Goal
A leading specialty retailer of premium pet food, supplies and services, with 1,500+ locations across the US, Canada and Mexico. Revenues for existing vet hospitals varied a lot, so the ability to correctly identify those more likely to yield higher revenues can deliver significant business value.
Approach
Developed segmentation model from customer purchase data, loyalty program, competitor traffic and market credit card usage augmented by 3rd party data to provide additional customer context.
By studying how customer pet store spend drives purchase behaviour in vet stores we were able to predict which pet customers would spend in vet for all locations (including pet store locations with no vet).
Due to the discovered relationship between the total vet revenue and the number of customers who spend in both pet and vet, the model can also predict total vet revenue.
Using SHAP techniques, each prediction can be explained with the specific attributes that resulted in the model making that prediction
By running predictions for all customers across all pet stores, a predicted vet revenue for each store can be used to order a list which is coupled with the SHAP explanations and additional context
Outcome
Delivered a prioritised (and explainable) roll-out action with production-ready application (and training) so that the retailer could continue to rerun and revisit the capability.