What Is This Thing Called Culture?

A recent post on LinkedIn discussed the reasons Publix was cited as the best place to work among very large companies. Having had dialogue with HR executives from Publix I sense a considerable amount of pride they feel when they tell people where they work. And since the award was the result of random polling of employees, rather than a carefully crafted mission or culture statement that was intended to make Publix look good it has credence. The organization is privately held… by employees. So that may be a reason for their positive reactions. Owners tend to want to maximize the positive PR, since shoppers will prefer stores where they are treated well by people who seem to be happy to be there. But what role does this thing called culture actually play in eliciting the behavior it wants?
Culture is like the water fish swim in. It has a major impact on the lives of the denizens even though they are not aware of it. If you ask a fish why the water is so blue you would probably get the response “the what?” So, if the setting in which employees work does impact their attitudes (and research indicates it certainly does) how do you define and evaluate an organizational culture? Having co-authored the book “Rewarding Performance Globally” with Fons Trompenaars, leading cross-cultural researcher and guru, I became a student of the relationship between organizational culture and workforce culture. I teach a Global Workforce Management course for DePaul U. in their MSHR and MBA programs and focus on the issues created when someone with a specific cultural orientation enters an organization with its own cultural orientation.

If the organization has a culture that results in rewards being tied to individual performance, but an employee has been socialized in a culture that is collectivist, there is a difference of perspective that may impact that person’s acceptance of the prevailing culture. It will likely impact whether they choose to stay with the organization and perform well. People will be inclined to behave in a manner that seems to be encouraged by the organization’s reward system. Investment bankers almost brought the economy down with instruments that they created and were richly rewarded for. Regrettably for the banks what the asked for (by rewarding it) turned out in many cases to be disasters. Wells Fargo employees exhibited fraudulent behavior because the pressure to meet new account goals was so intense. “What you measure and reward you most surely will get more of” is a principle management should consider when eliciting patterns of behavior. No one knows how the employees felt when they took these actions, but one might surmise those with strong moral compasses probably were not happy with their actions.

So how does an organization define the type of culture it believes will be optimal? Even more importantly how does one define culture? After sitting in many client meetings discussing culture I believe defining it is one of the most daunting challenges management faces. Numerous shelves dedicated to culture exist in most bookstores and libraries. The advice varies from encouraging behavior that emulates Genghis Khan to emulating Jesus. And how an organization creates the culture it desires is not clear. One of the findings that comes out of the extensive research is that management must exhibit the behaviors it wants employees to exhibit. If a multinational pays commission on foreign business that was obtained by bribing a government official it sends the message that business results take priority over lawful behavior. If managers get results by treating their employees like serfs that also sends a signal. And if the first step taken by an organization facing an economic downturn is to downsize the workforce it sends the message that employees are its most disposable, not its most valuable, asset. No matter what mission statements, policy statements or cultural definitions say they must be made real by behaving consistently with what they prescribe.

Perhaps Publix could not define its culture specifically if asked. But if the example set by management is consistent with how it wishes to conduct business and the employees can accept the direction provided it may not be all that important. And the fact that the #1 ranking was based on a random sample of employee opinions there must be a very high level of acceptance. Other organizations may rank being the most profitable or being the organization with the most rapid escalation in stock price above being seen as a great place to work. That is their choice. But increases in stock price or profits are most often the result of the workforce being competent and committed to the success of the organization.

My first job out of undergraduate school was with Johnson & Johnson. As a part of the training for those in the executive development program the J&J Credo was explicated… so deeply that I wondered what the payback could be from the big investment in understanding it. When the Tylenol crisis happened and everyone reacted immediately in a manner consistent with what the Credo stood for it became clear to me. That was the organization’s way of defining and creating the culture it wanted. Since it is not possible to prescribe the behavior desired for every possible occurrence the Credo served as a nudge in the right direction. The one-page Credo does not look like a detailed description of the desired culture. But it seemed to produce the desired effect when it counted.

Did You Hear The One About The Data Scientist And The HR Professional?

The technological advances in data analysis are presenting new opportunities for HR professionals to understand what causes what and to project what might happen in the future based on what has happened and is happening. But taking advantage of what technology offers is sometimes easier said than done. Most HR practitioners lack PhDs in quantitative methods. There is a limit to the number of courses in any curriculum leading to a BA, and even an MS, in a field related to HR and there is a lot to learn about the principles underlying sound workforce management, so it is unlikely HR practitioners are going to know enough about data science to take full advantage of what technology offers. Conversely, data scientists tend to focus their formal education as well, with limited exposure to behavioral science topics. So there generally is a limited amount of overlap in the knowledge possessed by HR professionals and data scientists.

