You’ve got a predictive analytics model, now what?

Getting unbiased information out of data is de rigueur these days that I’ll be surprised if you did not have a data analyst / team you can work with. Analysts do what they do best and give you a ‘model’ that will say help prevent customers from leaving. How a business manager reacts from here will decide how your future is going to be — in bed with data. Here are the three Goldilocks scenarios I’ve come across:

Bottom draw: either you were just given a set of charts or you do not like what you saw, either way you do not want to act on it. Sometimes the charts and the statisticians way of interpreting data might not be best suited to a practical business leader. Plus this model hasn’t been proven yet has it? However a model is a chance for you to have the ‘hand of god’ / the invisible help that you need. You and your team might be running fast and hard but 10% in the wrong direction. All your model will do is to identify the course you need to correct and charge ahead. Not looking at what data recommends because it is not your priority is a lazy approach to ignoring the obvious.

So, what can you do about it: Managers who do not understand the prediction can sit with the analyst who built the model and check for correlation i.e. what few strong behaviours of customers lead to churn. In one of the predictive models I worked on, if a customer paid by ‘cheque’ they were likely to leave. The typical practice as a B2B service provider was to bill customers on a monthly basis and customers either paid via auto debits or via cheques. This innocuous option meant that these customers had the opportunity to re-think the relationship on a monthly basis.

My cup runneth over: we are living the effects of the ‘great resignation’ so already have a lot on our plate. You could have been several years in this business and what can a young geeky analyst who has never met your customer help you with? Does he know what you had to do to sign the contract or know what it takes to retain them? It’s easy to get into the mindset especially if you have spent a few years and are in over your head with everything that it takes to keep lights on these days. This is a great way to make yourself irrelevant to your customers. If you truly cared whether your customers stayed or left, how can you leave any stone unturned?

So, what you can do about it: don’t have someone else take responsibility or don’t ignore the predictions.. understand the recommendations, have a small team work on the recommendations and see if this changes the outcome. In operationalising one of the predictive models, the recommendation was to meet more customers because data said that if we when we have absences of interactions, our customers were more likely to leave. Now we had a 30-member sales and relationship team but who were just too tied up at the moment. To start actioning on the feedback, we set aside 2 junior level customer service staff with a sales personality to call every high risk customer. This was more of a relationship call, just to connect and each month we varied the themes of these calls — one month it would be about a new product upgrade, another month it was about an upgraded service or change of call centre mode, or introduction of a new mobile app.. we found reasons to connect with customers. The response was overwhelming. Customers appreciated the contact opportunity, we got some service requests that they were meaning to call us about, we were able to de-escalate complaints because we proactively reached out, some were even new business opportunities etc. but we could see a perceptible traction with our customers. Most of all, we did something to move customers through data predictions.

Your Goldilocks approach: if you know a your data team is working on something, be curious, find out more. You have the opportunity to get this going the right way. Every predictive model is built specifically for that business requirement. While the model is being built, understand the assumptions that data has made, correct them to reality. Talk about your business priorities, what you know about your customer concerns and how your business is set up.

Have the right people (someone who knows the business well, has access to your own reports and are eager to learn and help) pair the analyst from the beginning, get regular updates. This will make the model and recommendations much more robust and practical to your needs.

Get your team leaders curious about the model, see how you can operationalise it when it’s ready. Measure everything that needs to be tracked so you know if the model is giving you the results. Most of all give it time, your analyst will give you a model that will work with a strong confidence, so trust in this, even though you might be surprised with some of the findings like the customers who paid via cheques, go with it and see what happens. Today we run a business where we do not accept cheques but more importantly, our double digit churn is at world class levels.