Predicting Analytics
Amit B. Mohindra and J. Allan Brown
INTRODUCTION
THE IMPORTANCE OF MEASUREMENT and analytics in the practice of human resources has grown apace since Jac Fitz-enz’s pioneering work in the 1970s. Further contributions have been made over the years by academics, consultants, and HR practitioners. The literature now includes numerous articles that recount how companies are beginning to reap a competitive advantage through the adroit use of human resources data, analysis, and insight. There is also a handful of recent books that serve to provide a framework for the application of human capital analytics.
Human capital analytics has evolved with the growth in sophistication and importance of the HR function.
Through the 1980s, HR was largely an administrative function (called personnel), and measurements were transaction-based and largely unremarkable—for example, how many people were hired last month or how many performance review forms have been submitted. These were descriptive measures and primarily backward looking. Since HR was not considered a strategic function at the time, there was no urge to project measures forward or to consider alternative scenarios. The focus—as Fitz-enz often recounts—was on building a conceptual measurement framework and getting HR itself to take metrics seriously.
As business conditions changed in the 1990s and resources were stretched in economic downturns, HR turned inward to focus on measuring its cost and efficiency and comparing these with industry benchmarks to which it aspired. As companies competed in the war for talent, the complexity and vitality of human resources—by now called human capital—became apparent. Thinking of the workforce as an asset rather than a cost implied the connection between human capital activities and business performance. It became incumbent upon HR to characterize and understand this asset using approaches established to manage other corporate assets. HR co-opted accounting terms and frameworks to ascribe a return on investment in HR programs and in the HR function itself.
The 2000s saw the emergence of human capital analytics, defined more specifically as the use of data, analysis, and systematic reasoning to make human capital decisions. HR data were more readily available with the emergence of ERPs and the associated move toward data warehouses; business intelligence software could as easily be applied to HR data; and business conditions required further optimization of companies’ investment in human capital. According to one head of HR at the time, “Analytics is the next evolution. If HR wants a seat at the table, analytics is going to become table stakes. If you can’t talk in your business leaders’ language, you won’t be invited into the conversation.”1
The publication of the first “handbook” of HR analytics2 signaled the arrival of analytics on the broader HR stage. All of a sudden conferences devoted entirely to human capital analytics were organized, and companies began to establish analytics teams and departments within their HR functions. It is now rare to encounter an HR department that does not apply at least some sort of quantitative approach to its operations or to meet an HR practitioner who is not intrigued by the latest application of human capital analytics. The quantitative and visual output of current HR analytics teams is a far cry from the transactional HR reports of a couple of decades ago.
By all accounts, the next big thing is predictive human capital analytics.
Human capital analytics has progressed in sophistication. Evidence is increasingly required to support decisions about the attraction, motivation, and retention of talent. In leading analytical companies, data have become invaluable, disparate claims are framed as hypotheses, models are built to test the hypotheses, and statistics are used unabashedly. Analytics allows all factors—employee attributes, operational parameters, and business results—to be examined and correlated. Data from employee engagement surveys provided the missing link of employee mindset and attitudes, enabling connections to be made between HR investment, employee perceptions, and business outcomes.
In the last few years, there has been an interest and growing demand for “predictive” human capital analytics along the same lines as predictive analytics conducted elsewhere in the business. For example, the marketing department constructs models to forecast revenue in different scenarios, based on integrated data on customers and their behavior, external economic indicators, and actual purchasing outcomes. The increasing amounts of data on consumer behavior available via web-based transactions and social media activities allow for richer predictive models. Business leaders expect these sorts of models from their organizations, are increasingly familiar with the models’ premises, and employ such models in strategic decision making.
It does not take a huge leap of imagination on the part of HR to wonder about the possibility of playing in this arena. Jac Fitz-enz’s most recent book challenges HR to do just this.3 Presentations and comments at HR conferences as well as recent human capital measurement and business analytics group discussions on LinkedIn all hint at this new frontier for HR. Indeed, once companies have a handle on their employee data and connect it to other enterprise information and business performance, the next logical step is to use the implicit models to make predictions about human capital as well as business outcomes. The ability to look ahead and “place bets” on human capital investment and practices with some degree of confidence in their effect on business outcomes is a powerful and attractive proposition. More and more companies will attempt to harness this capability as they progress along the human capital analytics continuum (Figure 1).
FIGURE 1. THE HUMAN CAPITAL ANALYTICS CONTINUUM.
