Predicting Churn is not Enough to Reduce it
The accuracy of churn prediction has improved greatly during the past seven years that we have been building churn models for our clients. But our customers, who are often large B2C companies, have been asking for more. They told us that our models make them anxious as they know well who is going to leave, but they do not know what to do about it. Churn predictions alone are of little use to them, without knowledge how to lower it.
There are two ways one can go about reducing churn. One way is to proactively contact the likely leavers with a retention offer. We have done projects where we have recommended which personalised offer (type, price, language, channel…) should be presented to which customer, so it is relevant and appealing to them. This has helped to improve the retention significantly. But this is costly and tempts some customers to wait for the offers. And importantly, the offers do not fix the problem of why these customers want to leave, as the models do not reveal the root causes of dissatisfaction.
This takes us to the second way of reducing churn: understanding what causes it and fixing it. Be Customer Smart have developed, together with leading academic researchers in this field, models that tell which factors drive customer loyalty and satisfaction and how much. This information allows to see which parts of the customer journey should be improved and what is an ideal target for each KPI.
To do this, we take as inputs data that organisations already produce in their daily operations. Depending on the industry, the market, and the company it can consist of things like CRM data (for demographic filtering to look at what is important for each segment), operational data such as how long did it take to solve a customer issue, and possibly also call centre recordings to understand how customers are served. A common push back we receive is that this data is not available, but we haven’t yet come across a case where there was no usable data at all.
In the results, the drivers are divided into two groups depending whether one can influence them or not. For instance, customers’ age can be a strong predictor for churn, but you cannot change that (except by focusing your new customer acquisition to only more loyal segments). But a list of the drivers that can be influenced is really valuable. It will tell you how much of the churn is explained by which KPI at each value point. For instance, how much churn or NPS changes if the call centre waiting time is reduced from five minutes to four. As the experience improvements cost money, knowing their revenue impact facilitates evidence-based decisions.
The current alternative to get these insights is to survey your customers, but as only few percent of them bother to answer – and usually those who either really happy or unhappy. As we cover your whole customer base, you get a full unbiased picture.
Knowing who is going to stop being your customer soon is nice, but only when you know why, you can do something about it. And we can provide this to you.