, ransomware, and crypto miners.

Anonymous networks

There are many reasons for employing anonymous network technology in cybercrime. The most prominent of them is the lack of traceability by law enforcement agencies and authorities. The use of anonymity in communications neutralizes the embedded blacklisting strategies in many security tools. This technology is clearly evident in the three phases (i.e., communication, extortion, and emancipation)...many ransomware families leverage a variety of anonymous networks like Tor and I2P to communicate with [the] C&C server for bypassing network traffic inspections.

Domain generation algorithms (DGAs)

Ransomware campaigns leverage various techniques to bypass security systems to prevent their C&C servers from taking down. This is where DGAs come into play as a secret mechanism for communicating with C&C servers. Applying this technology will make it harder to turn off C&C servers at least until the algorithm is completely reverse engineered.

Ransomware as a Service

Over the past few years, ransomware threats have soared dramatically. One of the reasons for this progress is the concept of ransomware-as-a-service, called RaaS. The emergence of RaaS platforms has equipped any users with malevolent intent with tools for creating their own ransomware variants, even sans previous knowledge. In this way, the main authors of ransomware focus on the development and promotion of malicious code and delegate its propagation to affiliates. By dint of easy access to RaaS, cybercriminals can comfortably move to the website providing RaaS and with little effort build their own ransomware variant. The RaaS provider groups pocket a portion of ransom revenue for every successful infection.

Defending Against Ransomware

Having thoroughly analyzed ransomware and its principal enabling technologies, the authors then note the various defense mechanisms possible at each stage. I summarize each briefly below.

Defend at the infection phase

The most operative strategy for circumventing any attack relies on preventing that in the first place. Understanding how various strains of ransomware infect devices is crucial for counteract such threats. Because most ransomware species are being delivered through malicious attachments or links included in phishing spam emails, analyzing, and protecting emails is an essential subject. 

Defend at the installation phase

Regardless of what the infection vector is, the malicious payload is delivered to the victim’s system, either as an executable binary or as a script or macros embedded in the file, and needs to be installed to continue the attack process. The best defense strategies in this stage are file and process monitoring at the endpoint.

Defend at the communication phase

Many cyberattacks need to communicate with their command center in order to complete the attack process, propagate and infect more devices, or exfiltrate the victim’s information. Ransomware is no exception. In many of ransomware families employing asymmetric cryptography, since the key pair is generated only after a successful connection with [the] C&C server, if communication is disturbed by security measures or never established, the attack will not arrive at the destructive execution phase.

Defend at the execution phase

Although the defense in the execution phase is somewhat late, it may be argued that the most practical defense solutions for previously unknown ransomware families are deployed in this stage. DoDR ransomware requires files’ read and write operations to encrypt or tamper their content. One of the most effective and commonly used methods for defending ransomware in the execution phase is to monitor file system activity by various approaches...also, in Crypto-Ransomware class, a defense strategy is to take advantage flaws in the design and implementation of cryptography algorithms. In the case of using symmetric cryptography, since the key remains in the victim’s machine until the user is online and the key is sent back to the C&C server, memory forensic tools can be leveraged for memory dump.

Defend at the extortion phase

The gangs behind ransomware attacks not only look for secure and untraceable, but also easy payment and exchange methods for victims. The most prominent digital currency used in cybercrime is Bitcoin. However, the footprint of Monero has also been seen in Internet attacks, especially crypto miners. Assenting victims to pay ransom demands can assist to fortify hackers behind ransomware, whilst there is no guarantee of recovering data and repeating this attack on the same victim. While many of the destructive operations have been carried out before the extortion phase and access to resources has been prevented, this stage can also provide an opportunity for defense. 

Defend at the emancipation phase

Identifying ransomware family via information included in the ransom note left on the victim’s machine or by analysis methods can be useful for unlocking or decrypting affected files. Because some of ransomware variants already have been discovered by third-party security companies and released their countermeasures. Therefore, all studies that have identified ransomware families can also be used at this phase...Because some of ransomware families have begun to target backup files as a part of their execution phase. Inspecting the accuracy of backups and recovering files from them in a testbed should be considered as a part of backup plan.

Conclusions

Ransomware is already an expensive proposition to victims. Indeed, about 75% of attacked entities do refuse to pay and as a result, suffer losses (on average) of about $623K per occurrence. Paradoxically, firms that do pay end up paying almost twice as much when all is said and done to return to their pre-attack state. This may sound counterintuitive but a recent report explains this outcome:

Well even if you pay the ransom, you still need to do a lot of work to restore the data. In fact, the costs to recover the data and get things back to normal are likely to be the same whether you get the data back from the criminals or from your backups. But if you pay the ransom, you’ve got another big cost on top.

These costs increase as local labor costs increase, notes the report, with Austria having the highest recovery costs in the world:

Figure 4: Average ransomware remediation cost by country (Source: Sophos)

Given that about one of four companies does pay some or all of the ransom, as well as the rise of anonymous payment mechanisms — not to mention RaaS — it's reasonable for most executives to expect ransomware attacks to continue in the near future. It has become, simply put, a cybercriminal’s most effective business model. Indeed, with average ransomware recovery costs quickly approaching $2M globally, there is a need for a clearer understanding of how these attacks move across their lifecycle as well as the technical and human phenomena that enable that progression. Moreover, while there are many technical issues that enable ransomware, the social engineering aspects, as I noted in a recent post, are sorely lacking in academic and practitioner focus. The taxonomy and sequential model provided in this research make a valuable contribution to those ends, and it would be useful to see this level of analysis and understanding inform the general business literature on this issue, which so often tends to focus on ransom amounts and worst-case scenarios rather than the tactical mechanics of, and solutions to, this growing problem.

The Research

M. Keshavarzi and H. R. Ghaffary. I2CE3: A dedicated and separated attack chain for ransomware offenses as the most infamous cyber extortion. Computer Science Review, Volume 36, 2020, 100233, ISSN 1574-0137. https://doi.org/10.1016/j.cosrev.2020.100233

In 2019, advertisers in the U.S. spent over $66B in commercials, so it would be safe to assume that this huge investment yielded a positive result to the ad buyers. This assumption is wrong, says new research from Anna Tuchman (Kellogg), Bradley Shapiro (Chicago Booth), and Günter Hitsch (Chicago Booth). Their newly published ad effectiveness analysis for 288 well-known consumer packaged goods should be required reading for advertisers everywhere.

Establishing the efficacy of television advertising has been a fiendishly hard task ever since the medium was invented. Over the years, different researchers have approached the problem in different ways. Some analyses focused on case studies, which were instructive but hardly indicative of overall patterns. Other researchers looked at the question from a high level, performing meta-analyses that were generally negative about the claimed impact of advertising but still left a lot of room for interpretation. New research from Tuchman and her colleagues is more promising.

Tuchman’s team started with sales data from 288 well-known brands and purchase data from 60,000 households from 2010-14, noted a recent summary of the work:

The researchers obtained data from Nielsen on the products’ sales at about 12,000 stores in the United States, as well as data on purchases by more than 60,000 American households, from 2010–14. The team also examined Nielsen data on traditional T.V. ads during the same time period. (Streaming services were not included in the data set.) For each commercial, Tuchman and her colleagues calculated the percentage of households in a particular local market that saw the ads (zones known as “designated market areas” or DMAs). 

To determine how much commercials drove sales, Tuchman’s team calculated a measure called advertising elasticity, which captures how much sales change with a given increase in ad exposure. This calculation is trickier than it sounds because ads can be affected by many things, such as seasonality and spot availability of products in a given DMA. The researchers adjusted their calculations to account for such variables and arrived at what they hope is an accurate indicator of ad elasticity.

After completing their analysis, the team found that the average elasticity across all products was only 0.0233 (while the median is 0.014).  In other words, doubling advertising implies a roughly 1% increase in long-run sales (for the median brand). This is a surprising result and much lower than previous elasticity estimates that ranged from 9-24%. Indeed, their new paper notes that the ROI of advertising in a given week, holding advertising in all other weeks constant, is negative for more than 80% of the brands studied. As the authors note:

The average ROI of weekly advertising is negative for most brands over the whole range of assumed manufacturer margins. At a 30% margin, the median ROI is -88.15%, and only 12% of brands have positive ROI. Further, for only 3% of brands the ROI is positive and statistically different from zero, whereas for 68% of brands the ROI is negative and statistically different from zero. These results provide strong evidence for over-investment in advertising at the margin.

Put another way, most money spent on advertising for these kinds of consumer products is wasted, or, as Tuchman puts it, the effect of the ad spend is “essentially zero.”

Of course, we all know that many, if not most, people today either skip or tune out commercials for many reasons, not least of which is the ubiquitous smartphone. Whatever the cause, the authors are curious to understand why advertisers continue to spend so much money with such poor results:

This raises an economic puzzle. Why do firms spend billions of dollars on T.V. advertising each year if the return is negative? There are several possible explanations. First, agency issues, in particular career concerns, may lead managers (or consultants) to overstate the effectiveness of advertising if they expect to lose their jobs if their advertising campaigns are revealed to be unprofitable. Second, an incorrect prior (i.e., conventional wisdom that advertising is typically effective) may lead a decision maker to rationally shrink the estimated advertising effect from their data to an incorrect, inflated prior mean. Third, the estimated advertising effects may be inflated if confounding factors are not adequately adjusted for. The last two explanations do not assume irrational behavior, but may simply represent a cost of conducting causal inference.

Indeed, the authors conclude that some advertising is probably better than none for only about 34% of brands analyzed, which, they note, should be “a threat to the survival of media markets in their current form, once knowledge about the small degree of TV advertising effectiveness becomes common knowledge.” Indeed, their main recommendation is that, given the reality of poor returns, advertisers should be skeptical of efficacy claims from broadcasters and agencies and conduct thorough analyses of all ad spend to increase their returns.

The paper ends with the following conclusion:

While improvements in targeting technology may theoretically increase the potential for higher advertising returns, they do not solve the underlying agency problems that allow sub-optimal advertising decisions to persist in the traditional TV advertising model we evaluate in this paper. Together with past research documenting similar results in digital advertising markets, our work should motivate economists to further study the managerial and agency issues in advertising markets.

Their recommendation for further analysis holds for CMO’s and advertising buyers as well, for even after accepting that this is only one study of a limited set of brands, the dismal ROI on television ad spend should generate a serious debate within the companies driving that $66B in ad spending. This conclusion is only enhanced when one looks at reports paid for by the TV industry itself, which are notoriously vague about the specific positive effects of their medium. This “white paper” is representative in that it offers 24 pages of positive claims for TV ad spend without ever providing a clear ROI baseline for their customers.

Given the analytical alternatives, perhaps unsurprisingly, the results of this new study are getting a good reading, with almost 4500 downloads as of this morning (an excellent figure for a journal paper). I’ll take the broad readership as a good sign that this team’s analysis has struck a chord and that there is more to come on this topic in the future.

The Research:

Shapiro, Bradley and Hitsch, Guenter J. and Tuchman, Anna, Generalizable and Robust TV Advertising Effects (August 2020). NBER Working Paper No. w27684, Available at SSRN: https://ssrn.com/abstract=3675236

The Interview

Back in April, the FT had an intriguing piece written by Stephen M. Fleming (Wellcome/Royal Society Sir Henry Dale Fellow at University College London) that examined the issue of metacognition in AI. The piece presented Flemming’s concerns about the most important way in which even the most sophisticated AI machines differ from people: they cannot doubt and so cannot question their correctness and, by implication, their decisions. As he notes:

AI researchers have known for some time that machine-learning technology tends to be overconfident. For instance, imagine I ask an artificial neural network — a piece of computer software inspired by how the brain works, which can learn to perform new tasks — to classify a picture of a dolphin, even though all it has seen are cats and dogs. Unsurprisingly, having never been trained on dolphins, the network cannot issue the answer “dolphin”. But instead of throwing up its hands and admitting defeat, it often gives wrong answers with high confidence. In fact, as a 2019 paper from Matthias Hein’s group at the University of Tübingen showed, as the test images become more and more different from the training data, the AI’s confidence goes up, not down — exactly the opposite of what it should do.

This problem matters because as AI machines become more common, their increasing familiarity has a significant downside because we tend to think of them as “optimized” and somehow more intelligent than humans. Flemming notes that this is a severe error, and that it is critical to consider the nature of AI systems carefully before we casually allow them to enter every aspect of our lives: 

The history of automation suggests that once machines become part of the fabric of our daily lives, humans tend to become complacent about the risks involved. As the philosopher Daniel Dennett points out, “The real danger . . . is not that machines more intelligent than we are will usurp our role as captains of our destinies, but that we will over-estimate the comprehension of our latest thinking tools, prematurely ceding authority to them far beyond their competence.”