PREDICTIVE ANALYTICS
Wikipedia defines predictive analytics very broadly as an endeavor “encompassing a variety of statistical techniques from modeling, machine learning, data mining, and game theory that analyze current and historical facts to make predictions about future events.”
Make predictions about future events! It’s no wonder then that HR professionals, being mortal, are interested in predictive analytics. Instead of just reporting what has happened to employee longevity, engagement, performance, turnover, or any other human capital variable, you can now predict what will happen next. What is often lost in the fervor, however, is how predictive analytics is actually constructed—using statistics, data mining, and game theory. It is worth summarizing these three key approaches to predictive analytics.
Statistics
Statistics is a branch of mathematics that is concerned with the collection and study of data. One aspect of statistics that is well known within HR is descriptive statistics—measures that describe a set of data. These include measures of central tendency (for example, average and median) and dispersion (for example, variance and standard deviation). Predictive analytics necessitates the use of another aspect of statistics—inferential statistics. Inferential statistics allows for modeling the randomness or uncertainty underlying a set of data with a view to drawing inferences—that is, predictions. Statistical models, therefore, are at the heart of predictive analytics.
Regression Models. Statistical models include simple correlations and regression analyses of various types and a wide variety of other techniques. Most HR practitioners are familiar with correlations and linear regression involving two variables. Relationships between two variables are convenient to depict via a scatter plot and to model via a regression line. It is easy to represent the correlation—either visually or in terms of the regression’s R-squared (also known as the coefficient of determination) value. The following figures, 3 and 4, represent a statistical model that tries to capture the relationship between years of experience and compensation. The first model is linear; the second model is nonlinear—it attempts to capture the slowdown in compensation with increasing years of experience. The equations are used to predict compensation, given years of experience.
FIGURE 2. A STATISTICAL MODEL THAT TRIES TO CAPTURE THE RELATIONSHIP BETWEEN YEARS OF EXPERIENCE AND COMPENSATION—LINEAR.
FIGURE 3. A STATISTICAL MODEL THAT TRIES TO CAPTURE THE RELATIONSHIP BETWEEN YEARS OF EXPERIENCE AND COMPENSATION—NONLINEAR.
However, the real world involves multiple variables (compensation is affected by many things, not just years of experience) and complex, nonlinear relationships. These are best modeled using multiple regression. However, multiple regression, especially nonlinear multiple regression, is not widely used in HR. One reason is that interpreting and explaining the outputs and statistical significance of the results often presents a challenge, since there is no visual analog of a scatter plot or regression line.
Logistic Regression. A very special type of multiple regression—logistic regression—is seldom used in HR even though it is arguably one of the most powerful weapons in the predictive human capital analytics armory. Logistic regression allows for a dichotomous dependent variable (i.e., a variable that has only two possible values), and its results are easily characterized as probabilities or odds that allow for easy and direct interpretations in a predictive model.
HR-related variables are often dichotomous or categorical—i.e., having discrete rather than continuous values. People either participate in a program or not; they either receive a salary increase or not; they are either male or female; etc. They can be characterized as one of five race/ethnicity categories; as having four levels of education; as being rated high, solid, low or nonperformer; etc. Logistic regression can combine dichotomous, categorical, and continuous variables (e.g., tenure in the company) into powerful models.
Logistic regression analyses have been used to predict turnover and success. These traditional models’ predictive power can be enhanced by incorporating data from engagement surveys and learning program participation, among other things. With adequate individual profile information, it is possible to predict the nature of a candidate’s fit with the department or company and the quality and effectiveness of teams in which employees participate.
Consider a logistic regression attempting to explain what drives turnover. Avoiding the complexities of modeling for the moment, let’s take a look at the output of such a model. In Figure 4, the regression coefficients are reported as likelihood ratios—the impact of each variable (in isolation, controlling for all other variables in the model) on turnover in terms of the contribution to the odds of turnover. Only a few variables are shown; the model can be enriched by including geography and business unit to capture variations across the company. The interpretation of each likelihood ratio is described. Note that the model can include multiple years of data; the more data, the more robust the model.
FIGURE 4. OUTPUT OF A LOGISTIC REGRESSION ATTEMPTING TO EXPLAIN WHAT DRIVES TURNOVER.
