10

Using Big Data to Manage Human Resources

Mark J. Neuberger

10.1 Introduction

The human resource function of any organization is especially well positioned to benefit from the business decision-making insights offered by Big Data. The ability to gain insight into the future and past behavior of employees is a key component to marshaling the power of people to better achieve desired organizational outcomes. This chapter

• Discusses the value of Big Data to enhance the human resource function of any organization;
• Reviews various case studies demonstrating how Big Data is already helping some organizations manage their human resources with greater efficiency; and
• Explains how certain legal concepts with an impact on human resource activities can create risks in connection with the use of Big Data.

Often, the human resource function in an organization is viewed as the enforcer of rules and regulations—something like an internal police force. In such organizations, the prevailing perception is that human resources are there to tell people what the rules are and to enforce them. Sometimes, human resources are seen as a proxy social worker/employee advocate. Neither of these views fairly recognizes the value an effective human resource function can deliver to the enterprise. When optimized, a human resource function has the characteristics described by Sharlyn Lauby, a human resource consultant on her Human Resource Bartender blog: “The value of human resources is that they are the great equalizer. Human Resource is the creator of balance between the interests of management and the interests of employees. When a proper balance is achieved, companies get the best performance from their employees. And employees are engaged with their work. It’s a win for everyone involved.”1 This value can be materially enhanced in any organization through the use of Big Data by enabling human resource managers to assist managers to make informed decisions regarding personnel onboarding, engagement, discipline, and termination and thereby support specific organizational goals, like increasing profitability and improving talent management and retention.

Historically, human resource managers have spent large amounts of time reporting on past events, including calculating the turnover of new hires, calculating the absenteeism rate of employees, and reporting how much money has been spent on compensation and benefits. Big Data represents a tremendous opportunity to shift the focus of human resource departments by leveraging data about employees to improve the efficiency and efficacy of an organization. The use of predictive analytics to help make better decisions regarding who to hire, who to promote, who to reassign or terminate, and how to compensate to motivate increased employee performance are just some of the benefits that can come from the use of Big Data.

As discussed in the examples that follow, often the use of Big Data will debunk conventional myths about the management of people in organizations. For example, in the quest to drive employee performance ever higher, organizations spend significant time and effort attempting to manage the balance between paying the least amount of money to achieve the greatest amount of performance. However, research studies going back decades have consistently proven that money frequently is clearly not one of the top factors motivating employees.2 In Drive,3 a 2009 book that examines multiple social experiments, the author demonstrates that intrinsic motivators like the internal desire to do good work, the ability to advance one’s career, and recognition by supervisors and peers are far greater motivators than money and benefits. Big Data represents a way to identify those intrinsic motivators within a workforce and to allow managers to better reward and motivate employees.

Many forms of analytical projects that have traditionally been performed by human resource managers can be managed using fairly basic databases. However, Big Data represents the opportunity to take data analysis to the next level; instead of reporting on what has already transpired, predictive analytics can help paint the picture of a likely future. However, those seeking to use Big Data analytics in the human resource function of their organizations must be sensitive to the likely perception that the organization is replacing sound human judgment from managers with “statistics” created by “number crunchers.” This perception should not likely prevail, however, because to optimize the value of Big Data, organizations will achieve a balance between the data scientists understanding the limits of their analysis and human resource managers understanding that the data is there to support, not supplant, the day-to-day decisions that they must make. The processes must also allow individual managers to apply their experience and judgment while debunking myths and prejudices so that decisions are made with empirical evidence informed by human judgment. When numerically based analytics are harmonized with experiential decision making, the introduction of Big Data into human resource processes will be optimized and more readily accepted.

10.2 Using Big Data to Manage People

If you are considering how to use Big Data to manage your people resources, there are already many excellent examples from leading companies that are instructive. Several summaries of how leading organizations have used Big Data in their human resource activities are presented in this section. These examples nicely illustrate opportunities that are available to most organizations.

10.2.1 Absenteeism and Scheduling

Google learned that during the annual flu season, people start searching the web for information on flu symptoms and flu remedies. Google describes the process they use as follows:

We have found a close relationship between how many people search for flu-related topics and how many people actually have flu symptoms. Of course, not every person who searches for “flu” is actually sick, but a pattern emerges when all the flu-related search queries are added together. We compared our query counts with traditional flu surveillance systems and found that many search queries tend to be popular exactly when flu season is happening. By counting how often we see these search queries, we can estimate how much flu is circulating in different countries and regions around the world.4

Google Flu Trends is not perfect; during the 2013 flu season, it overestimated the occurrence of flu in the United States by over 25%.5 However, when real-time estimates from Google Flu Trends and data from the Centers for Disease Control and Prevention (CDC) surveillance are combined, they reliably predicted the timing of the 2012–2013 influenza season up to 9 weeks in advance of its peak.6 The process will continue to be refined by Google and others. However, the potential for managers to use this type of insight can help businesses prepare for flu and other pandemics by anticipating the outbreak, allowing management to take preventive measures, including lining up contingent workers or advancing production ahead of an impending onslaught of employee absenteeism.