That phrase “far beyond their competence” might seem strange to say about a machine, but many serious AI thinkers agree with Dennet, who believes that "AI systems that deliberately conceal their shortcuts and gaps of incompetence should be deemed fraudulent, and their creators should go to jail for committing the crime of creating or using an artificial intelligence that impersonates a human being.” In his view, AI systems should be forced to make public all of their risks and shortcomings, much as commercials for medicines do today. 

Other researchers agree:

As automated systems become more complex, their propensity to fail in unexpected ways increases. As humans, we often notice their failures with the same ease that we recognize our own plans going awry. Yet the systems themselves are frequently oblivious that the function they are designed to perform is no longer being performed. This is because humans have explicit expectations — about both the system’s behavior and our own behaviors — that allow us to notice an unexpected event.

Flemming, for his part, thinks that we should encourage AI to question itself and to admit when there is a known unknown:

Let’s imagine what a future might look like in which we are surrounded by metacognitive machines. Self-driving cars could be engineered (both inside and out) to glow gently in different colours, depending on how confident they were that they knew what to do next — perhaps a blue glow for when they were confident and a yellow glow for when they were uncertain. These colours could signal to their human occupants to take over control when needed and would increase trust that our cars know what they are doing at all other times. 

As Dennet argues in his book From Bacteria to Bach and Back: The Evolution of Minds, we also need to prohibit AI providers from hiding their creations from us. As he puts it: "when you are interacting with a computer, you should know you are interacting with a computer." In his opinion, any system that hides its identity as artificial, something so many AI-based systems try to do today, "should be deemed fraudulent, and their creators should go to jail for committing the crime of creating or using an artificial intelligence that impersonates a human being."

The idea of “introspective robots” may seem far-fetched but then so did self-driving cars not too long ago. Flemming, Dennet, and their peers are correct in their warnings about pre-assigning AI a level of ability it does not deserve. Furthermore, until they fully deserve the trust of humans, there is a strong argument for the AI self-revelation that Dennet proposes. All in all, there is no better time than now to discuss not just giving AI systems the ability for self-doubt and self-reflection but forcing them to share their doubts and true identity with us.

We are living through a moment in which the question of bias is front and center in many business discussions. There are good reasons for the focus on this persistent problem. We know, for example, that identical resumes with names that "sound black" receive less attention than those with white-sounding names. We also know that female entrepreneurs face a harder path to getting startups funded than their male peers.

Because so much research has shed light on the potentially negative impacts of bias, an increasingly important question is what can be done to minimize the effect of bias in critical people-centric decisions. One of the more established response strategies is something known as blinding, which refers to the withholding of specific information, e.g., applicant gender or age, from a decision-maker until after a decision is made. This is a technique with a long history, and a famous case is the adoption of blind auditions by some symphony orchestras — a practice that started in the 1960s. The impact of blind auditions was such that by the 1990s 25% of symphony musicians were women — up from about 5% in the 1950s.

As with the symphony example, research has demonstrated that if job applicant demographic information is withheld from hiring managers, job applicants from underrepresented groups are more likely to get interviewed and, in some cases, to receive job offers.

Even though the literature on blinding is clear about its benefits, the practice remains the exception in business, which led Sean Fath (Cornell), Richard P. Larrick (Duke), Jack B. Soll (Duke), and Susan Zhu (Kentucky) to seek to understand why a technique with a long track record of reducing bias in decisions has not gained wider acceptance in the corporate world. Specifically, the researchers explored the factors "that might influence whether evaluators will choose on their own to use a strategy of blinding in their evaluations" and the efficacy of those factors once deployed.

The Studies

As noted above, despite the documented positive impact of blinding, very few Human Resources organizations have formal blinding methods in place. The authors confirmed this observation in a survey of over 800 HR managers. Overall, almost 60% of the managers said they were familiar with blinding methods in hiring, but only 19% said their company had any blinding policy in place and only 18% had ever worked for an organization that used blinding in HR decisions. Moreover, only 20% had ever been trained on how to use blinding in decision-making, though that number was slightly higher for respondents who worked for public sector organizations.

From years of research into the issue, the authors identify several reasons that stop managers who know the value of blinding from using it in their own decisions. One such factor is simple curiosity. The authors describe a study in which they asked participants to view a video of someone completing a specific task and then to gauge the person's performance. The participants had the option to see the personal profile of the person before the observation and grading. On average, half of the participants in this group chose to view the video. A second group, however, was asked if they should see the profile (before being asked if they wanted to do so). In this second group, approximately 90% of the participants agreed that seeing the profile could bias their judgment and chose not to watch the video. Thus, note the authors, "people have the insight that it is better to avoid certain information, but they need to trigger this insight by reflecting on what they should do.”

In a similar study, participants were asked to interview someone for a job. As with the study noted above, one group was given a simple choice: they could choose to see the applicant's name and photograph before the interview. A second group was asked the "should you see them" question. A third group, however, was given a third option: "to first judge the job candidate based on credentials alone and then see that person’s name and photo (with the option to revise the initial judgment)." The participants in this third group were more inclined to make a blind judgment first and see credentials after making their initial decision. Notably, only a small number of participants — fewer than 20% — revised their initial, blind judgment. In other words, "with their curiosity satisfied, most subjects chose not to adjust their assessments based on the potentially biasing information."

The authors note that in addition to curiosity, people also choose not to self-blind because “they honestly, but incorrectly, believe biasing information to be useful or helpful." In another (unpublished) experiment, the authors provided one group of hiring managers the option to see a candidate's professional headshot and credentials while another group had the option to learn a candidate's race and gender. The authors note that "even though a person’s photograph is very likely to reveal that information, we reasoned that the explicit option to learn a job candidate’s “race and gender” would be more likely to cue reflection about potential decision bias than the option to see a candidate’s “professional headshot”." In this study, 45% of the managers chose to see the headshot while 20% chose to see the race and gender information. The authors believe that "certain information, like a candidate’s name, headshot, or college graduation year, may fail to cue a desire to self-blind because the underlying, potentially biasing content — race, gender, age, and so on — is not immediately brought to mind."

After considering the results of various experiments across multiple studies, the authors conclude that "evaluators who can overcome or delay a curiosity-driven impulse to receive potentially biasing information about a target — and who understand that having such information tends to hurt rather than help decision-making — are more likely to choose to blind their own evaluations."

Conclusions

Given that so few organizations have adopted specific blinding policies, the authors end their analysis by suggesting two blinding-related techniques that can help reduce bias. The first approach is to "nudge deliberative thinking." Because curiosity is often the force pushing back against adopting blinding techniques, evaluators should be encouraged to reflect on their decisions and to consider how bias may be affecting the outcome. Just asking decision-makers to consider the question "What should you see?" can be enough to encourage self-blinding. Thus, managers should be trained to question what information they are using to make decisions — and when it is used — because this self-reflection stage can often be enough to make someone adopt blinding proactively.

A second technique is to "change the order of information." Thinking about when certain information is seen can also be a catalyst for self-blinding. For example, "managers can be asked to first perform a blind evaluation (such as evaluating an anonymized resume) and then receive the information that was hidden from view (the job candidate’s name, college graduation year, hobbies, and so on), with the option to revise their initial blind evaluation." In the same unpublished study noted above, the authors found that changing information order reduced initial decision bias and that less than 20% of participants elect to change a decision once they are shown a candidate's full profile. This finding suggests that a key to reducing bias is simply minimizing access to the kind of information that brings it to bear during decisions. 

The authors call this approach a "fair order" strategy, and it can be useful in many settings. For instance, venture pitches could be presented in two parts: first, a written description of the pitch (with information about the entrepreneur not included), and then a live presentation with the actual founder. As the authors note, "investors who first read and evaluate the blind version of the idea — the written description — and then see the pitch with the option to update their evaluation may be less likely to be swayed by the gender or the attractiveness of the entrepreneur in their final evaluation than those who learn about the idea from the pitch alone." While many venture funding processes look like this at first glance, the reality is that many VCs explicitly seek to be swayed by the founder profile right from the start of a pitch evaluation.

The good news is that researchers continue to make a compelling argument for the increased adoption of blinding in hiring, promotions, innovation leader selection, venture funding, etc. Adoption may be increasing slowly, but it is increasing. Until the practice gains wider institutional support, managers should consider self-blinding options, keeping in mind the possible negative effects of their own natural curiosity and the order in which candidate information is viewed. This evolution is a challenge for many managers because research has also shown that people sometimes choose to see potentially biasing information, due to the mistaken belief that such information may bias others but not them. Notwithstanding this uniquely human challenge, the authors make a solid case for at least considering how self-blinding can make our deliberations fairer and our decisions more equitable.

The Research

Sean Fath, Richard P. Larrick, Jack B. Soll, and Susan Zhu. Why Putting On Blinders Can Help Us See More Clearly. MIT Sloan Management Review. June 08, 2021. Access it here.

One of the more interesting debates in human resources over the course of the pandemic has been on rehiring employees who have left an organization. A quick search results in a long list of articles that present the pros and cons of what are typically called "boomerang" employees. The broad consensus of the articles in favor of the practice suggests that an employee who returns to an organization can help demonstrate that the job market is not as good as expected and that the company she returns to is a better alternative than others.

It is not difficult to imagine the reasons why boomerang employees may seem attractive to a hiring manager. Former employees are well-known to the organization, so the risk of getting an unexpected outcome should be lower. They understand the company culture and business, and thus they should require less time and effort from their rehire date to reach full productivity. Often, they are known to and liked by current employees. For these and other reasons, a practice that was once almost unheard of has been gaining in popularity, especially given the recent staffing shortages many companies are facing. In fact, many organizations, notably in the high tech and consulting sectors, maintain "alumni" networks to keep former talent aware of new opportunities and to make it easy for them to return to a former employer. 

Despite the seeming attractiveness of rehiring a former employee who left on good terms, behavioral consistency theory suggests that past behavior is a good predictor of the future. Someone who found reasons to leave an organization in the past is likely to find reasons to do so again. Likewise, an employee who was not worth keeping not too ago is unlikely to have transformed into a must-have human capital asset. 

Surprisingly, given the strong reasons for and against hiring former employees—and also the rising levels of boomerang hiring—serious analyses of the phenomenon are hard to come by. Almost all hiring literature on new employees assumes they are just that—new—and ignores what happens when former employees are rehired. Likewise, almost all literature on promotions assumes candidates are either (true) new hires or internal candidates. A new paper from John D. Arnold (Missouri), Chad H. Van lddekinge (Iowa), Michael C. Campion (UT Rio Grande Valley), Talya N. Bauer (Portland State), and Michael A. Campion (Purdue) provides what may be the first systematic analysis of boomerang employee performance, and it should become required reading for hiring managers considering bringing back former employees.

The authors set out to explore an important question: How do rehires really perform compared to internal hires and first-time external hires, both initially and over time? For their research, the authors used the employee lifecycle presented in Figure 1 below, and they explain the framework as follows:

Moving from left to right in Figure I, the boomerang lifecycle includes (A) an initial period with an organization, (B) a separation event, (C) time away from the organization (e.g., returning to school, working for another organization), (D) a rehire event, (E) post­ rehire performance, (F) a (potential) post-rehire promotion, and (G) a (potential) second separation.

Figure 1: Comparisons of the Boomerang Employee Lifecycle With Internal and External Hire Lifecycles (Source: Authors)

In addition to these events, the authors examined boomer­ang employees' post-rehire behavior in detail, "including their promotion rates (F) and the likelihood of a second separation from the organization (G)." They also used the framework to make "additional comparisons between rehires and the lifecycles of internal and external hires." 

The Study

The authors based their findings on an analysis of human resources data obtained from a large U.S. retail organization. As with many retail companies, the organization hires managers at various levels and has a high enough turnover average (30%, in this case) to make rehires a viable option. The company provided the research team data on approximately 31,000 employees hired/rehired into the manage­ment trainee position. 1,318 (4%) trainees were former employees who left and later were rehired, 20,850 (68%) were external hires, and 8,546 (28%) were internal hires promoted from a lower-level position within the organization.