Survival Analysis. Turnover gets much attention in HR, yet few companies employ survival analysis to model turnover. Survival analysis is used in a number of fields. Engineers refer to it as reliability analysis (where they would predict, for example, when a component would likely fail), and economists refer to it as duration analysis (where they would predict, for example, the duration of unemployment spells). It can be used in HR to predict tenure in a job or longevity with a company. Data needs for survival models are very modest; all you need is a start date (for everyone) and an end date (for those who have terminated), and the output is the probability that someone will terminate at any point in time. You can add covariates—i.e., variables that you think impact a person’s decision to resign—and build a more complex model that conditions the probability on these factors.
The mathematics underlying the model is quite complicated, but most statistical packages produce tabular and graphical output for survival models. Another variant of the model allows for “tenure tables” along the lines of demographers’ life tables that insurers use for pricing life insurance. The key point is that the probability of an employee exiting the organization at time X is conditional on the probability that they have not terminated until time X—just as the probability that a person will die at age Y is conditional on the probability that they have survived until age Y.
FIGURE 5. WHAT CAUSES THE SPIKE IN PROBABILITY OF EXIT FOR WOMEN AFTER THREE YEARS OF SERVICE?
Markov Models. HR practitioners often have to determine the strength of the leadership bench or the overall size and shape of a “pyramidal” organization. These kinds of population dynamics are best analyzed using what are known as Markov models. For example, Ward et al. include an application of Markov analysis in forecasting human resource requirements.4 In general, Markov analysis is associated with situations in which there are a number of “states,” there are known probabilities of moving from one “state” to another, and you are interested in what happens over time.
Think of the states as levels within an organization. In financial services organizations, levels (from junior to senior) could be analyst, associate, vice president, senior vice president, and managing director. Movement into states, between states, and out of states represents hiring, transfers, or promotions and terminations. You can then apply Markov analysis to predict the shape and size of the organization over time. You can plug in historical rates of hiring, transfers, promotion, and termination and then model what happens to the people in the organization as they move into, up, or out of the career ladder.
Markov models have been used for workforce planning since the days of “manpower planning” and may come back in vogue. Again, the underlying mathematics is complicated, the model requires certain restrictive assumptions in order for the math to work, and you need reliable historical data. An example of the successful use of a Markov model in a commercial setting occurred in 1999 when one of the authors employed a Markov model to adjust the size and shape of a global financial services firm’s investment banking division. The model enabled the division to control headcount moves in anticipation of changes in the underwriting, securities, and mergers and acquisitions (M&A) market and competitors’ hiring strategies.
Statistical models, useful though they may be in making predictions, are demanding tools. They need to be understood thoroughly in order to be applied appropriately. Their results need to be considered very carefully, and one must not fall into the trap of assuming that there is a causal relationship when there is a correlation among variables. The results also need to be qualified in terms of whether the model explains the data—i.e., whether there is a good fit between what the model predicts and actual outcomes—and whether the results are statistically significant. Last but not least, you need rich data to compute complex models. These data have to be relevant, accurate, and complete. They can be compiled manually from forms or extracted from the HRIS system, or even better, from an enterprise data warehouse that contains HR, financial, and operational data that have been “scrubbed” for use in developing metrics and analytics.
Data Mining
Data mining is considered by some as an alternative to statistical modeling for the purposes of prediction. It refers to the process of looking at vast amounts of data and identifying strong patterns. These patterns can then be used to make predictions. Economists are typically disdainful of data mining; they prefer to construct a model, test its validity using real-world data, and then either refine the model or discard it and start anew. The better the model and the better the data, the better the predictive power. Data miners trawl through real-world data, looking at variables in all sorts of combinations to extract a reliable relationship that is strong enough to warrant a prediction. Data mining has acquired a wide following through its discoveries using sophisticated statistical software and large quantities of data—e.g., customer behavior based on web activity. The important thing is to establish a strong enough relationship among variables that forms the basis for predicting a value for one variable based on the value(s) of the other variable(s). There are a wide variety of data mining techniques, each of which is suited to a particular situation or need.
Game Theory
Game theory is the science of strategy and, generally speaking, deals with “strategic” situations that are characterized by five elements:5
1. Strategy becomes necessary when your actions affect others’ well-being and their actions affect your well-being. If there were only one firm in the world, it would not need to engage in strategy. The first element of strategy is the notion of players—people or entities who interact and whose actions have consequences for each other.
2. Strategy becomes relevant when there is competition among the players. The players are rivals in some sense or another. Players are trying to outdo their adversary, who is trying to do the same to them. It is important to keep in mind that strategy allows room for cooperation and collaboration as well.
3. Strategy becomes interesting when there are choices and those choices involve trade-offs. It’s worth remembering that economics is all about choice—allocating scarce resources among competing ends.