One of the requirements for people from different disciplines to be able to work together is that they share language and knowledge. If one party speaks Chinese and the other English, there are tools for easy language translation available today. It is much more difficult to translate the technical language data scientists speak into a form that HR professionals speak. This has been an issue for a very long time when academics do research and practitioners attempt to understand and apply research findings. The scientific method is rigid in its requirements and compromising on those just to make research results accessible to practitioners is a breach of the rules (but there is no rule against trying to translate findings at least partially). One of the potential solutions to the lack of common ground is to have people with some knowledge of both fields do a translation of what research has found in the pop literature, where practitioners tend to get their information. But the translation often results in authors “cherry picking” research findings that support the point an author is trying to make, consciously or not. In some cases, the translator has a surface level of competence in interpreting research, and this can result in attempting to generalize findings beyond where they would apply.

A popular book attempted to support the claim that extrinsic rewards diminish the ability to experience intrinsic rewards. “Evidence” included a controlled lab study that found that people threw tennis balls at targets longer if they did not receive the very small rewards on offer. Even though the study seemed to meet the “internal validity” requirements for a valid study (it was well designed to determine what would happen under those specific conditions), it attempted to apply the lab results to employees working for many years, often doing things viewed as undesirable, to support themselves. The “external validity” requirements for a study to be generalized to a different context were definitely not met. To make matters worse numerous field studies done in contexts similar to real world contexts (thereby meeting the external validity test) have refuted the lab study findings. The bottom line is that whoever uses research needs to have the ability to determine if it applies to the situation being evaluated. And not knowing that research needs to meet both validity tests can mislead someone attempting to apply research findings where they do not apply. The growing adoption of evidence-based management in decision-making is a positive trend. But the evidence used must be sound and must apply to the matter at hand. Workforce analytics can be of enormous value to an HR practitioner. But there are two requirements. First, the analysis must be done in a manner consistent with the principles of the scientific method. Second, lab tests must meet the external validity test just discussed if their results are to be applied to the field.

One of the most common applications of workforce analytics is the attempt to predict who might be prone to voluntarily leaving the organization. Top performers in critical occupations will always be in short supply and will be both expensive and difficult to replace. So being aware of issues that may lead to turnover before they reach a critical level can enable the practitioner to adopt preventative strategies. When someone resigns they have often left (mentally) already and recovery strategies are much more difficult to pull off than preventative measures. But how does an HR practitioner get inside someone’s head and discover if they are edging towards the exit? By using data and finding causal relationships between factors leading to termination steps can be taken to focus on what might be most effective to address. Employee satisfaction certainly will influence whether someone is receptive to a call from a recruiter or an internet posting of an opportunity. So will employee engagement. And using technology enables organizations to get a reading on where their employees are on the satisfaction and engagement scales. Some models for predicting potential turnover also include characteristics of who have left in the past, either in the organization being studied or in other similar organizations (i.e., length of service, age, etc.). The use of big data, machine learning, and AI can be used to create prediction models.

Data scientists are apt to believe these models are strong predictors of what will happen. They create algorithms and apply them. But two potential limitations are often ignored:

  1. Data is from the past and the present and if the future is going to be different the predictive power may be diminished
  2. Data scientists are working with numbers, without consideration of human factors that may be better predictors.

If an employee is subjected to real or perceived mistreatment by a manager on Monday, a notice of departure may be tendered that week, even though that employee had not possessed the “prone to leave” factors, at least until that event. There are other signs of trouble brewing that may have been visible to an HR professional but not included in the prediction model. Behavior, events in an employee’s personal life, feedback from peers, subordinates, and managers… all of these indicators of an employee’s mindset may be recognizable by the HR professional. So in order for predictions to be as good as they can be it is prudent to augment the data analytics with the human stuff, either by building factors into the predictive model or by an interpretation of quantitative results using qualitative measures.

Recently I was asked whether a client organization should assign a data scientist to HR. I am in favor of collaboration between people from the two disciplines and believe that credible evidence can come in the form of “hard” evidence (numbers) and “soft” evidence (professional knowledge and judgment). But co-locating people with different perspectives does not necessarily result in the pooling of knowledge and consideration of both points of view. So should HR professionals add “speaks analytics” to their competency model? Yes, to some degree. Should data scientists add “respects the impact of human behavior” to their competency model? Yes, to some degree. Who has to walk the furthest out on the bridge between them to achieve mutual understanding will depend on the issues being dealt with and the parties involved. As a faculty member for DePaul U.’s MSHR program, I try to ensure students understand the need to respect both the quantitative and qualitative perspectives Although I do not know I hope the faculty developing data scientists are equally respectful of the “other way.” If nothing else, balanced approaches lessen the fear of being replaced by robots.

