Before you give in to the promise of letting data discover new hire potential, be sure you know exactly what's being promised.
Big data is supposed to solve all kinds of problems. Look deeply enough into the patterns of what you wouldn't normally have time to investigate and you can find insights that will make a critical difference to your company. Or so the theory goes.
As I've said before, big data can be useful... once you've mastered "small data," or the normal high-level information you can use to better direct business. But say yours is one of the companies that has done so. You run an efficient ship. Can big data help you? Maybe, but only if you approach it with some wariness and ask the inconvenient questions when it comes to the assumptions and methods put into play.
For example, Evolv is a software company that tries to analyze big data to improve workplace productivity and performance. One of its conclusions is that "technological aptitude is increasingly important in selection of the hourly workforce." Here is Evolv's director of analytic products Nathan West discussing the concept:
The result is something that is seductive: A simple test could help identify better candidates for jobs. Someone who has downloaded and installed a browser other than the one shipping with his or her computer or device shows a willingness to adopt new technologies. Using the browser factor as an indicator, the company found that employees who scored higher on tests of "willingness to adopt new technology as well as technical proficiency ... actually stayed 17 days longer, missed 15 percent less work, and adhered to schedule much better when they were at work."
Sounds good? Maybe, but here is where you have to ask the tough questions and not drift away on a happy cloud after listening to a company's pitch.
Check the baked-in data assumptions
Where did the data come from? What kind of workers and companies was Evolv able to follow? How well would the types of tasks and workers compare to your operation? Is there any kind of bias in the data selection that might not apply to your company? For example, are all of the sample companies large enterprises rather than startups? You'd have to test to see if the data still applied because the universe of companies is so different.
What assumptions does the analysis make?
Is the test reasonable? It's fine to assume that someone is more technically flexible if they downloaded another browser, but that's an assumption placed atop the data before analysis. All you can tell from someone filling in an online form is the type of browser they have used, not the type of browsers on their computers. What if someone prefers the feel of Internet Explorer to Firefox or Opera? It could be that the choice itself indicates something about inclinations, but that is different from the stated assumption.
How important are the implications?
Having people stay longer and miss less work is a fine goal, but drop this into a bigger context. First, that sounds like an average. What is the variation in the number across different industries and employee bases? In other words, how likely is it that you would see roughly similar results? Second, in the context of how you operate, are the improvements meaningful? Staying an extra 17 days sounds great, but what percentage does that add to the overall average length of employment? Does it matter that much? Do you have employees that get paid sick time? If not, and if you have enough employee coverage, does the time missed from work even matter? It might, but then again, it might not.
What are the hidden downsides?
You've selected for one set of characteristics, but as they say in math, you can't maximize for two variables at the same time. Do the people who install a separate browser fit other parts of your company's culture? Are all quality of work measurements equally strong? They may well be, but if you don't ask, you won't know.
When it comes to big data, be open to using it, but don't do so blindly. Ask tough questions of the data, the analyst, yourself, and your business. If you're going to implement changes based on some analysis, you want to be sure that you'll get what you are expecting, and not an unpleasant surprise.
More from Inc.com: