Dealership performance data
A private United States dealership dataset was prepared around advisor activity, customer-pay work, warranty work, repair orders, labor, and profit.
AI-enabled decision systems
Explainable Service Advisor Analytics
Can performance analytics show automotive dealerships not only who needs support, but why?
Published in the Journal of Retailing and Consumer Services, 2024
Why I built it
How can interpretable regression and clustering uncover distinct service-advisor performance patterns from real dealership data?
Service advisors translate customer needs into repair work and play a central role in dealership profitability. Simple rankings can identify high and low performers, but they do not explain the operational patterns behind those differences.
This project evaluates five regression approaches within finite-mixture models, clusters advisors into distinct performance groups, and uses SHAP to reveal how service measures influence predicted profit. The result supports targeted coaching, benchmarking, and resource allocation.
The dealership workflow maps how a service advisor receives a vehicle, translates customer concerns into repair work, coordinates technicians, and closes the service interaction.
How the system works
Mixture regression analysis
Five regression families were evaluated within finite-mixture models to represent heterogeneous relationships that a single global regression would obscure.
Explainable clusters
Gaussian mixture clustering, information criteria, silhouette analysis, and SHAP connect each performance group to actionable operational factors.
5regression approaches evaluated
3model-selection criteria
SHAPcoefficient-level explanations
Real datafrom a US automotive dealership
What this work contributes
From a research question to an inspectable system.
- A performance framework designed around the heterogeneity of human service work.
- Distinct advisor clusters that enable targeted development rather than one-size-fits-all evaluation.
- Comparison of regression approaches under shared finite-mixture conditions.
- SHAP explanations that show how operational variables move predictions within each group.
- A pathway from predictive analytics to coaching, benchmarking, and dealership decision support.
How I read the results
What the evidence means
Performance groups become useful when their differences can be explained. High-performing clusters can reveal transferable practices, while lower-performing clusters can receive support tailored to the factors that constrain them. The study also shows where quantitative models struggle to capture complex human performance.
Current scope
What this study validates
The analysis uses a private dataset from one United States automotive dealership. Its value is the interpretable methodology and decision framework, while broader validation would require data across additional dealerships and operating contexts.
Inside the system
More evidence from the research.
