06 Nov. 2013 | Comments (0) Share Follow @Conferenceboard
Login to the Human Capital Exchange website and watch Frank DiBernardino discuss his book, Optimize Human Capital Investments: Make The "Hard" Business Case, during our October Book Discussion webcast.
I skate to where the puck is going to be, not where it
~ Wayne Gretzky
It’s a struggle to swim against the current, and that’s just what one is doing when one takes on the concept of predictive analytics – a concept that is very much in vogue in the HR analytics space.
Predictive analytics attempts to relate what we know in the present to what we want to know about the future. Predicting future outcomes based on historical data is not only challenging, it’s dangerous. Just ask any weatherman, economist, actuary, underwriter, or statistician.
Maybe it’s my entry into the business world as a group insurance underwriter that causes me to be skeptical about the notion of predictive analytics. In this capacity, I underwrote all forms of group insurance, including life insurance, disability insurance, several forms of health insurance, and defined benefit pension plans – so it was my job to determine payable rates, and underwriting terms and conditions. As you would expect, I considered historical experience (among several other factors) in making these determinations. A year later, I would review the financial results and proceed through the renewal process. It was the very rare exception that the actual financial results were as I had projected. Contrary to what economists call Ceteris Paribus (all else being equal), there were just too many uncontrollable factors that impacted the outcome.
The same is true with human capital predictive analytics – there are too many uncontrollable factors that can impact outcomes.
Experts in the field of predictive analytics talk about observation periods and projection periods. During the observation period, they stratify the data to identify the characteristics and patterns that led to the outcomes. If those patterns continued during the projection period, then one can predict, within a narrow variance range, a future outcome.
Predictive analytics expresses the future in terms of probabilities, and while no application can predict the future with absolute certainty, its proponents argue that it helps management make decisions that minimize the risk and increase ROI. When properly applied, it will reduce variability and help minimize the likelihood of choosing a wrong course of action.
Potential Career Jeopardy
The use of predictive analytics, however, is quite risky and has significant pitfalls, as Amit Mohindra, Vice President Workforce Intelligence at McKesson, explained in a Talent Management magazine piece entitled: The Road to Analytics. “The problem is that relevant statistical models, such as logistic regression or data mining approaches, are complex. They are hard to explain, and the results need to be qualified with so many statements, the audience soon loses interest and confidence. Further, the outputs from predictive analytics are denominated in probabilities, odds, and likelihoods, numbers the average person isn’t comfortable interpreting. Last, but not least, because results are probabilistic in nature, there’s no guarantee a prediction will come true. When it doesn’t, no one will remember the prediction was qualified; woe to the talent manager unfortunate enough to make two faulty predictions in a row.”
A Prudent Path Forward
Returning to the stream analogy, how might one prudently embrace the concept of predictive analytics and successfully navigate its unexpected twists and turns? There are several ways:
1. Use a recurring cycle – The concept of predictive analytics needs to be an ongoing process.
Produce the metrics on a regularly scheduled, recurring basis. This is vital for trending and measuring the impact of human capital strategy changes.
2. Segment results – Strive to slice the results among homogeneous groups. Internal comparisons, especially over time, are a powerful and credible point of reference to identify those units performing well and vice versa. In addition, internal comparisons help establish realistic goals.
3. Dissect and analyze results – Endeavor to identify the factors driving the outcomes across the segmented results. For instance, what is it that high-performing business unit A is doing that can be applied to low-performing business unit B? This analysis requires the constructive cooperation of HR and operations.
4. Define human capital strategy changes – Based on what was learned in the first three steps of the process, what human capital strategy changes are appropriate?
5. Project the outcomes – Based on the proposed human capital strategy changes, estimate the impact on the human capital metrics and business outcomes.
Finally, avoid using the term “predict,” and embrace the terms commonly used in other business disciplines, such as sales and marketing, finance, and operations: Estimate, forecast, project. All three are conditional terms, leaving the user some wiggle room.
This five-step path described above is consistent with what sales & marketing, finance and operations do on a regular basis. Instead of predictive analytics being a fool’s errand, it becomes an integral element of the organization’s human capital analytics landscape.
View our complete listing of HC Analytics blogs.