Predicting hiring success

In our view of the world every event has a probability, and a more informed understanding of this probability enables better decision making which is ultimately the goal of our work.

Probabilities are expressed using a percentage, with a range of 0-100% and the sum of probabilities of all possible events summing to 100%.  Most people have some understanding of probabilities, much more than their understanding of more traditional statistical techniques and we like to describe event likelihoods as a probability.

One area where probabilities can be used is in predicting hiring success.  We’ve recently been using this technique to develop tools to guide recruiters in two scenarios:

1. as a tool which enables them to change the target level and understand their likely chance of reaching this target with the existing pipeline
2. as a tool to monitor their likely success as the underlying pipeline changes.

Both instances rely on previous experience, refined by current data to make the prediction.  We produce a set of probabilities using a simulation technique the result of which can be shown as a probability distribution:

Rarely are we interested though in predicting the chance of reaching a particular number of hires, our interest is in understanding whether we will reach at least a particular number of hires.  From the above example (which was based on a pipeline where 7 hires had already been made) we calculate the probability of the expected number of hires being greater or equal than the level we are interested in.

This probability is then expressed as a probability in a dashboard element as shown below:

blue bar shows current level, also labelled.  The orange bar is the target which can be adjusted by the recruiter using a slider-type control.  The probability of hitting the target is shown.

These probabilities will be updated as the underlying pipeline data changes, and those predictions can be reported to help increase the accuracy of our prediction simulation.  As the recruiter gets closer to the final target the prediction polarises towards 100% or 0%.

This technique can be used to predict almost any type of workforce question based on the underlying data.