Predictive analytics means one of two things. The common meaning among analytics professionals is that it involves some kind of statistical model. This model could be based on anything from multiple regression to machine learning. The usual example is flight risk. We see which factors are associated with someone quitting, and then we can look at our current employees and predict who is likely to leave. (Some analysts make a distinction between models for prediction and models for forecasting, but that need not concern us here.)
The second meaning of predictive analytics, the one intuitively understood by managers, is captured in the phrase, “So what?”
If you show a manager a dozen slides on absenteeism data in past years, then they are right to ask, “So what?” If your answer is, “If we reduce absenteeism in experienced employees by 10%, it will reduce overtime costs by 6% next year,” then that’s the kind of analysis managers are interested in.
Managers don’t really care what mathematical techniques you used to get to the “So what?” They are just interested in the implications the data has for what they should do in the future.
People analytics professionals will be interested in the origin of the mania for predictive analytics. In the past, HR was known for giving incredibly long PowerPoint presentations showing, for example, employee survey results or candidate experience for the past few years broken down by gender, department, age, performance rating, and various other data cuts.
Managers got frustrated at this backward-looking, descriptive data. They wanted to know, “So what?” They wanted to know what it meant for the future. Hence, predictive analytics — as opposed to descriptive analytics — became a kind of holy grail for talent professionals.
The point that HR and TA pros need to understand is that statistical models are not the only way to answer the forward-looking “so what?” question.
Looking at what will happen to a business unit if a trend continues is a useful kind of predictive analytics. Explaining that if the eastern business unit could get their overtime costs down to what has already been achieved by the western unit, then they would save half a million dollars is a compelling “so what?”
Another example: A descriptive finding that a lot of people leave after their first two years if they haven’t got a raise. So future managers should keep an eye on employees in that situation.
When you can do good statistical modeling, by all means go ahead. However, often HR doesn’t have the data, talent, or time to do so. The statistical modeling approach to predictive analytics is overhyped because there are simply not that many affordable, practical use cases in most organizations.
At the same time, the more rough and ready approach to prediction, simply using data to make a reasonable inference about what could happen, is underused because HR fears it does not feel scientific enough.
HR and TA should never share descriptive, backward-looking data if it does not lead toward some kind of forward-looking “so what?” Sometimes we may need statistical models to determine the “so what,” but often we do not.
Don’t interpret managers’ complaints that your data is just descriptive to mean that you need statistical models. They just want to know what the data implies that they should do going forward. If the back-of-an-envelope is the only mathematical tool you need to work out that implication, then embrace that tool and leave your machine-learning algorithms for other efforts.
