How to retain staff

It was bound to happen. As soon as the economy started to turn retention, a topic somehow off the radar for the last few years, suddenly becomes a hot topic.

There are lots of different calculations for the cost of replacing staff, but needless to say that for a big firm reducing turnover by 1% can lead to significant savings. Like customers, it costs a lot less to keep existing staff than to recruit new ones.

When looking at the issue of retention it’s important to remember that the probability of staff leaving isn’t uniformly distributed across the organization. Different groups have differing probabilities and differing reasons to leave. This is why measuring average retention rates is so absurd, and why targeting them can be so dangerous. What you need to do is understand where the pockets of high-risk are, the economic impact of the risk for particular groups, what is causing the risk, and what can be done to profitably reduce the churn.

Before making a blanket pay-rise to all staff or increasing equity options here is our suggested approach to improve retention:

Understand what is happening.

The first place to start is always by understanding what is happening. We prefer to do this by first looking at what the data says using various Data Mining techniques. Take your HR transaction data, employee details and supplement it with a range of other information such as economic data, consumer confidence etc. Whereas with classical statistical techniques we kept the number of variables small with data mining this isn’t an issue. You’re probably best starting with a tree-based algorithm and changing the parameters, pruning the tree until you have something which is manageable.

What the tree algorithms do is provide a set of groups that are statistically most at risk of leaving in a way that is easy to understand by the human eye. Using this you need to prioritise which groups are going to get your attention (and what you’re willing to spend, in both financial and resource costs) to reduce the risk.

Identify why it is happening.

Now you know who you need to target it’s much easier understanding key reasons. Certainly you should be looking at exit interview data. If this isn’t already being done in a way that enables easy analysis now is probably a good time to start. You’re looking for key reasons that are pronounced for your group, not only to understand what needs changing but also to understand if your predictive model includes the relevant variables. 

Fixing the issue.

By now you will have an understanding of who is at risk and some likely reasons as to why they are at risk. You can then starting targeting interventions that are specific and with an increased likelihood of success. 

Think of them in the same way as marketing would look at reducing churn – building offers that are relatively low cost (in comparison to losing someone), offered in a targeted way at the right time. Think, for example, of the male 30 year old who is just about to become a dad for the first time. You might realise, even before he does, that a new baby will likely cause him to re-evaluate his work-life balance. A friendly supportive conversation in advance about what opportunities for flexible work arrangements the company offers might be just the thing to reduce a risk later in the year. Most staff don’t understand what is available to them and just having a conversation, and leaving an open door can improve the odds of him speaking to the company before jumping ship.

The joy of this approach is that it can be maintained, and that the analytical model can be updated over time. Risks change throughout an economic cycle and cyclical throughout the year and are reasonably predictable. Therefore think of retention as a series of campaigns, targeted to the relevant individuals just a the right time.

Some extensions.

A model like this can also be improved to further reduce cost. Two examples are:

Understanding who to counter-offer.
Counter offers are widely used in several industries. A predictive model should be able to be developed to predict who to counter-offer. Of course a better approach would be to use the model to make the change before the resignation

Boomerang hires.
We know that on average about 25% of people leave their employer in their first year of service. Knowing this could be used to build a program to actively predict and re-hire certain ex-employees who you’d like to have back.