article » Big Data Comes to HR

Big Data Comes to HR

October 9, 2013
4 min read

Data mining can be an effective tool for making better hires, giving employers the ability to find new and unexpected relationships in the numbers. But there continues to be times when human judgment has to be factored in as well.

The Wall Street Journal recently published an article about the rise of "Big Data" in human resources, especially in hiring, and the story seemed to touch a nerve.

Big Data refers to the ability to look for relationships in big data sets, like the items bought by thousands of consumers in grocery stores over a year. Especially in human resources -- where the size of the data sets is much, much smaller -- a more accurate phrase for the new developments is "data mining," where we are fishing for relationships in data without a lot of guidance as to where to look.

Most of us have grown up with hiring processes that seemed pretty informal: A personal contact leading to some interviews and then to a job offer, with little or no data analysis involved. So it is surprising to hear that this was not always the case. A generation ago, a candidate for a white-collar job in a big company would have found themselves going through days of testing -- IQ tests, skills tests, interviews with psychiatrists, you name it.

That died off, in part because employers were no longer making lifetime hiring decisions, so the benefits of careful screening weren't so big, and in part because the new threat of being sued for discriminatory hiring practices. Companies backed away from formal hiring practices that were easy to test for discrimination in favor of informal practices that were more difficult to track.

Now a lot of those hiring tests are making a comeback, in part because they can be done online at significantly lower cost. Perhaps surprisingly, the interest has begun not with management jobs but at the lower end of the labor market, in call centers. One reason why is that these call centers hire so many people, largely because of high turnover, that it could pay to put in place the initially expensive screening systems. The other reason is that it is relatively easy to track individual performance in these jobs, which in turn makes it easier to use data-mining techniques to determine what constitutes good performance.

This approach is fundamentally different from the more common computerized application systems. The latter filter out the online candidates who do not have the basic qualifications for a job, and then turns those who do over to recruiters for a closer look. In practice, though, there is very little real evidence motivating those systems. The criteria used for the filters typically come from the gut hunches of hiring managers (e.g., "We need someone with a Ph.D for this job.")

Data mining is different because it is asking the empirical question, What actually does predict who turns out to be a good hire? Some of this is being done by companies studying themselves, Google being the most famous example when it discovered that their previous practices were not that successful. But most of it is being done by vendors, especially recruiting-process outsourcers.

What the data mining is finding is in part what those recruiting and staffing experts from the 1960s already knew. Beyond the usual qualifications and experiences, factors like IQ and personality can help find the better candidates. But they are also discovering new relationships that were not expected by hiring experts. One vendor, Evolve, found that job hopping among candidates was not a bad thing. Candidates who had changed jobs a lot were not more likely to change jobs when hired into the call centers they were managing. Why would you not want to know information

There is always a risk of messing this up, of course. The easiest way to do it would be for employers to rely on only one result from data mining, such as only using IQ to assess candidates. In practice, even the best of these statistical relationships account for only a small amount of candidate success. A lot of information is needed to make good hires; and a lot of human judgment is required as well, especially when the different pieces of information do not point in the same direction.

But looking at actual relationships between individual attributes and performance is a big step beyond going with the gut feelings of hiring managers and incorporating them into software.

We may like the idea that hiring should be about the plucky applicant who persuades a human recruiter to hire them. But with thousands of applicants for most every position and a lot of money on the line, it is not surprising that employers are once again turning to more sophisticated analyses to solve the problem.

For the full-length version of the article, please click here.