The authors used supervisor ratings to gauge performance in five competency areas (time management, communication, leadership, drive for results, and flexibility) and ten responsibilities (e.g., sales growth, inventory management, and customer service). Supervi­sors also considered the accomplishment of specific goals in their overall ratings. 

The authors recorded trainees promoted to manager as well as how long it took for the promotion to be made. For internal hires, "time to promotion reflected the number of days from initial promotion into the manager trainee position to the date of promotion to assistant manager." For rehires, "time to promotion reflected the number of days from rehire date to promotion date." In addition, the authors tracked whether managers remained with or left the organization, as well as the mode of (voluntary or involuntary) and reason for the departure.

The Findings

Perhaps the most important question of the study was whether rehires would demonstrate higher, similar, or lower levels of performance before and after being rehired. The authors found that performance after rehire was generally the same as in past employment. To illustrate, "62% of managers received the same performance rating on their final performance evaluation of their initial tenure and their first evaluation after rehire." In other words, the findings were consis­tent with behavioral consistency theory: the performance of boomerang managers tended to remain the same (rather than increase or decrease) upon being rehired.

One of the study’s hypotheses predicted that boomerang managers whose initial turnover was voluntary would demonstrate better post-rehire performance than boomerang managers who had been fired. Consistent with prior research, "rehires who initially left voluntarily did, indeed, have higher average post-rehire job performance rat­ings than rehires who initially left involuntarily." Interestingly, the authors also found that the indicated reason for the initial departure was not a significant predictor of rehire performance.

Speaking of performance, the authors found that in the short term rehires performed about as well as new hires. However, external hire performance was better than rehire perfor­mance at Year 2, though this effect leveled off by Years 3 and 4, as shown in Figure 2 below. In practical terms, "rehires initially outperformed external hires by 1% on average, but external hires improved enough by Year 3 to outperform rehires by an average of 4%." When compared with the internally promoted managers, the results were broadly similar. Rehires performed similarly to internal hires at Year 1. Internal hires continued to improve more than rehires at Year 2, but, again, differences in performance change leveled off by Years 3 and 4. 

Figure 2: Type of Hire as a Predictor of Job Performance Over Time (Source: Authors)

Taken as a whole, the authors conclude, both internally promoted and newly hired managers performed better than rehires in the long run. The authors also found that that rehires were also more likely to leave or be fired than new hires or internal promotions, as shown in Figure 3 below.

Figure 3: Cumulative Survival Rate by Hire Type (Source: Authors)

Perhaps not surprisingly, the results showed that rehires' reasons for turning over a second time tended to be similar to their initial turnover reasons. What is surprising is that, as shown in Figure 4 below, despite the inferior performance and higher turnover, rehires were 1.24 times more likely to be promoted than internal hires. To the authors, the promotion data suggests that "organizations may use promotions to try to retain boomerangs, even though such employees may not be as effective as other employees over the long term."

Figure 4: Cumulative Likelihood of Promotion by Hire Type (Source: Authors)

Taken as a whole, the authors conclude, "these findings call into question some of the proposed benefits of rehiring former employees."

Conclusions

A 2017 study looked at what happens when NBA teams rehire athletes who had played for them in the past and found that boomerang players' performance decreased once they were back on their former team. In the different world of retail, this study found outcomes that were broadly similar. Rehires tend to underperform over time, and they tend to turnover for much the same reasons they left in the first place. 

As noted above, the findings question the advice presented by articles recommending that companies bring back former employees. For the authors, any claimed benefits "may be short­ lived," given that other types of hires tend to improve more over time and are more likely to stay with the organization. Consequently, the authors conclude their paper by highlighting the advantages of promoting exist­ing employees: "internal hires are less likely to tum over (20.9%) than both rehires (36.6%) and external hires (33.5%)." Furthermore, "internal hires tend to require lower start­ing salaries than external hires, and organizations are thought to feel less pressure to promote internal hires due to their commitment to the firm."

The authors do note that in a limited set of circumstances—short ramp-up times or on short-term contracts—a rehire may be the best option. Overall, however, this first-of-its-kind study concludes that barring those special situations, companies are better off promoting from within or hiring someone new than bringing back a former employee.

Your Thoughts?

This paper is a novel study that will require future research to refine and confirm its findings. However, it would be interesting to hear from Thematiks readers—an overwhelmingly executive audience— your take on this paper’s conclusions. Below is a link to a short 3-question poll on boomerang hires:

Take Poll

I will share the results in my next post.

The Research

Arnold JD, Van Iddekinge CH, Campion MC, Bauer TN, Campion MA. Welcome Back? Job Performance and Turnover of Boomerang Employees Compared to Internal and External Hires. Journal of Management. 2021;47(8):2198-2225. doi:10.1177/0149206320936335

'Tis better to have loved and lost than never to have loved at all.

— Alfred, Lord Tennyson

One of the most studied issues in business research is negotiation strategy. Researchers have examined this issue descriptively—seeking to understand how negotiations develop—and prescriptively—seeking to define the optimal approaches for various negotiations types and scenarios. One important issue within negotiation research is the value of any options that participants have during the negotiation process, e.g., accept, walk-away, etc. One of the most important options of all is what is known as the "Best Alternative to a Negotiated Agreement," (BATNA) which is simply the best option available to a negotiator should negotiations fail.

All options are defined by three dimensions: their value, their risk, and their expiration. Take, for example, someone who is granted early admission to her second-favorite college but has to respond before the response deadline of her favorite school. As in other scenarios, this applicant has to consider the quality of her second-favorite school (value), the probability that it will accept her (risk) and the last day early admissions acceptance is allowed (expiration). A job applicant with a written offer letter for a job he is willing to accept negotiating with a second possible employer has a similar set of dynamics to consider.

Research has shown that agents who have an acceptable BATNA generally do better in negotiations than those who do not have one. Strong BATNAs lead individuals to ask for higher prices and to take a more aggressive negotiating posture, for example. They are also willing to seek more value from their counterparty and to maximize the value of winning. In other words, having a positive BATNA will generally allow someone to maximize the value of a "winning hand."

Given the positive outcomes that a good BATNA can create, it is somewhat surprising that researchers have not examined what happens if a good BATNA is lost during (or just before) a negotiation, something that occurs with regularity. Does losing a good BATNA lead someone to take a lower price or assume a weaker negotiating stance? Does it impact negotiating aggressiveness or willingness to “settle for less?”

New research from Garrett L. Brady (LBS), M. Ena Inesi (LBS), and Thomas Mussweiler (LBS) looks at these questions in a comprehensive set of studies with some surprising results. The authors’ interest lies in part from the observation that, from a rational perspective, negotiators who lose an alternative are in the same position as those who never had that alternative in the first place. Logically, the authors note, "one may thus expect a lost alternative to have little effect on negotiation processes and outcomes." However, "previous research has demonstrated that negotiators’ past experiences—such as reaching an impasse in a preceding negotiation—powerfully influence the intentions with which they approach a subsequent negotiation as well as the outcomes they obtain." In light of this research, the authors hypothesized that losing a BATNA constitutes a "prior negotiation experience," and that this loss would indeed impact how people behave once the negotiation begins.

The Studies

To understand the impact of BATNA loss, the authors conducted several experiments with a total of 2,538 participants. In the first experiment, 90 individuals were asked to imagine that they were selling a VW Golf they had owned for a few years. Participants were told that the market value of the car was £10,100 and that they were about to speak with a potential buyer named Taylor. Prior to meeting with Taylor, however, a subset of the participants was informed that another buyer, Brad, had made a non-negotiable offer of £12,706 for the Golf. But just prior to their meeting with Taylor, this group was informed that Brad had walked away from the deal and would not be buying the car.

The authors predicted that participants who lost Brad's offer would set a higher target price for the Golf, and that is exactly what happened. Even though Brad's option no longer existed, it influenced the actors to try to at least match Brad's price during negotiations with Taylor.

In their second study, the authors asked 220 participants to sell a limited-edition Starbucks coffee mug valued at £10 on eBay. Prior to meeting potential buyers, they were informed that an offer of £14.50 had been made for the mug. However, five minutes later, that buyer canceled the offer, and once again the BATNA was removed. As with the first study, the second experiment demonstrated the retracted offer's clear effect on sellers, who set higher opening prices and achieved overall higher final sale prices than those who never heard the £14.50 number. As shown in Figure 1 below, even a lost BATNA can lead to higher price aspiration and a better final outcome than not having had the BATNA in the first place.

Figure1: Study 2a Negotiated Outcome for Sellers.

Studies 1 and 2 demonstrate the power of lost alternatives in a negotiation. As the authors note, "although negotiators who lost their only alternative are objectively in the exact same bargaining position as those who never had an alternative in the first place, they obtain better outcomes." How far, the authors then wondered, does this advantage go? In other words, "will a lost attractive alternative, for example, be as beneficial as an attractive alternative that is still available?"

Study 3 was designed to answer this question, and for their third experiment, the authors recruited 495 participants who were asked to manage the sale of a pharmaceutical manufacturing plant. 

Participants were told that their company had purchased the plant three years ago for a “low” price of $15 million but that real estate values around the plant had declined 5% since the purchase. They were also told that failure to find a buyer would result in the dismantling of the plant and the auctioning of assets worth an estimated $17 million. In addition, participants were told that a company named Inergy had already made a $24 million offer for the plant. Unlike the other experiments, some participants were told that the Inergy bid was canceled prior to negotiation start but another group was allowed to keep this BATNA alive during the sales process. The experiment proceeded as follows:

In response to their opening offer, participants received a counter-offer from the ostensible counterpart. The first counter-offer the script gave was $16.50 million, followed by $17.50 million, $19 million, $19 million, $20.10 million, and $21.25 million. The final offer in round 7 was a take it or leave it offer worth $24.5 million. After each pre-set offer was received by participants, they had the option to either accept or reject it (and then make a counter-offer). If the participant rejected an offer (e.g., $17.50 M) and then provided an offer that had a value below the pre-set value for the next round (e.g., $18 M, which is lower than $19 M), the participant was informed that their counterpart had accepted the offer and the negotiation had ended. 

The results of the third experiment were consistent with the first two studies: "negotiators who lost an attractive alternative had significantly higher aspiration levels than negotiators who never had an alternative." Further, "negotiators who had an attractive alternative [i.e., those who did not lose the Inergy bid] had significantly higher aspiration levels than negotiators who lost an attractive alternative." Again, note the authors, "negotiators who had lost an attractive alternative set higher aspirations, made more aggressive first offers, and obtained better outcomes than those who never had an alternative." Clearly, losing the attractive alternative was better than never having had it in the first place. But, as shown in Figure 2 below, "keeping the attractive alternative is better yet." Indeed, "negotiators who still had an attractive alternative set the highest aspirations, made the most aggressive first offers, and obtained the best outcomes."

Figure 2: Study 3 Negotiated Outcome for Sellers. (Source: Authors)

One more study is worth noting. In their sixth experiment, the authors recruited another two groups and repeated the pharmaceutical plant sale experiment. This time, the BATNA group was told about the $24 million Inergy bid, which was then removed during the sales negotiations. Once the negotiations were over, the participants were asked to rate their overall satisfaction with how the negotiations turned out.

As expected the participants who lost the BATNA performed better overall in the sales negotiations than those who never knew about the Inergy bid. However, as shown in Figure 3 below, even though they had better outcomes, the BATNA group was less satisfied with the result of the sale. The results of this final experiment, the authors note, "demonstrate that the better objective outcomes that are typically obtained by negotiators who lost an attractive alternative do not necessarily lead to higher levels of satisfaction." Instead, these findings "are in line with previous research demonstrating that those who do better in negotiations because they set higher aspirations often feel worse about their superior outcomes." 

Figure 3: Study 6 Negotiated Outcome and Satisfaction Interaction for Sellers. (Source: Authors)

This last finding adds more nuance to the paper's overall story. Participants who lose an attractive alternative may do better at the bargaining table, but doing so is likely to leave them less satisfied with their results. Given that "satisfaction with a negotiated outcome is an important precursor of a negotiators’ engagement and success in future negotiations, the benefits of losing an alternative do come with a cost." In other words, because our satisfaction with past negotiations influences how we act in the future, the benefits a lost BATNA may confer could be diminished by sub-optimal actions in a future negotiation.