4. Strategy becomes complex when there is uncertainty. Players don’t know what the other player or players want or what they will do. There may be uncertainty about the state of nature—will it rain today? Will interest rates be high? Will unemployment be low?
5. Connected to the element of uncertainty is the notion of a future, or in other words, a time horizon or time frame. Your actions will likely be different if you are going to be in business for just one fiscal year or in perpetuity.
What does all this have to do with predictive analytics? When you need to model the behavior of an individual (say, an employee in the situation of a performance management discussion with a manager) or a group of individuals (say, workers intent on unionizing) to predict how they will behave in response to a situation or an offer, it is often useful to turn to game theory. When the situation is properly characterized in terms of all the elements at play, game theory provides a range of possible solutions. Some of the solutions are win-win and some are not. By addressing the right items, it is possible to move toward more preferable solutions.
PROMISE AND PERILS OF PREDICTIVE HUMAN CAPITAL ANALYTICS
Predictive human capital analytics fits right into the HR function’s need to be strategic and to be a recognized business partner. Successfully deploying predictive human capital analytics arms HR with confidence and credibility in a number of ways. HR models can incorporate financial, operational, and external market data, allowing HR to offer a more integrated view of the enterprise than other functions (arguably, finance could do the same, but let’s assume for now that HR has a competitive advantage with respect to understanding HR data). With the help of predictive analytics, HR can make more than just educated guesses about a host of business outcomes. Assuming a reasonable understanding of business and industry conditions, HR can comfortably participate in strategy and planning.
Becoming familiar with the methods and application of predictive analytics requires a way of thinking that lends itself to a strategic, enterprise-wide view of the world. You need to understand the data that you are using, particularly the data that reside outside HR’s purview in the finance, strategy, and operations departments. Detailed knowledge of this information and how it ties together will arm HR leaders with the confidence and credibility to go “toe-to-toe” with other leaders during business discussions. On the flip side, understanding the limitations of the data (since you know how they are collected, cleaned, and maintained) will give you a balanced perspective to differentiate between bona fide arguments, educated guesses, wild gambles, dissembling, and blatant falsehoods. Predictive analytics models require a “systems” way of thinking: What are all the factors involved? What are all the connections both internal and external? What are the feedback mechanisms at play? What are the current and future impacts? And so forth.
HR owns a rich data set—information on applicants, new hires, employees, and ex-employees. The information includes personal characteristics, employment history, pay and performance histories, promotion and other organizational move histories, learning interventions and outcomes, engagement scores and histories, and the like. Armed with this information and the appropriate statistical models, HR can get a complete grasp of what works and what doesn’t with respect to recruitment, motivation, compensation, engagement, and retention by examining cross-correlations (i.e., considering all factors at once, not just a single variable) and dynamics over time.
It would be possible to make predictions on any or all of the following. Note that the predictions are stated in terms of likelihoods (i.e., probabilities).
Likelihood of an offer being accepted
Likelihood of a candidate’s success in a division or in the company
Likelihood of an HR program’s success
Likelihood of a new hire’s longevity with the company
Likelihood of an employee voluntarily terminating at any point of time in the future
Predictive analytics is a powerful tool, but one should be careful to not get carried away by the hype. Virtually all descriptions of predictive analytics have dwelled only on the positive—i.e., that you act on the output of some predictive analytics work, the business wins, and HR’s stock rises. People seem to forget that there is a possibility of making an incorrect prediction. What would happen to the unfortunate analytically armed, newly strategic, soothsaying HR leader whose brilliant models just happen to spit out three dud predictions in a row? This possibility is all too real even with the best minds and models at work.
The emergent predictive human capital analytics movement can sometimes seem a little disparaging of traditional HR approaches and practices. Predictive analytics is viewed as the vanguard of a new kind of HR—smarter, more sophisticated, scientific, and way cooler than their non-numerate and less analytically inclined colleagues. What is often forgotten is that a great deal of judgment is necessary to formulate useful models and interpret the results. One CEO cautioned his analytically eager head of HR that if you take the judgment out of human capital decision making, HR is scarcely distinguishable from the finance department.
