The Wall Street Journal recently wrote an article called, “Meet the New Boss: Big Data”. The article’s bottom line was that, increasingly, algorithms based on predictive analytics are deciding who should be hired for a job, and who should not. Humans need not review the resume – machines will handle the decision making.
The fact is that analytics used for hiring are a lot more human than the article described.
When I read the article, images of Terminator robots taking over the world crept into my head. There is a fear, maybe legitimate, that analytics on talent data is the beginning of the end of civil society (and possibly a fear of terminators sent from the future to destroy us.)
Not all data is fair game when analyzing talent
Many human characteristics used by advertising, e-commerce, and insurance analytics are acceptable to analyze in these contexts, but unacceptable to consumers when analyzing talent. I realize that even in B2C, this is a debated topic… but hang with me here. An example used in the article was the distance someone lives from the job. Consumer marketing uses this data all of the time and it’s wonderfully effective. If I live closer to the big box store, maybe the analytics engine sends me more email promotions for grocery items than if I live 15 miles away. However, it is illegal (in the US at least) to base a hiring decision on where someone lives. Period. Full Stop. Companies can’t use that data to make a hiring decision through interviews, and analytics folks can’t use the data to recommend a hire.
There are lots of possible predictors of performance and turnover that a future employer could use, but don’t pass the sniff test. Frequently people leave their job after a divorce. Should we factor in the number of weekend getaways a couple has into our hiring decisions? A few years ago, an acquaintance was convinced that a leading indicator of an administrative assistant’s success at her firm was observing if the candidate’s handbag matched their shoes during the interview. That’s a test I would have failed.
There is a line that we, as analytics professionals, need to be wary of – and I know readers are, when it comes to using personal data.
Solving for the Right Problem
One of the cases in the article talks about hiring for a call center. You’ve probably interacted a call center enough to know that these are tough jobs; there is always turnover. I don’t know a call center manager today that wouldn’t love lower turnover. I’m not judging the particular company’s problem / solution, but we should not treat turnover prevention as a cure-all for customer service ills.
Low turnover is not what the business wants. They want happy customers, handled on the first call, customers who would buy again and tell their friends. Zappos offers a cash bonus to new recruits to quit after being trained just to ensure that highly committed people are working the phones. They encourage early turnover.
Solving for a single factor like turnover is an excellent way to hurt your business. One of my favorite turnover stories comes from a big box retailer a few years ago. A CEO called his analytics folks into his office alarmed. He saw that turnover was highest among his best sales people in the stores. The analytics team did a deep dive into what makes a top salesperson at their stores. They looked at age, education, psychometric scores, hours worked, how long they remained top performers, and their sales numbers.
They then walked into the CEOs office (a little nervously) to tell him that he’s worrying about the wrong problem. He shouldn’t fret about too much attrition of his high performers, he should worry that there isn’t enough attrition. Here’s why:
It turned out that for this business, their top sellers were college students who were tech savvy. They were smart, personable, and over-qualified for the job. These hungry college students were happy for extra cash while it lasted – their “top seller” status had a shelf-life. If they stayed too long, their sales shrunk.
What the analysts recommended was not to reduce attrition among this group, but to monitor that the right employees were leaving when the time came. The company should focus on creating a pipeline of strong future talent, and a set of “B players” – salespeople that were good, but not great, to handle training and provide consistent sales.
If this company tried to solve for attrition alone, they would have reduced their revenue, not increased it. There’s the folly of following one assumed good outcome in talent related analytics.
Hiring still needs help from analytics
Hiring is still an inconsistently managed function. We can do better, and analytics can help us. A few years ago, IIA’s own Tom Davenport debated Malcolm Gladwell on the use of analytics vs. gut instinct at a SAS conference. One of Malcolm’s points in favor of using data applies to hiring I think: there are so many possible choices, so many possible mistakes to make that analytics can help us focus our decision making on the things that matter most, particularly when humans can’t detect the differences. That’s what analytics does really well in hiring (when used correctly): they focus us on what matters most, and highlight signals we can’t see.
Skynet, whenever you do come to rule the world and tell us which jobs we can work in, I’ll be happy to revisit my stance on using every scrap of available personal data, probably because you’ve sent a scary robot standing behind me as I type. Until then, let’s help the community at large understand what analytics is really used for, without all of the scary stuff.
Originally published by the International Institute for Analytics