The Gender Pay Gap: How Large & Caused By What?

There has been a lot of attention paid to the gender pay gap recently. If there is a gap caused by gender-based discrimination it should be remedied. The law of the land and a sense of fairness dictate that this should be done. Many who are skeptical about the magnitude of a gap believe numbers in the press are overstated. Large numbers increase readership so the journalist’s temptation to magnify differences is understandable. But it is unprofessional to artificially magnify something in order to get noticed, particularly on sensitive subjects. Some of the more responsible analyses of the issue use statistically sound measurement techniques but determining how large any gap is and what causes it are tricky things to pin down. Before a pay gap can be addressed its causes and its magnitude need to correctly determined.

Calculating the gap can be done in many ways. Many of the articles published about the issue use an organization-wide average to calculate the relationship between the two genders. Regulatory agencies often use this approach when determining whether an organization is guilty of discrimination. Other studies aggregate the data nationally, by industry, by occupation and even globally.

Scenario 1:

An organization with 100 employees has an even gender distribution. The male CEO has a total direct compensation of $1,000,000 and five other male senior executives average $ 250,000 each, while the remaining 44 males and the 50 female employees each have an average pay of $ 50,000. When calculated on an organization-wide basis there is a significant gender pay gap between the average pay for all men and all women. Is there gender pay discrimination? Is it of the illegal type? Does the difference raise any concerns? What additional information would be necessary to decide on any action?

Subjecting the aggregate results to further analysis might reveal that a glass ceiling exists, which results in the highest ranking (and highest paid) employees being male. Gender-based discrimination in employee selection and career management is as illegal and unethical as pay discrimination between similarly situated employees but it has different remedies and should be correctly labeled. And regulators should do more in-depth analysis and pressure organizations to fix the real problems.

Differences in pay across genders can also be caused by different distributions across occupations. My PhD thesis was on the determinants of occupational worth. The relative value organizations place on jobs is supposedly based on their relative economic value to the organization. For example, a senior executive will have a direct and significant impact on the organization’s performance while a clerical person will have a much smaller impact. Job evaluation systems have been used for a century to enable organizations to determine the relative internal value of jobs and occupations. U.S. labor law (Equal Pay Act, Title VII of the Civil Rights Act) specifies that a system for establishing the relative value of jobs must consider the skill, effort, responsibility and working conditions required in the jobs. The gender, age and race of incumbents should have no impact. But in the U.S. there historically have been significant differences in the gender mix for many occupations. And if discrimination is caused by undervaluing jobs that are filled by women is that as odious as direct discrimination based on someone’s gender? If the concentration of genders in occupations is due to personal choice there is no basis for blaming illegal discrimination that is a result of such concentration. But when artificial obstacles exist that deny equally competent women or minorities access to certain jobs that is discrimination. And if the undervaluing of female dominated jobs is merely another way to discriminate there needs to be some remedy.

Some occupations have historically enjoyed higher pay than others because of the existence of influential blue collar unions. Blue collar jobs have historically tended to be paid more than white collar jobs, even though the latter might require more education and judgment. As the impact of unions has declined their impact has become less of a factor but historical patterns tend to persist. Early job evaluation plans were designed for manufacturing environments and placed heavier weights on factors such as physical effort, physical skill, adverse working conditions and hazards. When these plans were used to evaluate both blue and white collar jobs their design favored blue collar jobs. As gender discrimination became a political concern in the 1960s laws such as Title VII of the Civil Rights Act were passed to minimize this discrimination. Many job evaluation plans were revised so that heavier weights were placed on education, mental skill and mental effort. Few plans had been designed in such a way as to appropriately value both blue and white collar jobs and how much revision was appropriate to the weighting of the factors determining value was difficult to determine. A National Academy of Science study was done in the 1980s measuring the impact of job evaluation on gender-based pay discrimination. Principles were articulated that enabled those using plans to ensure their design was balanced, mostly ensuring that the factors gave credit for all types of skill (mental and physical), effort (mental and physical), and responsibility (influencing and managing). In some states (and in Canada) pay equity regulations were established that were aimed at discrimination, but it is unclear how much they have resulted in less discrimination.