A case in point is the recent financial crisis, whose roots can arguably be traced to the creation and indiscriminate spread of complex, derivatives-based financial instruments that seemed so brilliant and cutting-edge that people no longer questioned their validity. The so-called “quants” gained ascendancy, but their models were unable to account for human behavior. In a Harvard Business Review article entitled “The Big Idea: The Judgment Deficit,” Amar Bhidé concludes that statistical models “reveal broad tendencies and recurring patterns, but in a dynamic society shot through with willful and imaginative people making conscious choices, they cannot make reliable predictions.”6 He makes a number of arguments for caution in predicting human behavior based on statistical patterns without complementary case-by-case judgment.
When intelligently formulated, predictive human capital analytics can be very persuasive. However, not everyone is wired the same way to respond to predictive models. Ultimately, the executive decision is predicated on confidence in a stated probability. People have different degrees of comfort with uncertainty and risk, different appreciations for probabilities, varying levels of understanding vis-à-vis the power of test statistics. The decision maker could disagree that a 60 percent probability is good enough or that a 95 percent confidence limit is powerful enough. The characterization as a probability may actually defeat the whole argument if the decision maker has a strong hunch about an alternative. It is not always prudent to discount the subject matter expert’s or executive decision maker’s gut instincts if they do not align with a model’s prediction.
Executives don’t always have the patience to buy into a solution that doesn’t fit onto a PowerPoint slide. If executives like the answer and the analysis is supportive, chances are they will move forward with the recommendation and thank the team for a job well done. But if the recommendation is counterintuitive or countercultural, the data need to be combined with judgments about the non-quantitative aspects (like desired behaviors or values) and wrapped into a solution that suggests that the leap of faith is worth taking. When asking a CEO to make a decision “against her better judgment,” HR needs data to support the path and the likely result.
On balance, the promise of predictive human capital analytics is compelling, and HR ought to harness its power. Companies that position themselves to take advantage of predictive human capital analytics will have a head start in terms of HR having a direct impact on business strategy and decision making. They will also inculcate a broad and deep culture of analysis within HR and beyond. In the best of all possible worlds, assertions will require evidence; alternative strategies will be evaluated based on careful analyses; and decisions will be made based on all available information. Decision makers will understand the likelihood of success and be able to make strategic plans and decisions. Of course, predictive analytics offers the possibility of all these things, but things such as leadership and culture will influence its actual impact.
BARRIERS THAT HAVE HINDERED HR’S ANALYTICS INITIATIVES
In the enthusiasm around analytics that pervades the HR function today, the question of why HR arrived so late to the game is seldom asked. After all, in a very general sense, it has always been recognized that “people are a company’s most important asset” and often its largest expense; that employee-related data or at least the means to collect and organize such data have typically been in HR’s hands; that quantitative analysis has been applied to business by management consultants since World War II, and the popularity of business analytics has grown exponentially with the emergence of the Internet in the early 1990s; and that the basic statistical tools and models applicable to human capital analytics have been available for decades.
However, if we press on without some appreciation of why the adoption of analytics has taken so long in HR, we risk losing the opportunity to learn from our mistakes. Indeed, learning some of the lessons may speed up the adoption of predictive human capital analytics.
Despite how wonderful the techniques and models of predictive analytics may be and how powerful their influence might seem, they are neither new nor especially remarkable from a technical standpoint. HR’s delay in adopting predictive analytics mirrors in some ways its delay in adopting analytics in the first place—HR capability, HR’s credibility with the business, and data integrity. In particular, successful development and deployment of predictive analytics needs formidable data requirements, deep training in statistics and preferably in econometrics, and the judgment to apply the right model to a situation. Perhaps most importantly, it also needs the ability to succinctly and accurately describe a model’s prediction with all the requisite qualifications.
The fundamental hurdle in the adoption of analytics by HR has been its capability with respect to analysis and modeling.7 Other than compensation professionals, traditional HR roles have not required these skills and as a result have not attracted individuals who are numerate and analytical. Said one HR professional: “I didn’t join HR to become a spreadsheet jockey.” Even master’s level programs in HR have fairly modest quantitative course requirements, and from a cultural standpoint within HR, there is an aversion to getting too much into analytic detail. When was the last time you attended an HR conference presentation or webinar where the presenter voluntarily delved into some mathematics or the audience asked him or her to talk a little about the technical underpinnings of the presentation? HR audiences typically enjoy hearing about survey results and then begin relating their own anecdotal evidence. Seldom does anyone question the survey methodology and the potential consequences of small samples, sample selection bias, and measurement rigor.
