Our banking client was becoming increasingly concerned about the number of high value mortgage customers moving to competitors. The agreed strategy was to encourage mortgage customers to stay with the bank, by highlighting alternative deals and ways they can save money on their mortgages. However, to personally contact all mortgage customers would be prohibitively expensive and require too many resources
The marketing team decided to focus on high value customers who were most likely to leave (churn) and asked Sysware to help carry out a predictive modelling exercise to identify these customers.
By comparing the attributes of mortgage customers that had remained with the bank with those that had churned over a three-month period, Sysware identified over one hundred significant factors, related to customer, product and transactional data. As the required data was not stored in the bank's central data warehouse, Sysware developed extraction processes to collect data from a number of operational systems. The resulting data was collated, standardised and grouped and one long record created per mortgage customer listing all the attributes collected.
Mortgage customers were then split into two groups, one for all churners in the past three months and the other for non-churners. Through the application of statistical techniques, Sysware began hone in on the significant factors. For example, it was discovered that customers who were middle-aged, city-based, had gold credit cards and a professional job, tended to churn more than others.
The data was fed into a SAS predictive modelling tool that uses neural networking techniques to build a model to predict churn behaviour and factors with little or no influence were removed. Over many iterations using data analysis and trial and error, Sysware isolated the more significant factors, with the aim of reducing the number of factors in the model to around ten. This would prevent random swings in the data from making it too difficult for the modelling tool to achieve the best result.
The model was applied to all mortgage customers, ranking them in order of their likelihood to churn. Sysware graphed and presented the results, a "lift curve" illustrating that just 20% of the bank’s customers had an 80% probability of churn. If the bank aimed their marketing at this small segment of the customer base, Sysware advised they would receive maximum return on their investment.
The process was repeated over two months to confirm that those predicted to churn actually did so, and to verify the stability of the model from month to month.
A few months down the track, the model had proved its worth. The marketing campaign targeting those customers most likely to churn reduced the churn percentage by nearly half. For an outlay in the low six figures, the client retained many millions of dollars of mortgage business.
This project was tightly-focused and business-driven, giving it the greatest chance of success as well as the best return on investment. Such short term projects can provide achievable pay-offs within three to six months and provide an approach and basis for piloting bigger, higher risk projects.