When the gender pay gap issue heated up in the 70s and 80s there was consideration of the concept of “comparable worth.” That is, if two jobs were considered to be comparable in “value” they should be paid the same. U.S. law only required that “substantially equal” jobs should receive equal pay, but some politicians were amenable to including a “substantially equivalent worth” test. A committee of representatives from ASPA (now SHRM) and ACA (now World at Work) on which I served studied the issue of determining comparable value. We concluded that attempting to legislate this at a macro level would create clashes of subjective opinions about relative worth that could never be settled. As stated earlier, each organization is likely to view relative value differently based on whether occupations were central to their primary business and critical to their performance. I once asked representatives of Arthur Anderson and Anderson Consulting whether Accountants or IT Professionals were more valuable – and they responded with different answers. Accountants are generally considered to be more critical to accounting firms while IT professionals are viewed as more critical to IT firms.

Scenario 2:

Assume faculty members at a university are paid differently based on their field. STEM (Science; Technology; Engineering; Mathematics) field incumbents are paid 40% more than those teaching in the Liberal Arts fields.  A large percentage of STEM field faculty members are males and a large percentage of Liberal Arts faculty are females. Is this gender based discrimination? What factors should be considered when deciding?

This type of difference is often the result of supply and demand in the labor market, or at least that is the defense offered when relationships are challenged. There has been a shortage of STEM graduates in the U.S. for some time, causing demand to exceed supply. And short-sighted policies that restrict the use of H1-B visas exacerbate the shortage. As we learned in Economics 101 that will drive the market price up. So even when males and females are doing the same general type of work (teaching) the result of market forces and different gender mixes between fields can produce what looks like gender pay discrimination if these causal factors are not recognized. The same holds true within fields… Nuclear Engineers command a higher price than Civil Engineers, even though their education is similar. On the other hand, their relative value may change if a surplus of Nuclear Engineers manifests while Civil Engineers become scarce. One of the defenses for paying faculty differently might be the fact that there is a business necessity to respond to competitive pay levels in the labor markets.

Scenario 3:

Two Accountants were hired three years ago out of school with BS degrees. One is a male with a current salary of $70,000 and the other a female with a salary of $60,000. Is gender-based pay discrimination a problem? How would this be determined?

A sound pay administration system will determine pay actions based on employee competence and performance and if policies are administered equitably employee pay levels should be defensible. So if the two Accountants had received similar performance ratings and were equal in competence any difference in their pay rates should be explored. But there should also be attention paid to new hire and new entrant pay rate determination. If gender differences impact starting rates fair equivalent administration of pay adjustments going forward will not correct the differences. The problem was born at the point of hire. Research suggests females are often less aggressive in negotiating pay rates and this can result in start rate differences across genders. This can be a major concern in many public sector organizations using automatic step-based pay progression systems, since start rate differences are perpetuated. On the other hand, forcing everyone to start at the same step when entering a job may result in unfair practice when there are large differences in qualifications across incumbents. Recent legislation and supporting regulations have been implemented in the U.S. to preclude asking about past pay levels, which could moderate the perpetuation of past discrimination in determining hiring rates.

Unwarranted pay discrimination based on personal characteristics has detrimental effects, in addition to being wrong. If equally qualified incumbents of a job with equal performance are paid differently based solely on gender, race, age and other prohibited bases there is a clear violation of the law. And there are consequences. If I am a member of a gender or race that is discriminated against I will view my employer’s pay management system as unfair and inappropriate. This will lead to diminished satisfaction and engagement, which will tend to have a detrimental impact on my contribution. This makes discrimination something every organization should attempt to prevent, even if fairness and ethical behavior are not considered… it is just good business. Well managed organizations should have policies and systems in place that control pay actions, ensuring pay is determined by the value of the job and the contribution of the individual. What someone looks like has nothing to do with what constitutes fair rewards.

Scenario 4:

A department with ten females and ten males does a comparison of pay rates and finds that the males are being paid 15% more than the females on average. The typical length of service is similar across genders. However, the average performance ratings for males have been higher than those of females consistently over the last six years. Is there gender-based pay discrimination? How could this be determined? What might be causing the pay differences?

The majority of organizations claim to use merit (performance-dependent) pay administration systems. The policies regulating pay actions typically base an individual’s pay action on performance, as well as competence in the job and where the individual is being paid within the established pay range. One of the greatest challenges associated with performance-based pay is to ensure that those rating performance have equivalent standards and that their ratings are based on job-related results and behaviors. If personal bias influences ratings this can be a form of discrimination and should be eliminated. But changing someone’s biases is a tall order.

