Predicting employee turnover

Predicting turnover should be the first step towards managing it. If you understand that Jim is more likely to leave than Jane for whatever reason then you can do something about managing that risk. In many instances it isn’t desirable or efficient to extend the ‘offer’ that you make to Jim to all employees.

Certain industries have managed similar issues for some time. Telephone operators use churn analysis to predict when subscribers are likely to leave and to make them targeted offers to stay. For a telephone operator factors such as if one of your friends leaves the network dramatically increases the likelihood that you will follow. The best time to make an offer is when someone is thinking of leaving but has not made the move.

Employees don’t all share the same probability that they’ll leave. For a recent graduate in their first job the risk might be as high as 40%, for someone who has been with the firm for 20 years the likelihood that they’ll resign may be sub 1%.

Whilst some of the factors that determine risk are hard to identify others are probably already in your HR system. From these factors a conditional probability can be identified for each employee. From this it is relatively straightforward to identify which 100 people are most likely to leave.

These predictions don’t say what will happen but rather guide to where the issues are more likely. The objective of a prediction is to provide information which is better than an educated guess, or intuition at what will happen.

The most useful technique

All employees will eventually leave, the important part is to identify when they are going to leave, and then the associated driver. For reasons such as retirement that may be relatively easy, for other reasons less so. Whenever you need to predict a time-related dimension the correct analytic technique to use is survival, or duration analysis.

The key difference to survival analysis compared to other prediction techniques is that you use all the employee data, not just the cases where an employee has left. These non-leavers’ data is technically called censored and the approach of how to use this data is what differentiates the technique.

What survival analysis shows is that the reasons for leaving differs depending on certain time-related variables. The reason someone leaves in their first year of service is different from the reason someone leaves in their 10th year. As reasons differ, and can be allocated the HR team can start to develop initiatives that can be targeted effectively. This reduces the cost of deploying the initiative and probably the likelihood that the initiative will succeed. What is more the effectiveness of the initiative can be measured.

External factors

Some external factors play significant roles in determining turnover. Employees leave the organisation if they believe that the benefits offered by other alternatives outweigh the benefits offered where they stay. They will look into the future, but apply a discounting to less immediate benefits (this discounting differs by individual type). They understand that they don’t have full information so make a bounded rational expectation with what information is at hand. Finally, the job search has an associated cost, and most recognise that the risk aligned to a new employer exceeds in many instances ‘the devil they know’.

Confidence plays an important role in these assessments. Confidence of the individual that their skillset has external value, and confidence that the job search will be low in risk. As mentioned in an earlier article this job-search confidence is associated with consumer confidence measures, which reflect individuals’ perceptions of the health of the economy.

Where should employers start when analysing turnover data?

  • Capture good data about reasons for leaving. This should be done by an independent party with confidentiality
  • Don’t look at averages, especially the mean. Means show the central point of a normal distribution and turnover data is never normally distributed. Look at the distribution of data
  • Use survival analysis to understand how the reasons are related to time-dependent t variables
  • Use external economic indicators to place your organization in perspective of the related economy and likely talent competitors.