10.2.2 Identifying Attributes of Success for Various Roles

In its continual search to look for workers to staff its call centers, Xerox would historically focus its search for job applicants by looking for people who had previous call center experience. After running a Big Data analysis on hiring and turnover, Xerox discovered that the prior job experience of its candidates was not a strong indicator of success in the role. It learned that what made a successful call center employee was a certain personality type. Searching for those who were inquisitive and creative, the traits revealed through their data analysis, Xerox was able to reduce its call center attrition by 20%.7 Xerox further discovered that formal education was similarly irrelevant. Like most other employers, Xerox’s selection process historically focused on a candidate’s education and experience as revealed in an application form and resume. After undergoing a Big Data analysis, the focus of the selection process was changed to personality tests and data analysis.

The use of Big Data in hiring, especially in technology-related jobs, is causing more employers to look at candidates who, using conventional wisdom, they would have never otherwise considered. Increasingly, employers are looking at people who did not attend or graduate from college because their analytics revealed no benefit whatsoever to hiring people with college degrees. Alternatively, Big Data has shown employers there are factors that function as far better predictors of future job performance that allows for hiring people who do not fill the traditional mold of those with college-related coursework.8

10.2.3 Leading Change

ConAgra Foods became concerned when it realized that over 50% of its current employees would be eligible to retire in the next ten years. This triggered a need to ensure hiring people with the right skills to replace those departing workers. Knowing the extent of the organizational change that was coming, it was of primary importance to search for people who could easily adapt to job change. Using Big Data analytics, ConAgra learned that its conventional assumption that younger people were more adept at learning new tasks was false. Rather, ConAgra discovered that the ability to easily learn new tasks is basically an interpersonal skill that can be present at any age.9 By using Big Data, ConAgra was also able to predict which employees were more likely to quit. By analyzing data in the portions of the company that had significantly higher turnover, ConAgra was able to identify factors that cause people to leave. ConAgra found that lack of recognition and nonmonetary rewards drove people away more than any concerns over their pay and benefits.

The lesson to be learned from the Xerox and ConAgra Foods examples is that conventional myths regarding who best to hire and who best to invest in by way of training and promotions are being challenged by data. Businesses that can isolate what is truly relevant to ensuring better performance on the job are learning that traditional assumptions of the skills that make for an optimal new hire may not only be mistaken but also may be obstacles to hiring and retaining an engaged and more productive workforce.

10.2.4 Managing Employee Fraud

Another example of the successful application of Big Data is in the food service industry, for which analytical software is being used to cut down on employee fraud. By some estimates, employee fraud costs the US economy as much as $200 billion a year. In the restaurant industry, where profit margins are especially thin, anything that will eliminate fraud can have immediate bottom-line benefits for the business. Software programs have been developed that alert restaurants to billing irregularity trends, with one of the red flags for fraud being an unusual number of partially completed or voided checks. Additional red flags are a high occurrence of lower-priced menu items like beverages, all of which are statistical indications that an employee may be pocketing some of the money from such transactions. Tracking these trends reveals what has been commonly referred to as the “wagon wheel scam”: Waiters sell numerous sodas throughout a shift but keep the money for most of those transactions. They ring up a soda for one customer but leave the check open and temporarily transfer the charge to another patron. By studying keystrokes and trends in the checks being rung up through the system, restaurants can spot these scams and shut down the theft.10

These examples of how some companies use Big Data in the human resource function demonstrate that using Big Data may call into question conventional assumptions about how to manage people. There may be better and different ways to achieve desired organizational outcomes than those employed today.

10.3 Regulating the Use of Big Data in Human Resource Management

A number of federal laws that regulate the employment process may have an impact on the use of Big Data when applied to human resource management. Although due consideration must be given to the legal framework, none of these laws in and of themselves prevents the use of Big Data. Managers contemplating using Big Data to assist in employment decision making need to be wary of potential legal limitations on the use of the insights from their searches and structure their analysis and decisions accordingly.