The last decade has been technologically turbulent. New tools have come on the scene, such as AI, machine learning and evidence-based decision-making. When new technology emerges there is almost always a shortage of people with the skills to effectively utilize the new technology. That often leads to very large spikes in competitive pay levels. For example, workforce analytics is all the rage now and data scientists who can do the work command a premium price. If organizations employ people with these skills they will almost certainly create large differences between current employees and the new folks in the neighborhood. Although the other employees may view the magnitude of difference in pay as unreasonable an organization needs to meet competitive pay levels. But when even one person with hot skills is employed it creates a new “referent other” for all employees (referent other in equity theory is the person people select as a reference point to determine if their pay is reasonable). And if the people with the high priced skills are generally male (I have not found data on the gender mix for data scientists) it can exacerbate numeric gaps between gender pay levels. Over the last few decades more females have entered STEM (science, technology, engineering and mathematics) fields but statistics show the majority of incumbents to be male. This is a difficult difference to attack, since the choice of field is almost always voluntary.

Another trend is the increased use of non-employees to perform work. There are a number of reasons organizations are utilizing outsiders. One is the lessening availability of skilled people who are interested in an employment relationship. Those who work on gigs may find it allows them to focus on the specific skills that pay a premium… if they were to be employed they may have to do other things that are of less value. Another reason is that the supply of skills can be global if the work can be performed virtually, which more of it can. This avoids the necessity of relocating and also enables people to extract high pay while living in areas with reasonable living costs and avoiding long commutes. A final reason for using contractors and freelancers is that they do not enter the organization and remain hired hands. This might lessen the focus of employees on the gap between their pay and those working contractually. And it may even lessen the gender pay gap if male contractor pay does not enter into the calculations.


What is quantitatively measurable may not tell the story as to what lies behind pay differences. Governmental attempts to find illegal pay discrimination often result in findings that are not warranted by the facts. Some of their evaluations create statistical anomalies that do not represent true discrimination. The regulators also tend to label all differences as pay discrimination, even though they might be attributable to bias in selection or career management… or even people voluntarily choosing occupations that command lower pay. Statistical correlation does not constitute causation. If some of the conditions mentioned here exist they may produce results that look like pay discrimination, but that may be something entirely different. It is prudent for every organization to not only run statistical tests required by law but to explore much more deeply to determine the causal relationships. If differences cannot be attributed to anything other than personal characteristics a fix is in order. Waiting until an accusation has been made puts the organization on the defensive and is unprofessional, despite the fact that many attorneys recommend that discovery be averted. This argument is facetious… the facts are the facts and it is certainly easier to deal with them if the organization has defined the reality and explored in advance how it can be dealt with.

The subject of pay discrimination creates strong emotional responses. And people act on their perceptions, which become their reality, even when their perceptions are wrong. It is therefore important to measure the magnitude of pay differences accurately and to attribute it to the correct causes. If gender pay differences are due to occupational clustering caused by unwarranted barriers then the focus should be on alleviating those barriers. If the differences are due to personal biases this necessitates a different path to correction. If numeric differences are caused by factors that are not related to discrimination analysis should be sophisticated enough to avoid mislabeling differences. Working on the wrong problem is unlikely to produce an effective solution. And accusing organizations of things they are not guilty of will not result in a rational reaction. So let’s find the illegal and unwarranted discrimination and devise strategies to remedy it.

Replacing Human Judgment With AI: Danger Lurks

An organization is considering the adoption of a particular type of incentive plan to increase the motivation of its employees to extend their best efforts and to focus that effort on meeting organizational objectives. Using a data scientist the HR Director requests that research be performed that will predict the likelihood of success with a plan that was effective in another division. Using machine learning tools the data scientist examines the results experienced by the other division to conclude that the same plan will be effective if used in other divisions.

The HR Director is very familiar with the benchmarking principle that in order for results produced in one context to be likely to have the same results in another context the two contexts must be substantially similar. This principle is based on concepts from research that an internally valid study must also be externally valid (generalizable) in order for it to produce the same results in another context. A recent book argued that extrinsic rewards reduced intrinsic rewards based on a lab study that involved people throwing tennis balls at targets for short periods for insignificant rewards. When one compares that context with one where people work to support themselves over a career, often doing very unpleasant things, it seems absurd to claim the two contexts are even remotely similar.  Yet a machine learning algorithm may well make that error because its predictions lack any understanding of human nature. A Facebook algorithm was recently accused of gender discrimination because it sent ads for jobs in STEM fields much more often to men then to women. However, it was found that the decision was made on economic factors (it was more costly to reach working women), rather than on a judgment as to whether women were considered to be capable of doing jobs in the STEM arena. Regrettably on the surface it appeared that the results supported the latter assumption.

