Uncovering the true Customer Value by using Survival Analysis
Chloé Van Vreckem
4C is building ‘customer companies’ and we do it based on 4 values:
Our enthusiastic and motivated team scores very high on the first three values. In case of the fourth, inspiring part, we are expanding our ways!
A great opportunity to spread the word of innovation was to speak at the BAQMaR conference (on 16th of December 2016, after 10 years of sharing bright ideas, there was the very last BAQMaR conference - Red.). The theme of the year was to inspire business with innovative techniques and make sure that they go home with a fresh view of the future.
During the data science track, Marie Willo (CI manager at AXA Belgium) and I have presented
“Uncovering the true Customer Value by using Survival Analysis”
Survival analysis is mostly used and well known in biomedical research, however we can make the link with the business easily. In medicine, we want patients to live (or “survive”) as long as possible versus in business we want to keep the customer as long as possible within the company with his products. ‘In medicine, time is crucial” and “in business, time is money”. Why don’t we use this survival technique not more often in a business setting? And why should we consider this technique?
The three main benefits of survival analysis
Based on just one model we see a global view of customer behaviour throughout time
Based on survival curves, two marketing programs can be compared at each point in time. It can highlight moments in time where customers are at higher “risk” to churn or at higher “interest” to buy. This might have been missed out by using more static techniques where an arbitrary point in time has to be chosen. In example: did my customer churn after 6 months: yes or no.
We can use the entire population
In insurance there are a lot of customers who are not really churning but they are ‘out of risk’ because they don’t need the insurance anymore. In other techniques these involuntary churners are excluded making the sample smaller with loss of valuable information. In survival we censor these customers taking all available information into account.
Model variables give valuable insights for marketing campaigns
A good model is not a model that is only accurate and statistical significant. It has to be interpretable too. We have to explain to the business and the marketers which factors have an impact on the outcome of interest and in which order.
Because, in the data science track, the attendees of the BAQMaR festival were mainly data scientists we offered some exciting formulas and the Cox Proportional Hazard model too (which is used for our survival data). We hope we have inspired these “business scientists” to try out survival analysis additional to existing modelling techniques that they are currently using.