10.4 Antidiscrimination under Title VII

Title VII of the Civil Rights Act of 1964 is the granddaddy of antidiscrimination legislation.11 It prohibits all forms of employment discrimination on the basis of race, color, religion, national origin, and sex. Under Title VII, two theories of discrimination have evolved through court decisions: disparate treatment and disparate or adverse impact. The first, disparate treatment, is much easier to understand. Treating someone differently because of their race, color, or national origin is illegal. We all know refusing to hire anyone simply because the person is a member of one of these protected groups is illegal. Disparate impact, on the other hand, presents a more nuanced theory of discrimination. It holds that seemingly neutral employment practices may be illegal if they have a disproportionate, or more adverse, impact on members of a protected group. This theory prohibits an employer from using a facially neutral employment practice that has an unjustified adverse impact on the members of the protected category.12 Whenever a facially neutral policy is shown to have such a disparate impact on members of a legally protected group, an employer can defend its actions by proving that the policy is reasonably and rationally related to the job for which it is being required.

A classic example in which the disparate impact theory has been used to find illegal discrimination is in height requirements. Before the passage of Title VII, it was common for police departments to impose a specific minimum height requirement as a condition for admission to the police academy. On its face, the policy treats everyone the same. However, statistically women, and perhaps members of certain minority groups, are less likely to meet the standard. They will therefore be screened out at much higher rates, or disparately impacted. That makes for a prima facie case of illegal discrimination. However, to the extent the employer can demonstrate a rational basis for why people of that height are more likely to be better-performing police officers, they can rebut the prima facie case. Most police departments that used such standards could not make such a showing, and today, such requirements are no longer used.

One area in which use of Big Data could run up against the disparate impact theory of discrimination is through increased use of tests in making employment-related decisions. Any time an employer develops a selection device like a test, the device being used should be properly assessed by professionals trained in the use of human measurement devices. To withstand the inevitable legal challenge to the use of tests, employers must be prepared to produce to either plaintiff’s counsel or a prosecuting governmental agency like the EEOC (Equal Employment Opportunity Commission) the necessary statistical analysis. At a minimum, that analysis must include a showing that the manner and method of testing in each particular organization can be statistically shown to be both reliable (meaning it tests the same thing each time it is used) and a valid predictor. Typically, the development of such statistical analysis will require engaging a professional psychometrician. As seen in some of the case studies presented, employers who use Big Data may increasingly seek to make decisions based on such factors as identifiable personality traits. This may drive employers to the use of tests and other selection devices to isolate candidates with the desired traits. Often, such tests have demonstrated disparate impact. In fact, in the Griggs case cited previously in this section (see Note 12), the Duke Power Company was found to have historically segregated Black employees into the lowest classification of jobs. After the passage of Title VII, the company eliminated overt segregation but imposed passing an IQ test as a prerequisite for moving to a higher-classified job. Statistically, Black employees in North Carolina in the 1960s did not perform as well on such tests and therefore were limited in their career progression. When they were sued, the companies could not demonstrate that increased IQ was a successful predictor of future job performance for the jobs in the power plant in question. As a result, the use of these tests was found to violate Title VII.

The lesson here is not that all tests are discriminatory. Rather, if an employer wants to use tests, the employer must do so properly (see Table 10.1). To do that, employers must conform their practices to a comprehensive set of regulations known as the Uniform Guidelines on Employee Selection Procedures.13 In 1978, four federal agencies, including the EEOC and the US Department of Labor, issued this joint regulation. The guidelines apply to tests and other selection procedures used as the basis for any type of employment decision, including hiring, promotion, demotion, retention, and compensation. The guidelines establish how an employer, using a selection device like testing, must demonstrate that (1) the test adopted is both reliable and consistent among the parties to whom it is being administered, and (2) it is, in fact, a valid predictor of the performance it intends to assess. For example, the SAT and ACT college admission tests have been consistently validated to predict future performance in one thing and one thing only: one’s performance as a first-year college student. Therefore, an employer using the SAT or ACT to hire or promote individuals within an organization, absent some additional study and analysis by trained testing experts, would be deemed an invalid use and therefore an improper defense against a showing of adverse impact.

Testing for predictive indicators can be an extremely valuable tool in selecting and retaining engaged employees. As seen from some of the examples presented, employers can perform analytical research to isolate the skills criteria that are predictive indicators for success in their particular organization. In such circumstances, they then can test for those skills, be it among candidate pools or among their current workforce. However, use of criteria that cannot be statistically validated to support the job-related criteria, and the use of tests that have not been scientifically proven to be reliable measurements of how the test measures criteria which predict job performance will likely run afoul of the Uniform Guidelines. The takeaway for employers contemplating the use of tests to help make decisions in any aspect of the employment process is that if they are going to do it, they need to do it right. That requires consulting with testing and legal experts and not pulling tests off the shelf and doing what may intuitively make sense. After all, as we have seen, so much about the use of Big Data involves debunking myths and moving beyond conventional wisdom.

