Predictive Analysis in HR

We couldn't agree more with Max Blumberg on this: Don't Waste Money on Predictive Analytics: It Doesn't Exist in HR.

"Did you know that scientifically, the only way to guarantee predictive analytics is to set up a randomized double-blind controlled trial? In other words, this is the only way to guarantee that a particular set of predictive metrics is causing the desired outcome rather than some other factor that you hadn't thought of. This is why they're used when testing new drugs because the risk of incorrectly predicting the effects of a drug can be catastrophic.

Yet as we all know, even under carefully controlled randomized double-blind experimental conditions, we still sometimes read drug disaster stories in the news. Why? Because even under these stringent conditions, randomized double-blind control trials sometimes fail.

So therefore, in real-world organisations where so many hundreds of possible predictors which might be affecting the desired outcome we seek, the chances of true predictive analytics is virtually none. This is because randomized double-blind control trials are simply not usually an option outside of tightly controlled laboratories and even there they're highly suspect. 

Therefore as workforce analysts, we should really become more responsible and start using the term "correlational analytics" - because that is generally as good we're ever going to get in most organisations.

This is not just semantics. For example, if you believe that engagement predicts performance and spend £2m on engagement programmes to improve performance (as one telecom client did against good advice), you'll be in for a horrible shock - as they were when they discovered that in their culture, performance predicted engagement rather than vice versa. Thus their £2m investment had absolutely no measurable effect on performance.

The rules are:

  1. Accept that predictive analytics probably doesn't exist in most organisational settings and that you need to make do with correlational analytics
  2. Assess the risk of investing large sums based on analytics which are in fact correlational rather than predictive
  3. Always run a small pilot or two before spending big money on implementing the results of "predictive analytics" (which we now know are really just correlational analytics)."

Max Blumberg, 2015

(Source: https://www.linkedin.com/pulse/dont-waste-money-predictive-analytics-doesnt-exist-hr-max-blumberg)