Conclusions

With their general hypothesis largely confirmed, the authors take some time to discuss why it is that lost BATNAs have such a strong impact on future negotiations. The answer has to do with the oft-cited concept of anchoring. Lost alternatives "serve as powerful anchors”, note the authors, "impacting negotiator aspirations, first offers, and ultimately the negotiated agreement." Indeed, the anchoring literature suggests that "once an alternative has been considered, it will still have some effect on the negotiation, even if it is ultimately lost." This is the case because "anchoring is a remarkably robust phenomenon that can persist even if the anchor is uninformative or irrelevant for the judgment at hand."

How exactly does anchoring work in the case of a lost BATNA? It works because a negotiator who has seen a good BATNA will seek “anchor-consistent” information that sheds a positive light on his bargaining position. In other words, once the BATNA is seen, new information is processed in such a way that reinforces the "correctness" of that now-lost offer. Indeed, the authors found in another of their experiments that forcing people to confront evidence showing the BATNA was incorrect can be an effective way to "de-anchor" during a negotiation. Barring that tactic, even alternatives that are lost and can no longer be attained powerfully influence how negotiators think about, prepare for, and ultimately act in negotiation settings.

This research has some practical implications for business managers. First, negotiators need to exercise care when generating alternatives prior to the start of a negotiation, given the impact that even lost options can have on negotiating strategies. Moreover, should anchoring begin to lead to unfavorable outcomes, negotiators should consider de-anchoring techniques as a way of counteracting possible negative effects. 

Looking at the big picture, however, this paper comes to a positive conclusion. While losing a great job offer at the start of a job search is a shame, it may be of some comfort to know that just having seen it will probably lead to a better final outcome.

The Research

Garrett L. Brady, M. Ena Inesi and Thomas Mussweiler. The power of lost alternatives in negotiations. Organizational Behavior and Human Decision Processes 162 (2021) 59–80 https://doi.org/10.1016/j.obhdp.2020.10.010

Please take a second to rate this post.

• 🟢 This post contained useful insights.

• 🟡 This post was just ok.

• 🔴 This post was not interesting to me.

In Western management theory, there is almost unanimous consensus that price wars are best avoided because they can have devastating (often unintended) consequences for participants. In China, of course, price wars are an accepted part of business life. In fact, this year has seen Chinese price wars in various sectors, from e-commerce, to financial services, to automotive, to commodities. Vanguard's case is especially ironic: a pioneer in low-cost investing in the U.S., it had to pull out of China earlier this year, and its remaining robo-consulting joint venture with Ant Group finds itself under extreme price pressure from local rivals such as Huatai Securities.

Reading the recent price war headlines brought to mind a paper from Z. John Zhang (Wharton) and Dongsheng Zhou (CEIBS). Though the paper is over a decade old, its analysis of price wars is still worth reading to understand how they work and why they persist.

Anatomy of Two Price Wars

The authors start their analysis by noting that in China price wars are a legitimate competitive tactic. The authors hypothesize that managers there have a cultural affinity for thinking of business in military terms, which leads them to see price wars differently from their Western counterparts. As they noted back in 2007:

It is not uncommon for today's executives to talk about the "business arena" as the "battleground," and they do not just talk about it metaphorically, either. In fact, strategy in Chinese, "zhanlue," literally means "battle plans" or "combat strategies." The rub is that while Western companies seem to suffer whenever they start, or they are caught in, a price war, Chinese companies seem to thrive on price wars they start and many emerge from them stronger, bigger, and more profitable. 

The authors argue that Chinese managers’ view of price wars leads them to a different understanding of their dynamics and a better set of tactics to deploy when a price war breaks out. To support their view, the authors analyze two case studies from the early days of China's ascension to the world economic stage in the 1990s.

Changhong

The first case presented focuses on Changhong, a Chinese producer of consumer electronics. The company found its TV business in an untenable position in 1995. Foreign firms from Japan and elsewhere dominated the premium end of the market, while a growing number of local firms fought for control of the middle and lower segments. With tariffs set to drop significantly in 1996, analysts predicted that in two years' time an additional 10 million units of (foreign-owned) annual capacity would enter the Chinese market, which could be a fatal blow to the company’s TV profits. 

Changdong, however, had many advantages on its side. It had 17 production lines in one location, making it the biggest, the most efficient, and the most profitable TV producer in China. Changhong was also the largest manufacturer of many key components, such as plastic injections, electronic components, remote controls, etc. Moreover, as a highly vertically integrated company located in Sichuan, one of the less developed regions in China at the time, Changhong “enjoyed cost advantages and earned the highest profit margin among all domestic color TV manufacturers.” Moreover, Changhong was the first color TV manufacturer listed in China's stock market, with a high level of brand awareness and a high-quality image (among domestic brands). 

After considering the competitive and regulatory landscapes, Changhong's CEO, Ni Runfeng, decided that his best competitive option was a price war. He believed that a 10% reduction in prices would be enough to squeeze his local competitors, who did not have his margins and who could no longer count on government subsidies to keep them afloat. At the other end of the market, the foreign premium brands, who enjoyed a 20% price advantage, would face two choices. They could stay out of the price war and lose market share to Changhong, or they could cut prices, thereby lowering their margins while possibly damaging the premium position of their brands. 

On March 26, 1996, Changhong fired the first shot, announcing a price reduction of 8% - 18% for all its 17" –29" color TVs, and from there the price war evolved “mostly as Changhong had expected.” Some domestic competitors appealed to regulators for help, with no success. Other domestic producers cut their prices; however, they lacked Changhong's strong supplier network and could not compete across all product lines. Foreign brands such as Sony and Panasonic, refused to lower prices, as Changhong had expected, stressing quality and functionality to buyers. 

Within a few months, the winner and losers in the price war began to emerge. Changhong's overall market share increased from 16.68% to 31.64%. Some medium-sized local players, TCL and Xiahua in particular, who followed Changhong quickly, also benefited, but most small domestic players endured heavy losses. As the authors note:

During January – March 1996, there were a total of 59 local brands that had sales in the one hundred largest department stores in China. By April, this number dropped to 42. In the process, the market share for these small players dropped by 15.19%. Those big domestic manufacturers who did not follow suit saw their market shares dwindling, too. Panda's market share dropped from 7.6% to 5.8% and SVA from 5.5% to 2.6%. 

Foreign brands also suffered. Before the price war, imports and joint venture products accounted for 64% of the Chinese TV market. After the price war, "the market share of domestic products significantly increased with a total of around 60% by the end of 1996." By 1997, "8 out of the top 10 best selling brands in China were Chinese and three local players, Changhong, Konka, and TCL became the best selling color TV brands in China." The effect of the first-ever Chinese price war was clear to everyone: it "drastically changed the landscape in the industry in favor of Chinese companies and the CEO of Changhong, Ni Runfeng, became a hero for Chinese national industries."

Galanz

In mid-1996, it was possible to think of Changhong's CEO as someone who just "got lucky" with an audacious but very risky plan. However, the microwave oven producer Galanz showed once again the power of a price war executed correctly. From August 1996 till October 2000, "Galanz initiated five major price wars and, through them, became the world's largest microwave oven manufacturer, with about 30% of the worldwide market and 76% of the Chinese market."  

Galanz's decision to initiate the second price war in August 1996 was not without its own risks. However, the company's CEO gave the green light for three main reasons:

1. A significant portion of Chinese households was ready to modernize their kitchen, especially with the latest appliances.

2. As one of the largest manufacturers in China, Galanz saw a need to reorganize the industry for future growth.

3. A well-planned and executed price war could help Galanz create cost advantages that would be difficult for competitors to copy.

In sum, Galanz believed that a well-managed price war would increase its sales and eliminate weak competitors that were holding the industry back. 

Galanz launched its price war in August 1996 with a price reduction of as much as 40% on some of its key products. In some cases, note the authors, Galanz's price reductions were higher than their gross profit margins. Cleverly, Galanz picked August to start the price war, as it was "the off-peak selling season when manufacturers would generally downscale their production and distribution." Galanz's actions made headlines in China and were welcome by retailers, some of whom even supported Galanz by lowering their own margins to drive up sales of Galanz’s products. 

Galanz’s competitors were stunned. The small local players failed to respond initially and its largest foreign competitor, the joint venture started by U.S. company Whirlpool, was also slow to react. Consequently, by the end of 1996, Galanz's market share had grown from 25% to almost 35%, and its profits grew from cost reduction measures taken at the same time. The results of its plan were so positive that the company started four more price wars from 1997 to 2000, each time significantly cutting prices and production costs while dramatically increasing sales—a positive trend that continued, as illustrated by Figure 1 below.

Figure 1: Galanz’s sales Information for the years 1995 – 2003 (Source: Authors)

Throughout the process, Galanz used a specific approach for price setting that proved devastating to its competition:

It set its price at the break-even level for its nearest competitor. For example, if the second player's annual sales were 2 million units, then Galanz would set its price at the break-even level for the 2-million units. During price wars, Galanz's price would even go significantly lower than this breakeven point. Using this strategy, Galanz always made rivals reluctant to cut prices and thus it always stayed ahead of competition in capturing more volumes. As the process unfolded, Galanz encountered fewer and fewer competitors. In 1996, there were about 120 microwave oven manufacturers. By 2003, the three largest microwave oven manufacturers took over 90% of the market. 

The Art of the Price War

The success of Changdong and Galanz set the blueprint for how to win a price war in China, and the authors describe it in their paper. The framework is based on the use of Incremental Breakeven Analysis (IBEA), which is based on the idea that a firm can only benefit from starting a price war if its sales go up sufficiently to cover lost profits.

IBEA identifies the sales change that will allow profits (after the price change) to stay the same as before, i.e., the breakeven sales change. The formula is presented in Figure 2 below.

Figure 2: Incremental Breakeven Analysis (Source: Authors)

Returning to the Galanz case, the company planned to reduce its average product price by about 20% (thus, Δp=20%). Galanz's average contribution margin — the contribution per unit sales before the price change (price minus marginal cost) as the percentage of the pre-change price, was about 40% (thus, cm = 40%). The expected higher sales volume should reduce unit cost by 30-40% (thus, Δc= 35%). If one enters these values into the formula, Δq = 0.905 or 90.5%. In other words, "if the demand for Galanz's products would increase by more than 90.5% as a result of the 20% price cut, Galanz would make more profit by implementing the price cut." The only question left to answer for Galanz's CEO was if the price cuts would increase sales by 100%? He bet on the positive outcome, and Galanz easily beat that number. Thus, the authors conclude, the price war "was the rational thing to do."

The IBEA formula can be used for other analyses, the authors note, which help identify when price wars are more likely to occur. For example, across different industries, "the ones that have (unusually) high margins tend to be the ones where price wars break out, all else being equal." Additionally, in most sectors, the firm with the best margin is most incentivized to start a price war. These two insights, note the authors, explain why Chinese companies tend to start price wars when they enter the markets in the West. Chinese companies "have cost advantages and a favorable exchange rate, and they encounter a small number of competing firms in every market they enter." To them, "every business in the West is a high margin business!"

One other point is worth noting. In the IBEA formula, a larger Δp will lead to a larger Δq. This relationship simply suggests that a large price cut needs to generate a larger volume increase to break even. However, the authors explain that this relationship also tells us something about how price wars are related to product differentiation:

In a highly differentiated industry, it would take a huge price cut to persuade customers to switch from one firm to another. This, in turn, means that a huge increase in sales has to be expected in order to justify the price cut in the first place. Therefore, in a differentiated industry, price wars are less likely to break out. It is almost unnecessary to repeat the cliché here, except the fact that it conforms with the Chinese experience with price wars very well. Price wars almost always break out in an industry in China when products in the industry become standardized, with little room for further technology innovations and quality improvements.

Conclusions

"There is nothing intrinsically Chinese," note the authors, "about the calculus that Chinese executives use in planning and executing price wars." What is intrinsically Chinese, however, "is the fact that a whole generation of Chinese executives has grown up in a business environment characterized by growing markets, heterogeneous firms with a wide distribution of cost-efficiencies, and new technologies with significant scale economies." This business environment provides "many profitable opportunities for them to engage in price wars and to hone their skills.” By contrast, in Western markets, “oligopolistic competition among (mostly) equals in mature markets encourages more finesse in devising marketing strategies." In both cases, of course, managers make the rational choice.

