Clarkston recently completed a conversion rate optimization model for a global medical device company providing leading orthodontic treatments. Our digital experts optimized a model key to patient/doctor matching to maximize conversion. The first goal was to provide clarity into the current model’s functionality, as several transitions in ownership left this key model opaque to company leadership. The team identified and corrected a suite of previously unknown issues with the model. The team corrected key pre-processing issues, such as miscalculated/mis-aggregated variables, hard coded values, and other data quality issues that translated into sub-optimized conversion rates.
Clarkston’s team corrected a business rule-based process that was layered on top of modeling code and changed the model’s recommendations. Our team was able to demonstrate that this add-on code not only hindered model performance but also show how those business rules could be incorporated within the context of the model to significantly benefit performance. Finally, our team was able to engineer new features, which both increased model performance and circumvented fraudulent conversion figures in historical data.
Clarkston’s Digital, Data, and Analytics team proposed a suite of new model types, from more robust ensemble tree-based and Principal Component Regressions (PCR) as well as options maintaining the existing logistic regression to ease interpretability.
For legal stakeholders within the client, the team was able to balance maximizing performance and potential conversion while working with regulatory constraints around prioritization of doctors for incoming leads. For executive leadership, the team used advanced techniques to simplify model variable effects. The process for this involved using both Partial Dependence Plots (PDP) and Accumulated Local Effects (ALE) to create linear understanding of complex non-linear relationships between input variables and conversion.
Additionally, Clarkston’s team developed recommendations around long-term strategies for success. These included: identifying future opportunities around optimizing call center resources, translating findings to a customer facing version of the model, opportunities around incorporating patient-centric data, and a maintenance/testing/performance validation framework.