Wei Zhang
Senior Data Scientist | Pricing, Churn and Forecasting Models in Production
October 30, 2026
Dr. Ruth KaplanVP, Data ScienceEvergreen MutualSeattle, WARe: Senior Data Scientist, Retention and Pricing
Dear Dr. Kaplan,
Most churn models stop at a score. Your Senior Data Scientist posting at Evergreen Mutual asks for someone who can turn that score into a retention program with measurable results, which is the part of the work I enjoy most.
At Harborline Insurance I built a renewal churn model with an AUC of 0.81, but the real work was deciding how to use it. With the retention team I set up a randomized test of offers by risk band, and we found that discounts only paid off in the middle band. Targeting that group lowered the lapse rate by 2.4 points and now retains about $3.1 million in premium a year. The test framework I built for that project has since been used for 14 pricing experiments.
I also work closely with actuaries. Adding telematics features to our auto pricing model improved loss ratio prediction by 9% on holdout data, and I moved six models from notebooks into scheduled, monitored pipelines on Databricks so they no longer depend on one person running them.
Before insurance I spent nearly three years forecasting electricity demand for four utilities at Cascade Energy Analytics, with a day-ahead error of 3.2%, and presented model results to regulators in three rate cases. That experience helps in any regulated pricing work, where a model has to be explained as well as accurate.
I bring 7 years of data science experience. I would be happy to walk you through the churn project in detail, including what did not work.
Best regards,
Wei Zhang