Returning to the cases I noted at the start of the post, we see many of the same dynamics the authors detailed a decade ago still in effect. Especially in the financial services example, history repeats itself: a wealthy foreign entrant helps create a market for a new kind of product. It is attacked by low-cost local firms that push up quality, diminish product differentiation, and squeeze margins to drive out the weakest competitors. In time, the foreign innovator abandons China, leaving the local firms to fight until such time as a clear winner emerges. Because the past is prologue, the authors suggest that Western managers should understand the dynamics of price wars (as explained by IBEA) and focus on making price wars a losing proposition for Chinese competitors. Moreover, they should ignore the stigma that Western management thinkers place on price wars. In the end, a price war, in the right circumstances—and especially against a Chinese competitor—is a strategy that must be understood and, in many cases, clearly expected. 

The Research

Z. John Zhang and Dongsheng Zhou. The Art of Price War: a Perspective from China. Wharton Working Paper. 2010. Available here.

One of the competitive challenges faced by global high-tech manufacturers is maintaining extremely high levels of product quality while continuously bringing new technologies to market. Consequently, manufacturers typically invest heavily in a variety of quality control and assurance technologies to minimize defects in production environments. Researchers have focused extensively on this issue, noting, for example, the impact that quality management systems and techniques have on product performance. 

One of the issues that has not received as much attention from researchers is the impact that human capital dynamics have on product quality. Unlike service industries, where human capital is widely acknowledged to be a principal driver of quality and customer satisfaction, manufacturing workers have not generally been considered an important determinant of high-tech product quality for two main reasons. First of all, the products themselves are sophisticated enough that their ultimate quality is seen as less dependent on the human capital used in assembly. Second, the high degree of standardization and automation in high-tech manufacturing operations suggests the human factor is less critical in overall production line quality outcomes. Consequently, when looking at human capital in production, researchers have tended to focus on economic issues such as plant efficiency and productivity.

In a break from the past research trends noted above, a recent paper from Ken Moon (Wharton), Prashant Loyalka (Stanford), Patrick Bergemann (UCI Merage), and Joshua Cohen (Apple University) provides an interesting analysis of one way in which human workers affect product performance. Specifically, their research examines the impact that turnover in a manufacturing workforce has on product quality. 

The authors start by noting that high-tech manufacturers typically deploy two main strategies to control quality. As shown in Figure 1(b) below, the first strategy is that "received components undergo initial testing and inspection for defects, prior to being used in product assembly." Second, "the partially assembled devices and key components undergo controlled “burn-in” usage under stressful conditions, such as elevated temperatures and voltages, in order to force defect-based early failures to manifest." As shown in Figure 1(a) below, product failures tend to occur early (due to defects) or late (due to premature wear) in the testing process, giving rise to what is known as the "bathtub curve."

Figure 1: Conceptual illustration of failure hazard rates over the lifetime of a product unit (Source: Authors)

The bathtub curve, however, does not tell the whole story, because even stringent tests cannot always detect minute defects. Consequently, the authors note, even small errors in assembly could increase field failures. For example: 

The touchscreen multi-touch functionality in some mobile devices depends on the soldering connecting touch-related chips to a circuit interface. The resilience of the soldering is stressed under repeated customer usage, e.g., bending or dropping of the device. Similarly, a small imperfection in battery placement can lead to uneven power supply that degrades the screen display over extended usage. 

As a result of these undetected minute defects, "even world-class assembly line design and quality control may allow some variability in the quality of manufactured units that reveals itself only after repeated use, as minute imperfections in assembly degrade a product’s performance or accelerate its wear-out." Given the impact that assembly activities can have on product performance, the authors set out to understand to what degree worker turnover is related to the kinds of errors noted above and thus to overall product quality.

The Study

For their research, the authors collaborated with a major consumer electronics producer and one of its Chinese contract manufacturers. The manufacturer provided data tracking field failures for nearly 50 million devices, as well as weekly assembly line statistics and component quality data. Specifically, the authors looked at data such as "quarterly fallout rates for six key components: the device’s battery, display, housing, circuit board, rear camera, and front camera." Additionally, they obtained "the weekly pass rates of devices at the burn-in stations of the individual assembly lines" as well as human resources staffing data to track factory worker quits occurring on specific assembly lines and weeks.

For their quality analysis, the authors examined the percentage of delivered devices that "failed in the field," i.e., devices that needed any kind of service repair or replacement. For devices produced from September 2014 through June 2015, the authors analyzed all field failures as of July 6, 2019. Importantly, "because each device has been observed in the field for at least four years post-production, such field failure rates should capture virtually all variation in product reliability that originates in the factory assembly process."

The contract manufacturer provided the human resources data required to track production line assignments and worker turnover patterns. The authors defined worker turnover as "the fraction of workers on an assembly line who quit the factory during a given week."1 The contract manufacturer’s data covered 52,214 workers during the study period, and mean and median worker turnover were 5.1% and 3.2% per week. The mean assembly line headcount was 301, and staffing was relatively steady throughout the study period.

All told, the study data covers 40 weeks of production for 44 assembly lines in China, and there were no unusual events relating to production, suppliers, or labor during the study's time period.

The Findings

As shown in Figure 2 below, the authors found that the manufacturer's pay cycle has a major impact on when a worker quits. In fact, assembly line turnover averages 2.9% the week before a payweek and rises to 8.9% the week after a payweek. As Figure 2 also illustrates, there is a direct relationship between turnover levels and field failure rates.

Figure 2: Field failure rates and worker turnover rates over the manufacturer’s pay cycle (Source: Authors)

The payweek-turnover cycle has important implications. First, worker turnover "causes a net loss of experience in staffing the affected assembly lines, as relatively experienced workers exit and new workers enter." As shown by Figure 3 below, "the average level of experience is not higher among quits in post-payday weeks, but the total weekly outflow of experience is substantially greater (382 versus 171 years of experience on average)." Second, the replacement process itself "may be disruptive, requiring attention and effort from both supervisors and workers." For example, because only experienced workers can handle certain workstations, some workers must switch stations in response to turnover even as new workers are being integrated into the production line. Both factors could be expected to impact negatively the assembly line overall output quality. 

Figure 3: Experience-based composition of worker exit in post-payday weeks (Source: Authors)

The authors hypothesized that if monthly post-payday turnover leads to compromised workmanship, the facility’s weekly field failure rates "should manifest a pattern resembling an electrocardiogram," i.e, the field failure rate moved in regular, monthly cycles of peaks and troughs. Figure 4 below shows the expected result for the three assembly buildings at the contract manufacturer.

Figure 4: Rates of device field failures attributed to their weeks of production during Sept. 2014 to June 2015 (Source: Authors)

The authors note that "not only does each building and assembly line suggest a cyclically peaking pattern, the assembly lines’ field failure rates co-vary to a striking degree: nearly 41% of the variation is explained by the weekly averages." 

For post-payday weeks, each 1 percentage point increase in worker turnover rate is estimated to increase field failures by 0.74-0.79%. Thus, at the average rate of turnover for post-payday weeks, "worker turnover makes field failures 7.1-7.6% more common." Moreover, the authors found that workload also had an impact on quality: "a 50% increase in workload results in a 0.4-0.9% increase in the field failure rate, which when combined with the 50% expanded output amounts to a 0.55-1.34% increase in the weekly number of eventual field failures."

For non-payday weeks, worker turnover's effect on field failures is also significant: "field failures become 0.21-0.44% more common per each 1 percentage point increase in turnover." At the 5.1% average rate, "worker turnover raises field failure incidence by 1.1-2.3%, and similarly by 1.6-3.3% at the 75th percentile rate of 7.5% worker turnover."

In all, the authors found that worker turnover is detrimental to quality throughout the production calendar. As they conclude:

We find a surprisingly close relationship between product field failure rates and episodes of factory worker quits, as they rise and fall together. We find this relationship to hold consistently both (1) when some assembly lines experience greater turnover than others within the same week and (2) most strikingly when turnover periodically elevates facility-wide following paydays. 

To the surprise of the manufacturer participating in the study, the effects noted by the authors are significant enough that they suggest it should take concrete steps to curtail post-payweek resignations, especially among more experienced staff.

Conclusions

The authors believe that there are several practical issues for manufacturers to consider as a result of their research. First of all, manufacturers may want to reconsider the conventional wisdom that the high degrees of process standardization and product sophistication in high-tech products mean that worker turnover does not materially impact product quality. Second, manufacturers could benefit from raising compensation, increasing benefits, improving working conditions, etc., to reduce worker turnover. Indeed, the authors believe that the direct relationship between high turnover and increased field failures (which in turn drive higher warranty and service costs), makes a compelling business case for increased wages. Moreover, the authors note, that even in settings where skills are less important, "firms may benefit from incorporating additional training or skills progression into their assembly designs in order to better retain workers, and scheduling best practices can also aid worker retention." 

Another implication of this study is related to failure analysis for design and engineering teams, as this study's results suggest assembly worker turnover should be included in product failure analysis and warranty cost estimation. In fact, the paper makes a strong argument for directly tracing field quality issues back to workforce characteristics. As the authors note, "whereas traceability verifies the sourcing of goods to address issues such as adulteration or quality-based recalls, we use traceability to link data tracking workforce dynamics and working conditions to connected product data." Indeed, because connected products today can be tracked and traced directly by the manufacturers, firms have more opportunities than ever to connect product field performance back to in-factory workforce dynamics.

By taking a new view of the impact of worker stability on product quality, this paper provides manufacturing, quality, after-market, and quality leaders much to consider. As one of the authors noted in an interview:

A lot of turnover introduces uncertainty…. You don’t know who is going to leave when, and how that might trigger cascade effects on a production line when workers start to leave in part because their friends have left the firm. To deal with this uncertainty, you over-hire. That’s part of the analysis that we go through in the paper, and that has a very real cost.

In other words, when you start to control turnover, there are two main [results]. One is, “I raise compensation, and I keep people around longer [reducing disruption].” Your firm is no longer as much of a rotating door or turnstile. The second is, “I have less uncertainty, so I hedge less. I don’t need an excess workforce sitting around or to prepare for [production shortfalls caused by] turnover. I’m no longer putting out fires.” It becomes a much more stable and productive environment for the firm and for the workers. It looks to us like it could be a win/win.

From my experience with high-tech supply chain operations, I know that many other issues can draw managers’ attention from human capital factors when considering product quality at production sites or in after-sales and warranty modeling. This paper makes a strong case for re-assessing those priorities. Managers should look at human capital dynamics as a key variable not just of production economics but also the quality and performance of even the most sophisticated products.

The Research

Moon, Ken and Loyalka, Prashant and Bergemann, Patrick and Cohen, Joshua, The Hidden Cost of Worker Turnover: Attributing Product Reliability to the Turnover of Factory Workers (August 3, 2021). Available at SSRN: https://ssrn.com/abstract=3568792 or http://dx.doi.org/10.2139/ssrn.3568792

One of the trends making headlines these days is the “Great Resignation,” as workers leave their jobs in unprecedented numbers. As a recent HBR piece noted, the pandemic has radically shifted the way in which people see work, and many have simply decided to leave—or not return to—their pre-pandemic jobs. Reading about the shift in the way office work is being seen in 2021 reminded me of Scott Beauchamp's 2018 analysis of workspace history. Beauchamp's piece was an attempt to connect historical developments in workspace design with changing conceptions of capitalism, and it inspired me to reflect on how our offices mirror how we see work and the capitalist system in which most of us function. As the author noted:

The currently fashionable open-office plan—a design which attempts to incorporate the creative fluidity of a tech start-up with the stability of the “traditional” office—is only the most recent example of conflicting motivations inhabiting the workplace (or workspace, as they’re now called). The layout of such an office isn’t new, nor are the general conceptual arrangements associated with it. But the context has changed. Technology is shifting. New management styles have developed. The notion of the company itself is already different from what it was even a generation ago. To understand where we are and how we got here requires rummaging through office spaces of the past and reading them like hieroglyphics. It requires an archeology of the present.

Beauchamp begins his analysis by citing one of the earliest descriptions of office space, found in Herman Melville's “Bartleby, the Scrivener: A Story of Wall Street.” The office in the story, which is set around 1853, was a dark and dingy place, whose main purpose was to “separate the clerks from the laborers.” Of course, at this time most people were still farmers, and the idea of the industrial worker was just being born. In Melville's story, the firm's manager wants Bartleby (a writer of hand-written documents and records in an era before typewriters were invented) close but not actually visible:

I resolved to assign Bartleby a corner by the folding-doors, but on my side of them, so as to have this quiet man within easy call, in case any trifling thing was to be done. I placed his desk close up to a small side-window in that part of the room, a window which originally had afforded a lateral view of certain grimy back-yards and bricks, but which, owing to subsequent erections, commanded at present no view at all, though it gave some light. Within three feet of the panes was a wall, and the light came down from far above, between two lofty buildings, as from a very small opening in a dome. Still further to a satisfactory arrangement, I procured a high green folding screen, which might entirely isolate Bartleby from my sight, though not remove him from my voice. 

