Turnover is costly. Replacing an employee typically costs 50% to 100% of a year’s salary for that position. So employers are trying new practices that might reduce these turnover costs, using complex computer-based algorithms to identify the job candidates most likely to be loyal to the organization. Wall Street has recently pushed this trend, raising questions about how useful computers actually are in finding promising job candidates.
Hiring practices based on algorithms are not new. For decades, organizations have relied on candidate personality and aptitude testing to narrow down the interviewees who get job offers. The widespread use of the Internet has afforded job searchers greater ease to apply to dozens or hundreds of positions. With the subsequent flood of applications, organizations employed computer search engines to screen the thousands of additional resumes by using keywords that signal specialized skills.
The new wrinkle on these established hiring practices is using algorithms to find “fingerprint” sets of traits, values, and behaviors that are particular (or peculiar) to every company. These sets would be difficult to assess with traditional interviewing and resume scanning, and might be done more cost effectively with computer assistance (less need for HR employees). What makes these “fingerprint” profiles different than older, more established traits for hiring like intelligence, integrity, or conscientiousness is that they can be based on quite odd and idiosyncratic profiles specific for each company. Do you dislike paying high taxes in this country? Are you sometimes not sure what you actually believe? Strangely, your answers to obscure questions like these might get you the job, or might leave you without a call back.
Wall Street is now pushing the envelope even further, combining application, resume, and assessment data with its big data analytics prowess to find the hidden oddities of the hiring process. In the best-case scenario, algorithms identify important pieces of information about candidates that normally would go unnoticed or be readily dismissed by HR or hiring managers. By using data from existing and former employees, particularly variables that predict turnover and performance at that company, employers can optimize their new hires, improving the likelihood of loyalty and high productivity. It is a Moneyball-like approach, but customized to the particular cultures of each firm. These techniques are transforming how organizations manage talent.
What, though, are the potential costs of using these hiring algorithms?
It depends on how the algorithms are used. Like most tools, they make certain tasks easier and more efficient, but are also subject to two main problems that could cause the company to overlook better candidates. First, the output of hiring recommendations is subject to error and bias when designing the algorithm. Humans still decide on the inputs and how that data is processed. While the computers are not biased and not subject to error, the people creating the algorithm are a different story. Second, although the algorithm might be well-designed, these programs can find spurious patterns in the data that turn out not to be meaningful for business outcomes.
Diversity is at high risk to suffer as well. Diversity is typically thought of in demographic terms (race, gender, age, etc.), but for business performance, cognitive diversity is just as important. Companies use shortcuts in this regard, generally assuming that people with different demographic traits are also cognitively different. For instance, a woman and a man might not necessarily have different worldviews. They could actually be incredibly similar cognitively based on how they were raised, their values, their educational backgrounds, and so on. So just as algorithms can make a firm more diverse in terms of traditional measures like race and gender, the fingerprint profiles decrease diversity along these less obvious traits, making people more similar across those dimensions in companies. Firms then end up choosing candidates based on the traits that the incumbent employees have, making a good-ole-boy hire quantitatively justified.
Wall Street is in a particularly tenuous position using these new analytics. The dirty little secret of Wall Street is that it already has a major diversity problem. Karen Ho’s Liquidated (2009) and Lauren Rivera’s Pedigree (2015) are both substantial studies that outline investment banker hiring practices that tend to exclude women, racial minorities, and individuals not raised by wealthy families. Further, these studies show that firms employ extreme pipeline hiring tactics, targeting a very small number of schools, eschewing bright and talented candidates solely because of where they are, or were, enrolled. Ho points out that Wall Street banks already hire from established networks for their best, most high-potential jobs, and they fulfill their diversity quotas by hiring minority and female candidates to fill their low potential, low-pay jobs.
Will big data hiring analytics stem these discriminatory practices or reinforce them? It largely remains to be seen. How the analytics are employed will largely determine their impact on who comprises the firms of the future.
Rhett Andrew Brymer is the John Mee Endowed Assistant Professor of Management at Miami University’s Farmer School of Business.