Predictive Scoring Model Drives Revenue

Last week, I teamed up with a client partner to present a case study at TechTuesday in Birmingham. Below is a quick look at the project we discussed.

As part of a revenue growth initiative, a quick-service oil change retailer saw an opportunity within its oil change customer base to convert “oil-only” loyalists to its mechanical services as well; they wanted to be more of a one-stop shop. Across 300 stores, having a higher percentage of customers add extra services would mean a large increase in short-run revenue and a potential for long-term customer lifetime value and loyalty. The client team wanted to make sure they were approaching this challenge in the right way, so they called in our Data and Analytics team to create a scoring model for use in a mechanical coupon offer. The model would enable a view into which customers would be most likely to convert and target them specifically for the offer. Once live, redemptions could be matched back to the CRM list and campaign ROI evaluated.

The first step the data science team took was auditing the customer data that could prove valuable in predicting a customer’s conversion propensity. Then the client transferred its customer-level data to North Highland where it was landed and securely stored in our Insights Lab. From there it was off to the races diving into the data to mine out the predictors of conversion, and vetting those findings with the client along the way. The basic idea was to find behaviors of customers during their “oil-only” experience that correlate to future first-time mechanical conversion — and then put those together in an equation that the client could deploy into their data warehouse. In total, we wrangled and created 372 customer-level variables as inputs into the model including various RFM concepts. To supplement the 1st party data, customer PII was sent to Acxiom to enrich the file with 2,500 data points across a variety of demographic, lifestyle, automotive, shopping and census elements. This was truly a big data effort!

The final scoring model contained the best 10 predictors from the candidate pool of 372 customer variables. And of course, there were some surprises in our findings along the way. The time since a customer’s last oil change was a huge factor in the model, with those who recently had their oil changed as the customers who converted in the future. What was surprising is how recent it was – customers who were there just days / weeks before for an oil change had highest propensity. This further manifested itself in that customers with longer “in-bay” times were more likely to convert. So although their experience wasn’t as “express” as other customers, the technician was reviewing recommended repairs with them versus someone with a newer vehicle who may have been just in and out. We also found a significant decrease in the future conversion rate when customers received a recommendation to replace their air filter. This correlates to consumer sentiment that air filters are the most oversold product in a garage.

These findings – in addition to being helpful for the original task of determining propensity to convert – also prompted the client to look at how customer experience comes into play when a customer receives such a recommendation and will serve as an area for the company to improve their CX moving forward.

Related Reading