10.5 The Genetic Information and Nondiscrimination Act of 2007

The Genetic Information and Nondiscrimination Act of 2007 (GINA)14 is also administered by the EEOC. Under GINA, it is illegal for employers to discriminate against either employees or applicants because of their genetic information. GINA also prohibits employers from requesting, requiring, or purchasing genetic information about their employees. Under GINA, genetic information is defined in very broad terms and includes genetic testing not only of the individual but also of their family members. This includes information about potential diseases or disorders the employee or their family members may experience. Family medical history is also included in the law’s definition of information because, historically, it has been used to determine whether an employee has an increased risk of disease, disorder, or condition in the future. GINA prohibits discrimination based on the use of genetic information in any aspect of employment and further prohibits employers from harassing or retaliating against an individual because the individual has objected to improper use of their genetic information. Thus, the accumulation of anything that constitutes genetic information to predict whether an employee may be more susceptible to disease, or future performance issues because of their genetic makeup, will run afoul of GINA.

As employers struggle to contain the cost of providing medical insurance to their employees and try to maintain the health of an aging workforce, there has been an explosion of employee wellness programs. A wellness program is defined in section 2705(j)(1)(A) of the Public Health Service Act15 as any program offered by an employer designed to promote health or prevent disease. Certain types of wellness programs offered through employment-based group health plan coverage must now meet standards under the Affordable Care Act.16 There is a veritable potpourri of workplace wellness programs that run the gamut from benefits aimed to promote health-related behaviors such as free or discounted gym memberships, diet education or smoking cessation programs, to early identification and better management of chronic diseases like diabetes or epilepsy. To be effective, wellness programs typically include data collection to preidentify employee health risks, which can then be used to craft interventions to reduce those risks.

When used in the employee wellness area, Big Data may bump up against the variety of privacy concerns and laws described elsewhere in this book as well as GINA. GINA, however, provides a safe harbor for employers: Where health or genetic services are

offered by the employer ... as part of a wellness program; the employee provides prior, knowing, voluntary, and written authorization; only the employee (or family member if the family member is receiving genetic services) and the licensed health care professional or board certified genetic counselor involved in providing such services receive individually identifiable information concerning the results of such services; and any individually identifiable genetic information provided is . . . not [to] be disclosed to the employer except in aggregate terms that do not disclose the identity of specific employees.17

Thus, like the other employment-related laws discussed thus far, GINA does not preclude employers from using Big Data to measure and assess employee health, but it is a restriction that must be carefully navigated. Carefully analyze the various “safe harbors” contained in GINA and use them. Increasingly popular employee wellness programs are a common area where employers could run afoul of GINA. However, by ensuring the employer only sees aggregated and deidentified data about the health of its employees, the employees’ rights under GINA can be preserved.

10.6 National Labor Relations Act

Contrary to conventional wisdom, the National Labor Relations Act (NLRA) does not simply apply to employers who have unions. In fact, the NLRA’s legal protection extends to all the employees of those employers covered by it, which are those private-sector employers engaged in interstate commerce, excluding railroads and airlines. The NLRA allows all employees, whether or not they are represented by a union, to engage in what is known as “protected concerted activity.” Protected concerted activity is generally defined as two or more employees taking action over some aspect of their hours worked, wages, compensation, and other terms and conditions of employment. Under the law, even a single employee may engage in activities that are deemed “concerted” under the NLRA.

Those employees who discuss their pay and benefits or complain about their working conditions will in most cases be protected by the NLRA. In addition to providing employees with certain rights, the NLRA restricts employers from engaging in certain activities that are deemed to be unfair labor practices (ULPs). Employers commit an ULP whenever they attempt to monitor employees as they engage in their protected concerted activities. This is known as unlawful surveillance.

As employees increasingly use various forms of social media to communicate and express thoughts about their jobs and their workplaces, employers have stepped up their monitoring of such activities. When employers see what they perceive to be disloyal or disparaging comments by their employees as expressed in social media, they sometimes impose job discipline measures up to and including termination.

In recent years, the National Labor Relations Board (NLRB) has aggressively expanded its enforcement activities against employers who have sought to engage in surveillance of employees’ use of social media and to otherwise quash protected concerted activity. Thus, employers who use Big Data analytics aimed at identifying employees’ thoughts and perceptions about themselves, their jobs, and their workplaces may run afoul of the NLRA.