In this mid-nineteenth century form, the office was a place of drudgery, one step removed from a factory floor, in which the first generation of professional industrial managers kept workers under close inspection to ensure that work was completed and that workers, whom they probably did not trust, delivered the labor for which they were paid. Capitalism, at this stage, was crude and coercive: it was a simple exchange of toil for wages, and no one expected more from an office than a place in which to work under strict supervision. 

At the start of the 20th century, some sixty years after Bartleby's world, we find what Beauchamp considers the first corporate headquarters in our current understanding: the Larkin Building built for the Larkin Soap Company in Buffalo, NY. As Beauchamp noted, this new workspace reflected a maturing model of capitalism, replete with clerks and specialists, all having their space allotted in accordance with their contribution to the enterprise:

The unavoidable intimacy of Bartleby’s office had been replaced by a bland anonymity, an atomization that was only amplified by banal corporate team-building activities. There’s a YWCA on-site, as well as the opportunity to write for the employee-run newspaper. There are places in the building for employees to relax and rejuvenate. And the building itself, with its open central atrium rich with natural light, is as conducive to worker morale as it is useful for employers to observe and measure their employees’ every move. Every major aspect of the burgeoning spirit of twentieth-century capitalism coalesced in the Larkin Building, including the fact that it was more than a building. Designed by Frank Lloyd Wright, it was also a work of art, and an integral part of Larkin’s brand.

For Beauchamp, Larkin's capitalistic ideal was a paternalistic one, filled with managers who “knew best” and who created a world in which all employee needs were met within the workspace, and through which all employee actions could be monitored and corrected as needed. Larkin was not alone in this view, of course, and this was the great age of the “company town,” which gave rise to the idea of a “job for life,” a concept whose demise at the end of the 20th century continues to reverberate through our socio-political discourse.

By the 1960s, of course, the paternalistic capitalism of the Larkin era had been replaced by a modern version, shaped by the aesthetic and social aspirations of the space age, the rise of global brands, and a need to have offices reflect social, as well as capitalist, changes. As Beauchamp notes, the 1960s workspace was a reflection of Western society’s sleek new self-conceptualization:

Machine-crafted objects, sleek with minimal lines, embodied an ethos of modularity. Drop ceilings, developed to hide the raw innards of wires and beams, were industrial Mondrian. Cheap fluorescent lights flickered over cubicles deep within the windowless center of the building. And the file cabinet was itself a kind of office building in miniature. “Each office within the skyscraper is a segment of the enormous file, a part of the symbolic factory that produces the billions of slips of paper that gear modern society into its daily shape,” wrote C. Wright Mills. The modularity and order imposed on the office by the grand architectural minds of Europe perfectly suited the mid-century zeitgeist, and the new orthodoxy of office architecture theory proliferated in only slightly modified form in a million American office parks.

This version of capitalism, of course, was modern and optimistic, obsessed with trans-national conceptions of coolness, and workers were given much more freedom to shape their spaces. Conference rooms proliferated, lounges and flexible spaces entered the workspace, and the idea that workspace should shape, much more than reflect, corporate culture began to take hold. In the century’s last two decades, the first computers entered the workspace, initially in government and academia and then in the private sector, and their arrival challenged workspace designers to integrate people and machines side by side. Thus, an automated form of capitalism was born, and workers began to think of their work as tied to, and enhanced by, machines. Capitalism itself would from now on require not just human labor and intellect but “smart” machines that made us better and more powerful. 

With the social and media revolutions of the late 20th century, it should come as no surprise that the next reinvention of the workspace would be made by an advertising mogul and would involve the demolition of what was left of the walls that separated most workers. Legend has it that one day Jay Chiat, then-CEO of the advertising agency Chiat\Day, “had a vision on the ski slopes of Telluride.” On the basis of his vision, Chiat hired designer Gaetano Pesce to build a new office space in 1994, and, notes Beauchamp:

The result was something that resembled a cross between a warehouse, a conference room, and a college rec center. Without a need for walls, desks, offices, or file cabinets, employees seemed to “move through an improved dimension in a radically fluid arrangement of space” designed to keep the firm in a “state of creative unrest,” the New York Times reported.

Thus, the floor-plan office was born, and two decades later it was—pre-COVID at least—the dominant workspace concept across most of the Western world. Ironically, just as the pandemic emptied office spaces, more and more research pointed to its negative effects. A study from Steelcase, for example, highlighted the flaws inherent in most open floor-plan designs. As the report noted:

A constant din of sound serves as the backdrop to conditions in which workers are observed more intrusively than even the Taylorists could have imagined. And yet, contrary to what the Taylorists might have predicted, these panopticon offices are actually counterproductive in the literal, economic sense of the term as well. A growing body of research describes their deleterious effects on workers’ efficiency, with one study estimating that open offices cause a 15 percent reduction in productivity.

The Steelcase report added that:

More than ever before, workers are going public with complaints about their lack of privacy at work. Blogs and online chat rooms are chock-full of soliloquies about what everyday life in an open-plan workplace is like: how easy it is to be distracted, how stressful the environment can be and how hard it is to get any individual work done.

The open office, said its critics, had become a cacophony of voices and sounds. Workers were inundated with noise, so much so, said the Steelcase report, that many respondents “say they literally can’t hear themselves think.”

Perhaps in response to these complaints, the latest offices from the likes of Google, Facebook, etc., have returned to a modern version of the Larkin ethos. Daycare, recreation spaces, parks, etc. are an attempt to de-stress the open office, much as Larkin's then-revolutionary laundry and atriums was an attempt to de-stress the corporate building. These new conceptualizations of workspace, shiny and new as they are, have as much to say about our version of capitalism, notes Beauchamp, as the workspaces of Bartleby and the Mad Men eras.

For Beauchamp, the implications of pre-pandemic office design are at once subtle and disturbing. To explain why this is so, he cites the South-Korean philosopher Byung-Chul Han, who believes that work has changed from a dialectic around what I, as a worker, am allowed to do by my company, to what I, as a worker, am capable of doing through my own efforts. In Han's view, companies replaced a workspace shaped by a “disciplinary” society, in which bosses set rules and workers were measured by standards of conformity to those rules, with an “achievement” society, in which workers are measured by their self-driven productivity. In other words, for Han, the victory of the open office space is one that signals a total internalization of capitalism, in which employees no longer need to be boxed, segmented, and monitored in the office to perform. Freed from their physical spaces and constraints, employees are left only with themselves as motivators and judges, pushed by their fellow workers/competitors — and increasingly AI — to meld personal and work lives into one unified, always-on/always-working, existence. As the author concludes:

An attention economy dissolves the separation between the personal and professional, between entertainment and information, all overridden by a compulsory functionality that is inherently and inescapably 24/7. What this suggests is that as the office walls come down, so will the temporal and ideological barriers separating work from nonwork. The office of the future, in other words, won’t be a place, but an identity. The office of the future will be your most intimate conceptions of self, somehow put to work.

The pandemic, of course, made Beauchamp’s conception into a global reality. The office of 2021 is indeed an identity, a state of mind, enabled virtually through technology and completely integrated not just into our minds but into our private lives and homes. We now, as Han predicted, work at home and live at work. The pandemic has brought about the ultimate fulfillment of Chiat's desire that workers live in a constant state of creative unrest.

This is a gloomy outcome that has led to feelings of exhaustion and isolation all over the working world, but there is hope on the horizon. Research indicates that people feel a need to reclaim a clear demarcation between work and home as the pandemic ends. This dissociation of two worlds that the pandemic fused may, in turn, force us to reconsider the version of capitalism that dominated before the COVID crisis. Indeed, a recent report from Steelcase suggests that the most important change the pandemic brought to workers is an increased desire for autonomy.

Most people want a hybrid model to remain and they want to retain the autonomy over work-space and work-time that the pandemic gave them. Moreover, as AI technologies increasingly threaten white-collar work, some workers are starting to wonder about life outside the full-time capitalist system, moving to gig work that allows them to shape their own lives independent of a single company or employer.

In Beauchamp’s dystopian present/future, capitalism was, in 2018, making the leap from a system to a state of mind, from a conceptual commercial framework to a holistic way of life. In our future, then, managers would be obsolete, as workers merged their private and work lives completely, immersed in a capitalist system that was with us from sunup to sundown. Ironically, in some strange way, the pandemic may have stopped that dark vision from becoming a reality.

Workers have had over a year to reflect on where capitalism was headed pre-pandemic, and many did not like what they saw. As they return to their offices over the next year or so, perhaps they will do so with a different concept of what work means to them. Indeed, in a recent article Han notes that the origin of the word crisis is the Greek term krisis, meaning a turning point. Perhaps, writes Han, the pandemic may “allow us to reverse our fate and turn away from our distress.” If so, this new path will require a radical redesign not just of workspace but of work itself. Hopefully, capitalism is capable of yet another evolution through which we regain our independence, both physical and psychological, from those spaces that can so easily come to reflect, and even define, who we are.

In a recent post I touched upon the idea that as Artificial Intelligence (AI) and Machine Learning (ML) applications become more common, we are likely to discover unintended consequences from their adoption. Despite their power, these technologies remain in their infancy, and their creators are finding that even extensive forethought in design does not insulate these systems from failure. That failure may be severe enough that the AI actually makes a situation worse through its intervention.

A new paper from Shunyuan Zhang (Harvard), Nitin Mehta (Toronto), Param Vir Singh (Carnegie Mellon), and Kannan Srinivasan (Carnegie Mellon) describes just such an outcome with a price optimization algorithm introduced by Airbnb in 2015. The idea behind the roll-out was straightforward: because AI can do a better job than humans at analyzing massive amounts of supply and demand data across the rental platform, it should help landlords set the optimal price for a given property in the system. When the pricing algorithm went live in 2015, hosts were given the option of letting the pricing algorithm set rental prices automatically after it had evaluated a series of factors that included property features, seasonality, comparable pricing, etc. Because Airbnb has far more data and superior computational resources than any individual host, its pricing algorithm should be more effective than the average host at finding the revenue-maximizing price.

The authors note that the Airbnb project was not without its challenges. The ML system is not easy to understand for the average host. Moreover, the ML’s logic would have to consider that Airbnb’s interests do not always align with those of its hosts; thus, the algorithm should ideally choose the hosts’ interests over Airbnb’s should they ever conflict. A greater challenge was that—even for similar properties—there is a disparity in the income earned by white and Black hosts across the platform. A 2017 study1 found that:

• Across all 72 predominantly Black New York City neighborhoods, Airbnb hosts are 5 times more likely to be white. In those neighborhoods, the Airbnb host population is 74% white, while the white resident population is only 14%

• White Airbnb hosts in Black neighborhoods earned an estimated $160 million, compared to only $48 million for Black hosts—a 530% disparity

• The loss of housing and neighborhood disruption due to Airbnb is 6 times more likely to affect Black residents, based on their majority presence in Black neighborhoods, as residents in these neighborhoods are 14% white and 80% Black.

The authors accept the reported inequalities but note that race may not be the only factor driving that phenomenon: “It is plausible that differences other than race (e.g., education and access to other resources) make it more difficult for Black hosts to determine optimal prices.” Therefore, a successful ML tool should be able to help Black hosts reach parity with their white counterparts if it were accounting fully for the factors that drive unequal racial revenue outcomes.

The Study

To understand the impact that Airbnb’s algorithm had on host revenue the authors looked at 9,396 randomly selected Airbnb properties across 324 zip codes located in seven large U.S. cities. The data included key property characteristics, host demographics, each host’s monthly revenue, and date of algorithm adoption (if any). Of the 9,396 properties studied, 2,118 hosts adopted the algorithm at some point during the observation period, the distribution of which is shown in Figure 1 below.

Figure 1: Histogram of the timing of algorithm adoption (t = 0: November 2015). y-axis: density of event; x-axis: month (Source: Authors)

Findings

The data presented several key findings. On average, hosts who adopted the algorithm saw a downward price trend of 5.7% that increased overall revenues by 8.6%. In other words, the tool made properties less expensive, which in turn led to more rentals. The authors highlight that:

Before Airbnb introduced the algorithm, Black and white hosts charged similar prices for equivalent properties (in terms of observed host, property, and neighborhood characteristics), but white hosts earned $12.16 more in daily revenue than Black hosts. The revenue gap was caused by a difference in the occupancy rate (rental demand): 20% less for Black hosts’ properties than for equivalent white hosts’ properties. The algorithm benefited Black adopters more than white adopters, decreasing the revenue gap by 71.3%.

