Making a Little More Sense of AI in HRTech

On January 4, 2018, in HRExaminer, by John Sumser


“I’ve been looking at a ton of ‘flight risk predictors’ over the past year. None of them addressed their tool’s ability to surface evidence of sexual harassment (i.e., increased attrition from women in specific departments). None. Of. Them.” – John Sumser

An opinion piece in the year-end issue of the Economist (paywall) notes an important consequence of sexual harassment on the job:

“The victims went on to earn less than other women; of those who had been verbally abused repeatedly or physically touched at least once, 79% left the company within two years.”

I’ve been looking at a ton of ‘flight risk predictors’ over the past year. None of them addressed their tool’s ability to surface evidence of sexual harassment (i.e., increased attrition from women in specific departments). None. Of. Them.

When do you suppose we will start to see flight risk forecasting that is focused on systemic issues rather than on individuals who are likely to leave? It’s reasonably clear that flight risk forecasting that focuses on individuals produces more attrition than it reduces. Attrition forecasting that focuses on the ‘system’ shifts attention towards more intractable problems than individual employee motivation.

And, individual flight risk forecasts are potential weapons in the hands of a harasser. HR is wise to carefully consider exactly who has access to this kind of forecast. Enabling a bully, no matter how well credentialled and powerful, is not likely to be tolerated going forward.

Gender-based inequities are the observable tip of the iceberg. Flight risk forecasting might better be considered as a primary measure of organizational health, employee experience quality, and the impact of other morale-crushing social inequities. As with many early-stage technologies, initial assessments of the value of predictive tools have been off base.

Focusing on individual flight risk makes it impossible to see the real causes of attrition. The very same data is much more effectively used to diagnose system-wide issues. The idea that the focus of HR’s work will become the identification of productivity improvement opportunities seems farfetched. But, as the impact of the overall ‘system’ becomes clearer, things are going to change.

Employee sentiment is more fully expressed in behavior than it is in survey responses or free-form text. One area of investigation would be the comparison of survey and NLP outputs with actual attrition rates and other behavioral data. There’s a real dissonance between what is said and what is done. Surveys and free-form survey inputs are always subject to distortion. Telling management what it wants to hear is a time-honored survival mode. Automation and intelligent parsing don’t change that.

It’s clear that the earliest “AI” offerings are primitive answers that simply get the conversation started. What’s most interesting is that the data can be used (as in the case above) to see bigger things. Most employee success is the result of system dynamics (although that’s not what you’ll hear). The next horizon is using predictive tools to better understand the organizational dynamics that drive success. 

As a conclusion, it’s worth noting that data models are evidence and create evidence. It is super important to take a hard look at intelligent tools and see them from a jury’s perspective. The question is ‘what should you have seen’, not ‘what did you see.’


Read previous post:
HRIntelligencer v2.01

The longer the hype goes on, the fluffier the material. For the most part, this week's articles continue our routine...