Machine learning tools could be incapable of predicting the success of an incentive plan considering the impact of differences between the two divisions mentioned earlier. For example, if one division had a hierarchical culture while the other allowed employees to take the initiative to deal with issues this might result in different levels of motivation to excel. Recognition by employees in the top-down control culture that they were not empowered to impact results can certainly depress initiative. That would certainly limit the motivational impact of the plan in one type of culture. So the differences in the cultural contexts might make comparisons across the divisions inappropriate. When working in the Middle East I had an HR executive ask me if U.S. style incentive plans would be effective if employees believed results were out of their control and that it would be irreverent of them to assume they could have an impact on outcomes that were determined by a higher power. Lacking an answer based on experience I had to apply behavioral science principles and suggest the cultural difference would certainly impact individual decisions about taking the initiative. A manager might be able to convince employees that their talent was a gift and that it was their responsibility to use the gift but the different cultural orientation certainly changed the game.

Melanie Mitchell, a researcher at the Santa Fe Institute, observed in a recent New York Times article that AI systems lack an understanding of situations and the meaning of the differences across situations. Again, what works in one context needs to be understood before projections can be made about what results would be in another context. Her long experience with AI systems has made it clear to her that even minute differences in contextual characteristics can have a major impact on outcomes. Someone hacking into a system and making changes that are so minor that they would not be detectable by humans could dramatically reduce the effectiveness of the algorithms. And if the target is a control systems for a city’s power network this could be too important to overlook. The fact that machine learning algorithms have been trained in specific contexts makes them vulnerable to erratic results when contextual details change.

Back to the HR Director deciding whether to adopt the incentive plan from one division in another. The division that had used the plan had included all employees as participants. But the division executive considering adopting wishes to carefully select plan participants based on their ability to significantly impact results (perhaps including only managerial and professional employees). This decision would be driven by human judgment, applying the principle that those lacking the latitude to vary their behavior and control their results would not be affected by participation. The algorithm based on the division that had used the plan may in this case be a bad choice to project the success in the other division. Interestingly the algorithm might find the new application to be a good choice because it would cost less (smaller participant payroll), when in fact the trimming of participants may result in strong negative reactions by those excluded. The lower cost would be detectable by the algorithm but it would be incapable of detecting the probable angst felt by non-participants, which is attributable to human characteristics.


Mitchell points out in the NYT article that humans can generalize what we know, form abstract concepts and make analogies. So the HR Director would have to consider whether the exclusion of some employees from the plan would result in negative outcomes that would be significant. Current AI technology cannot help with that judgment, lacking the “fuzzy logic” we use to combine black and white in a way that produces the best shade of grey. Human judgment is certainly not infallible and cognitive bias is always leading us to make less than optimal decisions. But in complex contexts where multiple factors impact systems someone has to come up with the best choice among imperfect solutions. AI is certainly gaining in its ability to help us extend our human reach, as well as to enable us to turn work that is fully defined and requiring speed, precision and accuracy over to systems that will almost certainly be better at doing that kind of work. Our biggest mistake would be to assume AI will surpass human cognition in all types of endeavor. Organizations are not likely to let AI make decisions about selecting the next CEO without significant human intervention.

We Are All Biased: Acknowledging That Can Help Moderate The Impact

The work of Kahneman & Tversky on cognitive bias led to a Nobel Prize. It also provided us with a window into the inner workings of our mental processes. Whatever the cause our biases influence how we view things and make decisions. Knowing we are biased can help us pay attention to tendencies that may lower the quality of our decisions.


  • Think others agree with us more than they actually do
  • Accept data/hypotheses that agree with what we believe/want to be so more readily… resist contradictory evidence
  • Accept too readily conclusions based on inadequate samples

“four employees complained about this policy last week… angst is rampant”

  • Rely on literature that reports only successes and not on failures
  • Believe that intuitive conclusions are valid
  • Overuse similarity as a simplifying heuristic

“candidate is great (thinks/looks like me)”

  • Assign weight to evidence based on its availability

“recency bias; familiarity bias”

  • Do not adequately consider regression to the mean

“why did the best performers last year seem to do worse this year?”