Algorithm price reductions were similar for both white and Black hosts, but it was Black hosts who saw their occupancy rates increase the most. For this reason, Black hosts who adopted the tool obtained more value from its adoption than white hosts. This outcome, the authors note, “supports our theory that Black and white hosts face different demand curves; the demand for Black hosts’ properties is more responsive to price changes than the demand for equivalent properties owned by white hosts.” 

The findings, however, illustrate two challenges. First, Black hosts were 41% less likely to adopt the algorithm. When combined with the overall revenue-lifting impact of the algorithm, the lower adoption rate means that the revenue of non-adopting Black hosts decreased relative to the host population average. In other words, since more white hosts took advantage of the algorithm’s power, revenue inequality on Airbnb increased over time. The second problem, the authors note, "is that if Black and white hosts face different demand curves (as our data suggests), then a race-blind algorithm may set prices that are sub-optimal for both Black and white hosts, meaning that the revenue of both groups could be improved by the incorporation of race into the algorithm." 

The authors believe that the findings of their study point to some important recommendations for AI regulators and Airbnb. For example, current U.S. law prohibits the use of "protected attributes" (such as race) in the development of predictive algorithms. While well-intentioned, this regulatory approach may end up hurting those whom it intends to protect. Regulators, the authors argue, "should consider allowing algorithm designers to incorporate either race or socioeconomic characteristics that correlate with race, provided that the algorithm demonstrates an ability to reduce racial disparities." As a recent paper from AI ethicist Alice Xiang noted:

Ironically, some of the most obvious applications of existing law to the algorithmic context would enable the proliferation of biased algorithms while rendering illegal efforts to mitigate bias. The conflation of the presence of protected class variables with the presence of bias in an algorithm or its training data is a key example of this: in fact, removing protected class variables or close proxies does not eliminate bias but precludes most techniques that seek to counteract it.

For Airbnb managers, the hosts argue that "while Airbnb cannot overturn a racial bias that is ingrained in society at large, it could try an intervention that prevents guests from knowing the host’s race until they book the property." Moreover, Airbnb could make a greater effort to encourage the algorithm’s adoption by Black hosts. Otherwise, "a racial disparity in algorithm usage may end up increasing the economic disparity rather than alleviating it."

Conclusions

Reading this paper reminded me of a complex system implementation I managed for a global high-tech client some years ago. As we prepared to go live, I sensed that my client expected that the new system was going to make most of their production problems go away. My expectation was quite different: the new system would significantly add to the list of issues the company’s managers would have to manage. My conclusion was based on two reasons:

• Pre-system, a lot of problems were being solved by human interactions that did not leave any digital evidence of the problem or solution. These problems would now appear in the system records.

• The system itself would ask new questions and demand new processes that would, in turn, create brand new challenges for my client.

Realizing that it was quite possible that in the end, despite the project's success, this particular company would end up with more challenges to solve than it started with, I told them the following (true) story.

My roommate in graduate school was a Flemish engineer who lived in the lovely town of Ghent, Belgium. He had been a devoted audiophile since we met, and his dream was one day to own a Linn stereo system. For those of you unfamiliar with Linn, it is an esoteric manufacturer, based in Scotland, with a legendary reputation. One day my friend called me to tell me that he had finally bought “the Linn” and wanted me to come to Ghent to hear it.

Soon afterward, when I happened to be in Europe, I took a detour to experience this amazingly expensive system. I sat in his listening room, prepared to be amazed by the sonic brilliance of the technology. But a funny thing happened when the music started. I was speechless, but not for the reasons I had anticipated. Instead of sounding great, the music sounded terrible, much worse than in my pedestrian system at home. The more I listened, the more annoying the sound became. 

Seeing my reaction, my friend asked me what I thought of the Linn. "I hate to say it," I replied, "but it sounds awful." He exclaimed, "Yes! Isn't that amazing?" Confused, I asked him whether the whole point of the tens of thousands of Euros he had spent was to make his music collection obsolete? He replied, and I will never forget what he said: "No. What you just heard were all the flaws in the original recording, which were inaudible to you on your normal stereo. Only the Linn can faithfully reproduce all the errors. That's its beauty."

I told this story to the project's steering committee and explained that what I call the "Linn Effect" occurs anytime a new technology solves a problem (or set of problems) and simultaneously creates an equal or greater number of new problems. As evidenced by Facebook’s latest problem with Instagram, social media is a great example of the Linn Effect, and we can find examples of it in smartphones, drones, autonomous vehicles, and many other recent inventions—including, perhaps, the internet itself.

Returning to our paper, it seems to me that Airbnb’s smart-pricing algorithm is yet another example of the Linn Effect at work, this time in AI. It is hardly the only one, however. In 2017, Amazon abandoned an AI project to decrease gender discrimination in recruiting because it actually increased the bias. The AI failed because its models were trained to vet applicants based on a decade’s worth of resumes submitted to Amazon. Since most past resumes were from men—not unusual in a tech company—the AI decided that male candidates were preferable. In line with this conclusion, the AI downgraded resumes that included the word “women,” for example, as well as graduates of certain all-women’s colleges.

The Amazon, Airbnb, and Facebook cases all highlight one critical aspect of getting AI right: correctly training the ML system during development. This task is so difficult and expensive today that only the richest firms can afford it at all, and wealth does not guarantee success. Given the immense challenges inherent in creating fully successful AI platforms, I suspect that we will see more cases like the one this paper thoughtfully describes in the foreseeable future.

The Linn Effect is not all bad news, however. I once described it to a radiologist friend, and her response was that the Linn Effect is real but ultimately beneficial. As she put it:

While your premise is true, I don't think the effect is necessarily a bad thing. For example, when the first MRI machines were made and used, the images were not very good for multiple reasons (poor signal, weak RF gradients, inhomogeneities in the magnetic field, various artifacts that sometimes mimicked pathology, etc.). Tackling each and every one of those problems resulted in more problems, yes, but also more innovations to fix those same problems. This cycle continued until we reached the unbelievably detailed and even beautiful images we now can acquire.

In her view, the Linn Effect leads to a “Linn Valley”—the cycle of problem discovery and solution creation that every great technological creation must cross before it reaches its full maturity. As this new paper illustrates, AI is still in the early stages of this journey, and its creators would do well to remember the full implications of this reality.

The Research

Shunyuan Zhang, Nitin Mehta, Param Vir Singh, and Kannan Srinivasan. Frontiers: Can an Artificial Intelligence Algorithm Mitigate Racial Economic Inequality? An Analysis in the Context of Airbnb. Marketing Science 0 (0) https://doi.org/10.1287/mksc.2021.1295

Almost from the start of the Industrialized Age, the concept of a team has been a part of the working world. It is safe to say that in large organizations today, virtually all managers and employees participate in some sort of team structure, either as leaders or members. Team leadership, consequently, is one of the most common topics in business research today. However, the sheer amount written about teams and team leadership means that researchers sometimes struggle to produce insights that are interesting in themselves and relevant to practitioners. However, a new paper from Inga Carboni (William & Mary), Rob Cross (Babson College), and Amy C. Edmondson (Harvard) looks at teams from a different perspective, and this shift in perspective produces new insights that are worth considering.

The basic hypothesis of the paper is that while the structure of work has dramaticall changed— and continues to change— the management of work has not. Specifically, most teams are managed as if they were a small group of people working together towards a common goal. The reality, the authors, argue is quite different:

In most organizations today, however, this type of small, dedicated team is the exception rather than the rule. Teams are no longer lone islands of activity. Instead, individuals, particularly at more senior levels, routinely lead teams of 20, 50, or even several hundred people across multiple continents and time zones. Teams are not only bigger, but they are also more permeable, more fluid, and more pervasive than in the past, and working on many teams simultaneously is increasingly common. Indeed, senior-level managers might work on as many as 25 project teams in a given week.

For the authors, the traditional focus of team leadership research has ignored this changed reality and has not updated the understanding of the new concepts and techniques needed to manage teams in today's employee networks. As the authors note, "in a workplace in which no team is an island, managers are still using management techniques designed for bounded, dedicated teams with stable membership." Because individual employees can be part of multiple teams and because the shape of those teams can change constantly, team leadership needs to change also from a set of static practices to a collection of dynamic methods that adapt as teams evolve.

The Study

In order to understand how the best team leadership models are evolving in today's networked organizations, the authors interviewed approximately 100 successful team leaders in 20 organizations. The goal of the interviews was to uncover and understand the new practices used by this set of successful team leaders. Given the fluid nature of teams in today's complex organizations, the authors "sought to describe alternatives to the traditional practices that work for bounded—that is, island—teams."

The organizations in their study included a wide range of industries, e.g., financial services, high tech, consulting, manufacturing, food services, and hospitality. The companies ranged in size from several thousand to hundreds of thousands of people. The teams analyzed worked in a variety of functions, including sales, research and development (R&D), human resources, production, and public relations. The authors specifically selected interviewees "who were identified by their senior leaders as having successfully led multiple teams over at least ten years," because the researchers "wanted to create a typology of practices-in-use rather than test causality." 

As the interviews began to suggest new ideas, the authors presented those concepts to the interviewees as well as to "small groups of senior business leaders at roughly two dozen other organizations to validate and refine our understanding.” An important point is that interviews occurred both before and after the onset of the pandemic, allowing the authors to "capture in real-time the adaptations that leaders were making to implement their practices remotely."

The Findings

From their research, the authors discovered a variety of techniques these successful team leaders use to manage their teams' networked ecosystems. I have synthesized their extensive documentation and analysis into five general techniques below:

Technique 1: Build the connections

Rather than building external relationships as opportunities arose, successful team leaders define the purpose of any missing external connection and then acquire it systematically. Importantly, the best leaders divide this task among team members, aligning networking responsibilities with team member profiles. As the authors explain: "As long as the purpose is clear, any member of the team can initiate and nurture the relationships that bring value into the team"—indeed, this distributive approach "further increases the efficiency of the team's ecosystem management."

Technique 2: Shape the work

The leaders the authors interviewed are deliberate in shaping the nature of their team's work. Most team models assume that work is assigned to teams and that they have little say in shaping what they do. The best team leaders, however, devote energy to shaping their team's work before it ever gets to the team. For example, "one senior sales leader in the life sciences industry described engaging key financial sponsors early in the funding process to better align the work that comes into his unit with the aspirations of his employees." He shaped his team's work not with a slide deck but with a single slide that framed serious conversations about his team's future work and approach.

Technique 3: Find the benchmarks

As an experienced team leader knows, it is easy for teams to insulate themselves from outsiders, especially if the team is struggling. The best team leaders, however, use their designed networks to constantly seek and evaluate external practices that might be of use to their teams. As the authors note: "the leaders we interviewed identified and brought best practices into their teams through their or their team members' connections with people who were doing similar work in different geographies, functions, or organizations." In other words, the best team leaders cast a wide net for good practices and then carefully compare them with their own team's needs. They do this not just to find new ideas to improve performance but also to seek out potential collaborators and ideas for future projects.

Technique 4: Expand the support system

It is rare that a team succeeds without external support of one kind or another. The support may be a CEO championing their work or a competent steering committee that removes obstacles to progress. Whatever the form, teams depend on outside help, and the best team leaders devote serious effort to designing and maintaining their team's support ecosystem. This outreach takes various forms, each with its own specific value. Formal decision-makers are an obvious part of this outreach. However, the authors found that good team leaders also invest a great deal of time with two groups that less-effective leaders sometimes overlook. 

The first group is positive influencers—people whose perspectives have a big impact on the future of success of their team's work. Counter-intuitively, the leaders interviewed "tended not to focus their efforts on painting a picture of the worst-case scenario and what would go wrong if the team did not receive adequate resources." Instead, "they or other members of their team drew decision-makers and influencers into an exciting vision of what the team could achieve, given the right support." 

The second group is negative influencers, i.e., people who may hold negative opinions of the team's work. Negative influencers often "were colleagues with different priorities driven by functional commitments, incentives, or personal values in their work." The best leaders either engage with these individuals directly or assign that task to a team member. The important goal is to understand the source and potential impact of negative opinions and to try, as much as possible, to turn negative perspectives into positive ones, at best, or neutral ones, at worst. As one interviewee stated: 

I always ask my team whom they should be talking with about a particular initiative. Their voice will go up, or they'll give some indication as to who their favorites are or whom they're closer to. And so I know, "Okay, those are our influencers for the resister."