The NLRB has gone after employers who have taken disciplinary action against an employee based on that employee’s Facebook postings critical of the employer. In a case involving a BMW auto dealer in suburban Chicago, the employer terminated one of its auto sales reps because his Facebook posting implied the employer was cheap, presumably because it served only hot dogs and chips during a new model event held at the dealership, whereas a competitor Mercedes dealer provided hors d’oeuvres served by waiters. Although ultimately the discharge of the sales rep was upheld, the NLRB found in that case, as it has in a number of others, employers who have broad policies against negative social media postings will run afoul of the act’s guarantee of all employees to engage in protected concerted activity, even where no union exists.18 Thus, any Big Data analytic that monitors employees’ use of social media, like Facebook or Twitter, should avoid analyzing comments about the workplace or terms and conditions (wages and benefits) of their employment. Because the law under the NLRA tends to be very fact specific, employers who in any way seek to survey or monitor the use of social media as part of gathering Big Data analytics about their employees or organization will be well advised to consult both the current case law under the NLRA and various case law compendiums issued by the general counsel of the NLRB.

In implementing social media policies for its employees, all employers must carefully consider the laws on employee privacy as well as the latest pronouncements from the NLRB. Nonunion employers who previously never considered NLRB ramifications need to fully understand and consider the latest pronouncements from the NLRB before disciplining or discharging any employee for alleged improper use of social media. Given the NLRB’s increased enforcement of employees’ right to engage in protected concerted activity, even nonunion employers must be prepared to defend against potential charges filed with the NLRB by aggrieved employees.

10.7 Fair Credit Reporting Act

Increasingly, employers are using background checks of candidates’ criminal records and credit histories as part of the interview and hiring process. The use of such checks is regulated by the federal Fair Credit Reporting Act and is discussed in detail in Chapter 4, “Privacy and Big Data.”

10.8 State and Local Laws

There are a multitude of state and local laws, too numerous to discuss in this chapter, that replicate those discussed previously; often, these go further than the federal standards. Managers must consider these when using Big Data analytics to manage their human resources.

10.9 Conclusion

The human resource function is one area that stands to greatly enhance the quality of business decisions through the use of Big Data. The legal framework that regulates the employment process must be considered but should not be seen as a barrier to the use of Big Data. Like many other areas of data management, the law lags the technology, which makes compliance more difficult, but not impossible. Through careful planning, Big Data analytics can take human resource management to a new capability level.

Notes

1. Sharlyn Lauby. Human Resources Adding Value to the Company. December 16, 2012. Available at http://www.HRBartender.com.

2. Edward E. Lawler III. Pay and Organizational Effectiveness: A Psychological View. McGraw-Hill, New York, 1971.

3. Daniel R. Pink. Drive. Riverhead, New York, 2009.

4. Flu Trends: How Does It Work? 2011. Available at http://www.google.org/flutrends/about/how.html.

5. Declan Butler. When Google Got Flu Wrong. Nature, February 13, 2013. Available at http://www.nature.com/news/when-google-got-flu-wrong-1.12413.

6. New Methods for Real-Time Influenza Forecasting Appeared Effective. Infectious Disease News, December 20, 2013. Available at http://www.healio.com/infectious-disease/influenza/news/online/%7B9183c557-f289-496c-8a5e-f19d2314b77a%7D/new-methods-for-real-time-influenza-forecasting-appeared-effective.

7. Joseph Walker. Meet the New Boss Big Data. Wall Street Journal, September 20, 2012. Available at http://online.wsj.com/news/articles/SB10000872396390443890304578006252019616768 (subscription required).

8. Don Peck. They’re Watching You at Work. The Atlantic, December 2013.

9. Rachel King. Data Helps Firms Find the Right Workers for the Right Jobs. Wall Street Journal, September 15, 2013. Available at http://online.wsj.com/news/articles/SB10001424127887323906804579036573250528100 (subscription required).

10. Lamar Pierce, Daniel C. Snow, and Andrew McAfee. Cleaning House; The Impact of Information Technology Monitoring on Employee Theft and Productivity. MIT Sloan Research Paper No. 5029-13. November 12, 2013. Available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2318592.

11. 42 U.S.C. 2000(e) et seq.

12. Griggs v. Duke Power Company, 401 U.S. 424 (1971).

13. 29 C.F.R. Part 1607.

14. 29 U.S.C. § 2000ff et seq.

15. 42 U.S.C. 2705(j)(1)(A).

16. 42 U.S.C. 300gg-4(a), (j), and (m).

17. 42 U.S.C. 2000ff–1(b)(2).

18. Karl Knauz Motors, Inc., 358 NLRB No. 164 (Sept. 28, 2012).