As a result of this capability gap and the resultant lack of quantitative analysis or argument from HR, the field has had to wage an uphill struggle to be taken seriously when it has finally produced metrics and analytics. HR’s credibility in this area is not helped by the lack of HR equivalents to standards like generally accepted accounting principles (GAAP) around data and metrics. The Society for Human Resource Management (SHRM) is currently addressing this gap in conjunction with the American National Standards Institute (ANSI). The standards they develop can be used by companies in sharpening their HR metrics and analytics and by investors examining the quality of human capital management in the company.
By far the most challenging impediments have been the availability and quality of data. Absent a demand for data to feed metrics and analytics, there was no urgent need to ensure that HR data were captured universally, accurately, and systematically. In the past, HR data were stored in different locations and systems that didn’t necessarily talk to each other, immensely complicating the task of creating a comprehensive, live employee database. Even when databases were available, there were issues with the quality and reliability of the data. Data privacy regulations that limited the storage and spread of data across geographies created additional challenges. As a result, initial steps toward creating metrics and performing analytics had to be careful and tentative at best. Anytime holes were poked in an argument due to poor data, HR took two steps back in terms of credibility.
Even when HR has developed analytics that are inserted into reports and dashboards, a frequent complaint is that the information is not always actionable. The charts and numbers look lovely and are sometimes even compelling, but unless there is a clear and visible cycle of discernment, action, and results, the HR data ends up being window dressing to the “hard” numbers on revenue growth, customer satisfaction, market share, and the like. The need to generalize the information for the top-line view that dashboards require takes the focus away from identifying issues within segments of the company or certain topical areas. Drilling down from the top-line view uncovers messy details that require a lot more attention than the nice round macro numbers on top.
As a result of one or more of these factors, HR’s ability to capitalize on analytics is either stymied at the outset or stalls earlier in the evolutionary path than desired. It is important to note that the requirements for successful predictive analytics are even more stringent across all the factors. Data needs are greater, more sophisticated software is necessary, and specialized skills are required to build predictive models.
Typically, predictive analytical models require rich data sets. This entails large volumes of data—more data on more individuals over multiple time periods. More data on individuals, including performance histories, learning interventions, and career progression allows for more complex modeling that incorporates all the relevant information for the issue being studied. The quality of the data is important, too. There can’t be too many systematic gaps in the data since those observations will have to be thrown out, resulting in skewed estimates. The large number of observations also helps to improve the statistical results in the sense that tests of significance have more “power.” The ability to track cohorts of employees over time allows for the creation of so-called “panel” data, which facilitates dynamic modeling, that is, modeling changes over time. Of course, the nature of the “sample” (it’s not really a sample in the true sense, since there is no random selection; it’s typically the employee population or a defined subset of it) is important to understand so that appropriate adjustments can be made to the analyses and results can be interpreted properly.
Constructing predictive analytical models requires specialized training and skills. Prerequisites are an interest in and mastery of basic statistics at the level of a junior- or senior-level undergraduate course. But statistical training and propensity are not enough. Some training in econometrics, biometrics, or psychometrics is useful for an effective human capital analytics shop that builds predictive models. Econometrics is the field of study that applies quantitative and statistical models to economics and other fields (econometricians are known as biometricians in the health sciences area, for example). Econometricians are able to build models and test them using real-world data (known as the parametric approach, since the intent is to estimate the value of the model’s parameters) and also to examine data to unearth patterns and relationships (known as the non-parametric approach, where the data is unencumbered by structural constructs). Econometrics is taught at the undergraduate level with a handful of courses, depending on the size of the institution and its economics department. One or two years of graduate training in economics provides the necessary econometric skills and insight to comfortably apply the skills to a corporate setting.
Microsoft Excel has been the weapon of choice for most analytical work, and successive versions have incorporated more and more useful tools. However, performing predictive analytics requires more sophisticated software that manages data, runs predictive models, and outputs customizable charts for import into presentations. There are a number of packages available, but the best are SAS, Stata, SPSS, and S-Plus. An open source alternative called “R” is gaining traction as a result of its cost-effectiveness and charting flexibility. SAS has been used by corporations for some time, and it’s likely that companies already have a license that can then be extended to the HR or analytics team. Stata is very user-friendly, but it is not geared toward a corporate environment. It is most popular in health sciences and economics research. SPSS is a recent entrant into business analytics. S-Plus is very powerful, but it is also relatively more complicated to use.
