Merit Matrix: Using Predictive analytics in building Compensation Programs

Merit Matrix, is there still any merit in it?
You can build different merit matrix model, but your underlying employee indicative data remains the same.

A What-if projection on compensation budget is built on, current Performance rating, current salary and market pay range. The budget alignment is computed by an iterative increase, to arrive at an approved budget.

A model that is built using above information has a flaw, when forecasting, the data set used for building budget should be in the future, instead of current.

With the advancement of predictive analytics and machine learning, the ability to build more accurate budget programs that aligns with your organization is what HR should be looking at. The magic 3% budget should be exposed in a way, its usage is more actual than accrual.

The merit in Merit Matrix can still be retained using the traditional method of building budgets only by applying modern heuristic algorithms.

Enough talking where am I going with it?

Let’s take any traditional Merit matrix and its components, Performance rating, Compa-ration (driven by Mid Pay) or Quartiles or Range penetrations. All this data, boxes employees into a matrix and an iterative method of applying increase percent guides to build budgets.

What if we have the ability to use the future data and build our budgets on top of it, that means our budgets are more accurate and close to reality. Let’s for a moment assume we have the power to predict employee pay increases, then we can build bottom up budgets using increases predicted by machine learning algorithms.

Using regression algorithms we can predict salary increases to a fair degree of accuracy, it gives you a fixed budget that is more accurate and close the actual requirement vs the 3%.

Now using the predicted pay increases we could box employees into the appropriate quadrants in the matrix based on the predictive distribution. Now HR can build guidelines and rollout a more accurate matrix to managers.

Once you complete the merit increase cycle, you can also compare the predicted increases vs manager input increase. A bias report, an outlier report and more finding, they are of immense value to HR.

Does this really work? The proof of the pudding is in the eating

We built few models for different organization with varying organization size from 300 employees to 30,000 employees, the organizations are from different verticals ranging from manufacturing to services to specialized job. For each organization we used different data points and so our models were unique by nature and the training data for building the program was specific to each organization. We validated our logic on the data sets. Our predictions was cross referenced to historic information of increases. We could arrive at a high degree of accuracy. We are now in the process of intervening user experience to influence the dependent variables used in the predictive analytics. I am working on to build a model that can be shared by customer but still has uniqueness in delivering specific value.