Technique 5: Put in the time

One of the more surprising findings of the study was the amount of time that successful leaders put into their network-focused efforts. The techniques noted above were not left to chance; on the contrary, the time needed to execute them well was carefully calculated into management strategy. In fact, when the authors asked the interviewees about the amount of time that they spent managing external relationships, "the team leaders frequently mentioned at least 50%—sometimes 60%—of their time, far beyond what most team models indicate." This high percentage gave the authors cause for a wider conclusion:

Reflecting back on the many conversations we've had with successful team leaders over the years, with our new findings in hand, we now realize how many of the leaders who seemed focused on team-building were actually supporting their team members' efforts to build the external relationships that brought value into the team.

Conclusions

Taking a step back from the specific techniques discussed above, the authors suggest that the way we think about team leadership needs to evolve in two fundamental ways in order to reflect todays' working environments—especially in our hybrid present.

The first evolution required is to reconceptualize teams as working groups embedded within larger organizational networks. Doing so, the authors believe, "highlights the permeability of team boundaries and emphasizes the structure and quality of interpersonal relationships and their impact on work." Given that one person might be on a dozen or more overlapping teams, each with its own set of stakeholders, team leaders should focus on managing not just the work of the group but also the "relationships among team members and between team members and external interests."

The second evolution required is for managers to devote as much time and attention to the ecosystem that surrounds their team as to the team itself. In the traditional model, a team leader's role was generally bounded by the team members and a team's scope of work. Managers often invested time in outsiders—e.g., stakeholders, customers, partners, opinion leaders, and decision makers, etc.— with undefined hopes of finding value in some way. For the authors, however, in today's highly networked organization, team leaders need to systematically manage their external connections, a strategic imperative in which team members should also play a crucial role.

"No team is an island," is the author's fundamental point, and "high-performing leaders in today's collaboratively complex organizations recognize that managing their teams' ecosystems is essential work." Today's best team leaders use their knowledge of the networked ecosystem to cultivate the right external relationships, bring value into their teams, and drive them toward high performance. 

Reflecting on the research, I remember that in every project that I managed successfully at least half of my time was spent exactly as the authors describe. Whether meeting with project sponsors, positive or negative influencers, or external partners, I saw the creation and maintenance of our external ecosystem as a critical part of my job. Indeed, I used to say—only half in jest— that on any big project a major decision has a half-life of two weeks before some force, benign or destructive, starts to undermine the consensus. I was able to focus on the ecosystem for two important reasons, however. One is the nature of how consulting firms operate, which has always been in decentralized hybrid physical/virtual working models. The other is that I was fortunate to have talented team members, the sine qua non of the approach the authors document in their study. No team is an island, and neither is any team leader.

In closing their paper, the authors observe that the pandemic, financial uncertainty, and social unrest are "leaving people more dispersed and more in need of leadership than ever." Relying on old management models, "unaware that the structures and needs of teams and organizations have changed in fundamental ways, is a recipe for failure." This thought-provoking paper not only provides managers with specific and actionable ideas for improving their perspectives and practices, it also suggests new directions for researchers seeking to define and enhance what makes teams work at the highest level. 

The Research

Carboni I, Cross R, Edmondson AC. No Team is an Island: How Leaders Shape Networked Ecosystems for Team Success. California Management Review. September 2021. doi:10.1177/00081256211041784

One of the most difficult moments in life occurs when confronting a question one wishes had never been asked. From job interviews to meetings with bosses to business negotiations, there are endless examples of questions where a fully honest response could result, at best, in an economic loss or, at worst, in the destruction of an entire relationship.

Take, for example, a recently-married female job candidate who is asked about plans to have children in the future. She could answer honestly: “I plan to have children in the next year.” She could refuse to answer: “I would rather not say.” She could lie: “I don’t plan to have children any time soon.” Researchers have examined the implications and consequences of each of these three types of responses, but a fourth option also exists —deflection, i.e., responding to the original question with yet another question that attempts to take the conversation in a different direction.

The consequences of choosing deflection in response to an unwanted question are the focus of research from T. Bradford Bitterly (Hong Kong Technical University) and Maurice E. Schweitzer (Wharton). Their research conclusions are worth considering, given that difficult questions are still very much a part of everyday life.

Their paper begins by noting that declining to answer a question directly risks two possible penalties. First, someone may incur an “interpersonal” cost, e.g., the risk of being seen as less forthright or trustworthy. Second, someone may incur an economic cost, e.g., receiving an offer of a lower salary. These risks arise from the information asymmetry that can be present in certain discussions, e.g., where one party has information that, if revealed, could benefit a counterparty or even cause them harm. The authors label these interactions strategic disclosure interactions. They note that in these situations “individuals are motivated to conceal sensitive information, but the likelihood of disclosure may be profoundly influenced by contextual factors such as competition, social pressure, financial incentives, and even the medium of communication.”

In these settings, even individuals who want to disclose all the requested information may not do so for fear that doing so may bring a heavy cost. As the authors note:

In a negotiation, for example, someone who fully discloses their private information may be exploited by their counterpart. Similarly, in a job interview, someone who truthfully responds to a sensitive question (e.g., how much they made in their last position or about drug use) may be offered a lower salary or fail to receive an offer. In settings involving sensitive information, people often feel compelled to respond when they are asked a direct question, but may suffer economic costs when they do.

The challenge for people asked an uncomfortable question is to respond in such a way that does not put them in jeopardy. Refusing to answer is not an ideal response, since research has shown that people who decline to answer direct questions are viewed as less trustworthy and less likable than individuals who disclose sensitive information. In addition, individuals who decline to answer sensitive questions often reveal information just by declining. For example, someone who responds, “I do not want to answer that question” after having been asked, “Have you ever been convicted of a felony?” suggests an answer that is not very hard to divine. Alternatively, note the authors, individuals may respond to a question by engaging in lying, but individuals who engage in deception risk harm to their long-term relationships should the deception be discovered. Indeed, lying and subsequently being found to be a liar is probably the worst possible outcome.

Given that lying is such a risky (and unethical) option and that refusing to answer may be almost as bad as answering, what, the authors ask, are the consequences of deflecting?

The Studies

To answer the main question of their research the authors conducted a series of studies to assess the impact of various response strategies. The three studies were broadly similar in approach and methodology. In the pilot study, for example, they recruited 99 individuals to ask a job seeker to disclose his salary at his previous job. Some respondents gave an honest answer, and some deflected with the question: “Will the answer impact who pays for the coffee?” 

The 99 participants then rated each candidate along several interpersonal dimensions such as warmth, competence, trustworthiness, and likability. The authors also asked participants to rate the extent to which a candidate’s response was “appropriate,” “funny,” “humorous,” and “suitable.” 

The authors found that job seekers were rated as warmer, more competent, more trustworthy, and more likable when they deflected than when they refused to answer a direct question. Participants also rated the deflection as “funnier and more appropriate” than a direct refusal.

Their next study was more complex. In this experiment, the authors asked 232 people to imagine that they were the owner of an art gallery negotiating the sale of the last available painting in a set of four works by a well-known artist. The participants were told that a buyer who had the first three pieces in the set would pay much more for the last piece than someone who did not have other works in the collection. Thus, a buyer’s other holdings in the series became a key piece of information they needed to gather. When asked, some buyers refused to disclose other holdings in the set: “I’m not prepared to discuss my collection right now.” In other cases, the buyer admitted to ownership: “I did purchase the other Hearts pieces in the collection.” In a third group, the buyer deflected: “How much do you want for this piece?”

As in the first study, deflection proved to be the response with the best interpersonal and economic outcomes. Deflectors paid less for the painting than those who answered directly and deflectors were perceived as more trustworthy than those who refused to answer. As the authors note:

In the deflection condition, only 12% of the participants asked the buyer more than once if they had the other pieces in the Hearts collection. Deflection was also very effective at concealing information. Individuals thought that individuals who deflected questions were less likely to have the other pieces in the collection than individuals who explicitly declined to answer the question and individuals who disclosed information about their collection. The disclosure of this sensitive information directly increased economic surplus for individuals who deflected. 

In their final study, the researchers introduced a different response strategy known as paltering. Paltering occurs when one uses statements that are in themselves true to create a misleading impression. For example, someone asked if she planned to have children palters by replying “children are a big responsibility” or “children need parents who are secure in their careers.”

Figure 1: Summary of the methods of responding to direct questions during strategic disclosure interactions. (Source: Authors)

In this last study, 304 people completed much the same process in the study noted above. However, in this final experiment, both deflection and paltering were tested. Some buyers deflected: “Can you tell me more about this piece? What price are you asking for it?” Some buyers paltered: “I’ve been looking to buy one.” A final group lied in response to the ownership question: “No, I do not have any other pieces in the set.” 

The results of the final study indicate that deception created the most economic value and favorable interpersonal impressions up until the moment the lie was discovered. At this point, the outcomes became the worst of all. That is, “after revealing the truth about the buyer’s history and interests, the sellers’ ratings of the buyers’ trust and likeability were significantly lower.” In contrast to engaging in deception, however, “deflection had lower interpersonal costs after participants discovered that the buyer had other pieces in the collection.” Moreover, much as with deception, paltering yielded positive outcomes until the truth was discovered, after which it created similar outcomes as lying.

Figure 2: Summary of the economic and interpersonal costs of the methods of responding to direct questions during strategic disclosure interactions. (Source: Authors)

All in all, across the various studies, participants were more willing to negotiate with individuals who deflected because sellers viewed that negotiator as more likable and more trustworthy. That is, they viewed the buyer who deflected as “having concealed less information about their collection than the buyers who engaged in deception.”

Conclusions

The late linguist Paul Grice described rules that people intuitively follow to make their communicative efforts effective. The four Gricean maxims are:

1. The maxim of quantity, where one tries to be as informative as possible, giving as much information as is needed, and no more.

2. The maxim of quality, where one tries to be truthful and does not give information that is false or that is not supported by evidence.

3. The maxim of relation, where one tries to be relevant, saying only things that are pertinent to the discussion.

4. The maxim of manner, when one tries to be as clear, brief, and as orderly as one can in what one says, avoiding obscurity and ambiguity.

The authors note that deflection paradoxically both violates and invokes the second maxim. By redirecting the conversation, deflection enables individuals to conceal information they do not wish to share, while at the same time requesting similarly valuable information from their counterpart. The authors’ research suggests that when faced with an unwanted question or a question that would put someone at a disadvantage in a negotiation, deflection is the optimal response. As the authors conclude: “By responding to a question with a question, individuals can maintain favorable interpersonal impressions, capture economic surplus by avoiding revealing potentially costly economic information, and avoid the risks inherent in using deception.” Moreover, the value of deflection is even higher in those situations where even a non-answer becomes an answer. For example, “when an employer asks a prospective employee if they have ever been convicted of a felony or a negotiator asks the other party if they have other offers, explicitly declining to answer reveals information.”

In closing their paper, the authors caution that effective deflection is not so easy as it may seem. As they note, conversation norms guide individuals to answer questions directly; thus, "it may require effort and practice to both violate and invoke this conversational norm by deflecting a difficult question." On the other side of the conversation, "interviewers, negotiators, and debate moderators should anticipate and guard against deflection." In addition, questioners should recognize that deflection may yet convey information; specifically, “someone who deflects reveals that [he] would prefer to avoid discussing the topic.” Moreover, if deflective questions are delivered defensively or with anger, they could signal the very thing that the deflector is trying to avoid saying. That said, the authors conclude their paper by noting that sometimes the best way to answer a question may be to pose a new one. 

Of course, no discussion on deflection is complete without noting the arena in which it is encountered most often: politics. Deflection has become such an overused tactic in political debates and discourse that anyone might wonder if it is still of any value in other settings. Unfortunately, the authors do not address this question; nor do they address the ethical implications of avoiding a direct answer when one is possible. Indeed, it is interesting to consider which is worse: deflection or hurting someone with an honest answer? The latter may be the purer response, but even the most severe moralist might admit that there are questions in life that are best left unanswered.

The Research

Bitterly, T. B., & Schweitzer, M. E. (2020). The economic and interpersonal consequences of deflecting direct questions. Journal of Personality and Social Psychology, 118(5), 945–990. https://doi.org/10.1037/pspi0000200