There are ways to overcome these barriers, as demonstrated by some leading-edge companies. The challenges suggest that the speed with which HR adapts analytics and moves into the predictive phase will depend on how well HR organizations can do the following:
Shift the mind-set with talent from outside HR. When two engineers educated in operations research and management science interviewed for intern positions with Google’s People Operations group, several HR professionals suggested that their skills were better suited to marketing or web analytics and wondered what they would do for HR. By the end of the summer, the application of their skills was clear. They weren’t doing deep analytics yet, but with a ready supply of work that went beyond data compilation the stage was set for building on the capability and executive buy-in soon followed.
Start a data quality program. Delving into analytics with a focus on data integrity may seem like a foregone conclusion, but it takes only a few mistakes in this area to set back credibility or give the impression of a false start. It is surprising how many HR organizations don’t have a data warehouse or “single version of the truth.”
Build on the analytics capability over time. The promise of analytics and Google-like success with talent management might lead some organizations to make the leap to predictive models. We shouldn’t discourage that work, but time is often needed to bring together data that works into the right combination of analyses, some of which have already been bought into. Too much complexity at once, even if done accurately, may not be bought into by senior management. More often these approaches are viewed as an attempt to “boil the ocean.”
Allow for missteps within the analytics function. Let them make mistakes. Nearly every analytics exercise is interesting; not every model will be useful or relevant. Choose carefully the work you put forward to ensure it has an actionable outcome or suggested path.
Prioritize the work toward critical pain points relevant to the business and define the questions you hope to answer. Start with a thrust in one or two areas to build the capability and organizational learning. If, for example, growth is a strategic priority, then the appetite for staffing analytics may be high.
There has been some interesting work done in predictive human capital analytics. Google in particular has been at the forefront recently and has a number of advantages in its favor, including management appetite and expectation, thoughtful HR leadership, availability of data, and skilled resources to conduct research and develop predictive analytics. Other companies, such as Morgan Stanley, have recently established human capital reporting and analytics functions. Still other companies dabble with predictive human capital analytics where there is a happy convergence of demand for such work and supply of internal talent and resources. In some cases external consultants fill the void as well as develop off-the-shelf approaches and analytics, such as Mercer’s internal labor market and productivity suites.
PREDICTIONS ON PREDICTIVE HUMAN CAPITAL ANALYTICS
Given what we now know about predictive human capital analytics—what it is, what its strengths and weaknesses are, and what resources are needed to develop and use them—we can turn our attention to thoughts about how its advent might play out in the next few years. At the current time, it is reasonable to say that there are only a few companies that employ predictive human capital analytics, the applications are limited, and the modeling expertise is in relatively short supply.
A few companies will remain in the vanguard (Google comes to mind immediately). These will be larger companies with diverse business footprints—both geographic and product. The argument for investment in predictive human capital analytics is that understanding the relationships between various inputs, assets, and outputs allows them to make optimal decisions in each segment as well as reap advantages through scale. Investment is possible because the HR leadership team is business-oriented, numerate, and willing to hire nontraditional HR staff that have analytical backgrounds. Business leadership has high expectations of its ROI on human capital investment and is willing to let HR “experiment” and work toward optimality.
It is safe to say that predictive human capital analytics will be used more and more. On the one hand, more companies will begin to experiment with and ultimately use these models, and those companies that have already adopted them will begin to apply them to new situations. Consultants will emerge in the marketplace who specialize in developing customized predictive human capital analytics. Some products will become standardized, and there will be some benchmarks created. Companies will hire consultants or develop in-house capabilities. The in-house capabilities may either grow organically as demand for analytics is met through hiring specialist staff or may be purpose-built and staffed as a centralized unit.
Some companies will focus on so-called macro-analytics that attempt to model the entire enterprise. These will be akin to the computable general equilibrium (CGE) models so popular among economists in the latter part of the last century. The data, technology, and capability exist, so why not try and model the entire workforce or, for that matter, the whole company? These, like the CGEs of yore, will likely fall apart under their own weight. The real treasure will be in the micro-analytics that help a company understand its myriad internal dynamics, incorporating people, operations, and financial perspectives. Analytics will best be applied on almost a project basis to “get under the hood” of an issue, process, or outcome.
Specialists in predictive human capital analytics will begin to emerge as a formal job description. Initially, there will be a bimodal distribution of talent. On the one hand, there will be the technically adept analysts who can build the models and run the numbers. These individuals will initially come from graduate programs in economics and industrial psychology with formal training in statistics and econometrics as well as consulting companies. On the other hand, there will be more experienced, more senior managers and executives who understand the power of analytics and which situations are amenable to modeling. These more senior people understand the business context for analytics and the specific needs vis-à-vis predictive human capital analytics. Over time, the technical specialists will get some practical experience under their belt and through their experiences will develop the judgment necessary to properly apply their models. They will become a new generation of highly quantitative HR leaders with an innate strategic instinct. The seat at the table will have been warmed up for them and they will take the HR function to new heights.