“giving a QB a huge raise because of a great year may prove to be foolish”

  • Assume quantitative data has higher validity than subjective data
  • See puppies and other shapes in clouds… this is System 1 thinking (intuitive/automatic/emotional); realizing that it is silly is System 2 thinking, which takes considerable effort
  • Think the world makes much more sense than it does
  • Sometimes answer a simpler question when we have no answer to a complex question – System 1 is puzzled and System 2 goes to work
  • Can learn to apply System 2 control over impulsive conclusions reached by System 1 with training but emotional intensity of issue makes it harder
  • May use repetition of anything, even falsehoods, to make something seem more feasible
  • Are overly prone to assume causation when we see correlation
  • Believe we are too intelligent to be prone to bias… believe that rationality is used and overrides impulse
  • Know less about how we think than we think we do
  • Are primed to make associations by recent/repeated experience
  • Falsely believe that a “hot hand” in sports or in gambling exists
  • Believe a lot of experience makes us wiser – a lot of experience is generally good but we need to organize the experience to make it relevant/useful


Clear thinking requires hard work… and overcoming what comes easily.

Overcoming Resistance To Sound Evidence-Based Recommendations

We all have made recommendations based on sound evidence, only to find the decision-maker did not share our belief that it was the right thing to do. The human tendency is to push harder, emphasizing the evidence that supports the recommendation. But the behavioral scientist Kurt Lewin found that decreasing resistance was more effective than increasing force.

When decision-makers are considering actions that involve significant change barriers are common. People must believe that the benefits to them outweigh the costs to them if they are to consider change. Resistance is apt to be very strong if the decision-maker must reverse a prior decision (s)he made. And if an admission that the prior decision was wrong is required the intensity of the resistance escalates. As evidence-based management becomes more widespread it is likely that improved analytics will increasingly show prior decisions were less than optimal. Many of those prior decisions were made using experience, intuition and judgment, since tools like workforce analytics had not yet been sufficiently developed or accepted as another valid form of evidence. As technology develops at light speed decision-makers are often overwhelmed by the volume of new forms of evidence and they may be ill-equipped to understand the data science tools that are used to generate it. Consequently, they may feel their decision making competence is being challenged by things they do not understand.

Practitioners can be intimidated by a specialist in another discipline using “tech speak” since they do not speak the language or understand the underlying technology. Understanding how financial analysts, IT specialists or data scientists have formulated their recommendations helps immensely when one is deciding whether or not to accept them. That is not to say that one must be able to perform a multiple regression analysis on a data base to discover what factors impact something. If I am told that education, experience and performance appraisal ratings strongly predict someone’s pay I may accept that because it is reasonable. If I am told gender and race are strong predictors of pay it can alert me to something about which I should be concerned. After all, acknowledging that gender or racial bias is a problem in the organization is both difficult and distasteful.

It is not necessary to be able to do the analysis in order to use the information it generates. But when data analytics produce surprising results which don’t align with common sense or professional knowledge it is likely that the user may be reluctant to act on the evidence without being certain it is valid. And, as said earlier, if the evidence calls into question strongly held beliefs or shows a prior decision to be flawed the level of reluctance escalates even more. So how does a practitioner acquire the knowledge and skill that is necessary to integrate data science into the decision-making process?

If two parties do not share a common language it is like the time I tried to find insect repellent before going into the jungle in a city where no one spoke English. When I found someone who was multi-lingual to act as a translator the problem was solved. But finding someone who is competent in both data science and the discipline of the decision-maker is a big challenge today. This historically has been a challenge when IT people have dealt with neophytes. It also has been common for academics and practitioners to have trouble communicating. The IT challenge was dealt with by creating Systems Analysts, who knew enough about both what was needed by the user and what it would take for Programmers to meet that need. So the intermediary approach is a feasible way to resolve the communication problem. But this still requires finding a multi-lingual party.

Teams are widely accepted as an effective way to resolve issues, assuming the needed skills and knowledge are possessed by someone on the team. So replacing a sequential process by a concurrent process can improve the creation and utilization of data science efforts. Having data scientists do their technical work and then presenting the results to practitioners can result in ineffective communication… the translation challenge. But if both parties work together from the start it is possible for the practitioner to convey what is needed to the data scientist and for the data scientist to inform the user of the alternative approaches to performing data analysis and how each would be likely to address the needs. If two parties are separated by a canyon building a bridge is the way to bring them together. But in order to fully integrate the two parties it is necessary for each to go halfway across the bridge and to conduct an open dialogue in the middle.

