The start of any year is the time to focus on trends. Predictive analytics is one of the trends that looks “hot” for 2012 in the world of human resources measurement. The promise of predictive analytics is enormous. For example if we look at turnover and were able to predict how much turnover was coming from where, and when, it would allow us to focus our recruiting resources far more effectively, reduce vacancy rates and better support the business. The ability to predict outcomes brings to HR the opportunity to be proactive and demonstrate the value that can come from well run people systems and processes.
Strategy is a future-oriented view of the organization and the ability to deliver predictive insight brings HR ever deeper into the strategic realm of the organization. However, as with many such trends, there is a lot of focus on the concept and the promise, and less on the mechanics of how you actually make it happen. I have heard and read a few thought pieces recently that are asking for good predictive numbers and decrying the more standard HR measures, such as turnover, because they are not predictive. This is not true. The ability to predict comes from having robust and deep historical data. There is no shortcut to predictive measures. Prediction relies on history, and without history, any predictive attempt will be shallow at best, and more likely wrong.
The value in predicting turnover is clear. Hiring resources are scaled to needs and are therefore efficient; key roles are empty for shorter periods of time, reducing productivity losses; panic hiring is reduced based on more realistic expectations, leading to potentially better hires. The question becomes, how do you know what your turnover might be without knowing what your turnover is now?
There are some good emerging models which look at several factors relating to each employee, such as length of employee commute, time since last role change, performance ranking overtime, etc., which are demonstrating good predictive function for voluntary turnover. These models have come from years of study of actual employee behaviour, and for them to work in each specific organization, they rely on being tested and refined for that specific context. For example, a commute of 30 minutes maybe too long for employees. In other geographies, a commute of 45 minutes may prove to be too long. The factor (i.e., the commute) is a valid factor, however, without understanding the history of how this factor plays out in your organization, any prediction is at best a guess.
This is our message for the start of 2012: prediction is a worthy goal. It can offer incredible value and should be a goal for any HR person or department looking to continue its shift into the strategic realm of business. However, it is not possible to jump from no measurements to predictive measurements. It is also not correct to decry the measures of the past. Yes, they only tell you what has happened; however, prediction is best done by understanding the patterns of the past and projecting them into the future. Without this deep historical picture, there is no ability to predict. Therefore, if you want to pursue the goal of prediction, your first step is to build your historical set of data. Without it your pursuit of prediction will have no future.
Ian J. Cook
Latest posts by Ian J Cook (see all)
- HR analytics process: Ask better questions - January 9, 2013
- People analytics for business: In high heels and backwards - December 12, 2012
- The winds of change: Making HR measurement happen - November 13, 2012