CONCLUSION
Human capital analytics have come a long way. Their sophistication has grown and their rate of adoption has accelerated in the last few years. As more companies adopt analytics and more success stories are publicized, the interest and investment grow will grow further. Many companies are now engaged in predictive human capital analytics. These statistical models allow users to make inferences about future outcomes and therefore participate more confidently in business strategy, a realm that HR has always aspired to.
As companies develop human capital centers of excellence (COEs), they should be mindful not to wring all the analytical work from the rest of the company, nonintuitive though this might appear. The centralized resources are a good way to attract and develop talent, create quality standards, and come up with solutions across the enterprise. However, in the long term, companies can truly harness the power of analytics via its dispersion throughout the company. What is to be avoided is a tendency toward central planning and all data flowing to the central body that makes decisions and issues orders for execution. Just like the Soviet planning committees, centralized organizations cannot know everything and take everything into account. Decisions that drive the business are made every day by employees and managers, and it is as important for them to see the analytics so that the right decisions can be made.
While the possibilities are large and exciting, some words of caution are in order. First, predictive analytics can do only so much. It’s sobering to keep in mind that no one really understands the macroeconomy or the market, even though the best and brightest minds in the world have been designing and applying predictive analytics to make an academic reputation or a business fortune. Second, no matter how good a model is, it is still a representation of reality built upon assumptions and past facts. Unpredictable things can happen and models may never be able to account for these events. Third, the importance of the judgment of seasoned HR professionals cannot be underestimated.
Note: The views expressed in this article are the authors’ personal opinions. The authors are grateful for fruitful discussions and insightful comments from Jamie Herslow and Lawrence Gallant.
Notes
1. Bill Roberts, “Analyze This,” HR Magazine, October 2009.
2. Laurie Bassi, HR Analytics Handbook, Reed Business, 2010.
3. Jac Fitz-enz, The New HR Analytics, (New York: American Management Association, 2010).
4. Dan Ward, Thomas P. Bechet, and Robert Tripp (editors), HR Forecasting and Modeling. The Human Resources Planning Society, 1994.
5. Amit B. Mohindra, “Closing the HR Capability Gap,” HR Executive Magazine, September 2010.
6. Amar Bhide, “The Big Idea: The Judgment Deficit,” Harvard Business Review, September 2010.
7. Mohindra, “Closing the HR Capability Gap.” Thomas H. Davenport, Jeanne Harris, and Jeremy Shapiro, “Competing on Talent Analytics,” Harvard Business Review, September 2010. Alexis A. Fink, “New Trends in Human Capital Research and Analytics,” People & Strategy, 33, no. 2 (2010).
Amit B. Mohindra is the founder and managing director of Nelson Touch Consulting, LLC. The firm was established in 2008 and advises clients on HR strategy, incentives, and analytics. In 2011, Amit served as research director at the Institute for Corporate Productivity (i4cp). He has worked in economics research, compensation consulting, and corporate human resources for organizations that set the agenda in their industry: the World Bank, Towers Perrin (now Towers Watson), Lucent Technologies (now Alcatel-Lucent), IBM, and Goldman Sachs.
Amit received a master’s degree in economics from Brown University, where he specialized in labor and development economics, and undergraduate degrees in economics and business (AB) and electrical engineering (BS) from Lafayette College, where he was a McKelvy Scholar and a member of the varsity crew team. He has written articles on HR capability, workforce analytics, and workforce planning in publications such as HR Executive Magazine, Talent Management, Training & Development, and Chief Learning Officer.
Jeffrey Allan Brown is vice president of human resources for Marvell Semiconductor. Allan joined Marvell in May 2008 as the director of compensation and benefits and assumed leadership of worldwide human resources in January 2012. Prior to joining Marvell, Allan was Google’s first director of recognition and HR systems, where he led the development of reward practices and productivity systems that supported the culture and helped grow Google from 3,000 to 22,000 employees during his tenure. Allan’s experience also includes leading the HR analytics team at Microsoft, compensation consulting with Towers Perrin and Mercer, and financial analysis with Wells Fargo Bank. He holds an MBA. from Rice University and an AB in economics from the University of California, Berkeley.