This will require practitioners to acquire at least a basic understanding of the scientific method and of how the analytical tools work. It will also require a data scientist to learn more about the issues faced by the practitioner and the type of analysis that will help resolve dilemmas. For example, if the stakes are high the degree of certainty produced by the analysis may need to be high as well. But if the practitioner is seeking an approach that will improve the ability to predict something (i.e., an employee with critical skills leaving the organization) the data scientist might be able to produce an adequate result with data that is of lower quality. Knowing whether approximation or precision is required helps to guide the approach to analysis. As artificial intelligence and machine learning techniques evolve it is increasingly possible to use qualitative data and even unstructured data to produce an adequate result when a precise answer is not required. By approaching the problem with continuous dialogue the two parties can both become more efficient and effective because they share a common understanding of what is needed and how needs will be met.

Practitioners have a wealth of options for developing a basic understanding of data science. Online courses are available and books are being released continuously, both containing the type of knowledge needed by the practitioner. It is somewhat more difficult for the data scientist, since much of the knowledge about practitioner issues is accessible only by hands-on experience. But by using a continuous dialogue between team members this bridge can be crossed. It is of course necessary to convince both parties that they need to add to their reservoir of knowledge but continued frustration in attempting to communicate across languages without a translator should make it clear that this is the way to ensure that the best evidence is used to make decisions and that the decision makers do the best job of using it.

Cultures That Are Hospitable To Data Science

Workforce analytics are all the rage. AI and machine learning are expected to transform the way decisionsabout managing people are made. But not all cultures are open to the potential contributions data science can make. Much has been promised in terms of new tools that can better inform decisions yet the actual realization of these benefits has not yet lived up to its potential. And it is being found that organizational culture can be a potent impediment.

The obstacles that appear to slow down adoption of data science are:


  1. Data scientists often do not understand the needs of practitioners, the culture of the organization and the way in which decisions are made
  2. Practitioners do not have the background to understand how the data scientists do their work and how their findings can be applied
  3. Data scientists speak a different language, that is difficult for practitioners to understand… and they are sometimes impatient with the lack of knowledge practitioners have about the scientific method
  4. Practitioners are often suspicious of the data being used, the mysterious black box analytical tools and the scientific methodology that is being used
  5. Decision-makers are over-confident in the quality of their intuition and judgment and view it as being superior to quantitative analysis

Most professionals know that if they present evidence that contradicts what is currently being done or conflicts with the beliefs held by decision-makers there will be resistance. Cognitive bias is the villain here… we are by our nature over-confident and we welcome confirming evidence more readily. If analysis produces evidence that past decisions are flawed it puts the person that made the decision in a tough spot. Acknowledging mistakes is hard for everyone but if resources have been expended as a result of a bad decision there are more consequences than just wounded pride. And when intermediaries who use the findings of data scientists to make recommendations to executive management they must decide how much to convey and how strongly. It is hard to enter into contentious discussions when one is out-ranked and/or when the advisor knows the decision-maker holds onto beliefs tenaciously.

Cultures that require everyone to tell the truth, rather than what is politically correct, encourage advisers to present all evidence, even when it is contradictory. As long as the adviser behaves respectfully it may be permissible for the unpolished truth to be presented and for the adviser to aggressively fight for what is believed to be correct. But that is not to say it should be done in a way that publicly exposes flawed decisions without offering some face-saving options. For example, it can be pointed out that the prior decision may well have been correct, given all the facts known at that time. The discovery of new evidence can be used to change everyone’s understanding of the context within which a decision will be made. However, in some national cultures it will still be difficult or unacceptable to show that a higher ranking person’s decisions were wrong (i.e., in high power distance cultures). It may be necessary to fabricate the appearance that this is a new decision, unconnected to the prior one. Face-saving is a very important consideration in many cultures and globalization has made multinational organizations realize that it must be respected. But considering the feelings of someone about to be shown wrong is a good strategy anywhere. The Golden Rule learned by Westerners during their socialization advised them to “Treat others as you would like to be treated” but the Global Golden Rule is “Treat others as they would like to be treated.”

Unfortunately inadequate attention is paid to culture when technological advancements are considered for adoption. Hiring a data scientist into a hostile culture is a sure formula for failure. So someone should consider what has to be changed in order for the organization to reap the benefits of technology. And rarely is the job title VP of Culture found on organization charts. Changing the culture seems to be in no one’s job description, which means it is hard to recognize when culture needs to be changed and who must take the lead. HR is a function that seems to be ideally suited for this role, but the entire C-suite must be on board as well. HR must be given the mandate and must have the knowledge needed to operate on the culture.


Data science is racing ahead and offers the possibility of dramatically improving decision-making. Evidence-based management is being embraced by more organizations. A key requirement for it to succeed is not only the generation of high quality evidence but the willingness to accept and apply the evidence. An appropriate culture is to decision science utilization as pure water is to fish… necessary for survival.