The problem
Executive questions about pricing, churn, and account mix cut across CRM, CPQ, and service-order data that live in different systems with different grain. Answering them by hand produces a different number each time someone asks.
Leadership needed analyses they could act on for pricing strategy, pipeline visibility, and revenue planning, and dashboards they would trust enough to run meetings from.
What I did
-
Modeled the data in SQL against Snowflake and MS SQL Server so CRM opportunities, CPQ quotes, and service orders join cleanly, with definitions agreed up front for what counts as churn, a price increase, or a given service class.
-
Built the analyses the executives asked for: price-increase impact, churn, and service-order classification for major accounts such as Kaiser Permanente, then turned the recurring ones into Power BI dashboards.
-
Went past descriptive reporting where the question needed it. Estimated price elasticity with regression on the order-line panel using robust standard errors, and modeled win rate against relative price with logistic and Poisson GLMs clustered by facility, so pricing conversations had an effect size and a confidence interval instead of an anecdote.
-
Built an opportunity-tier classifier in LightGBM with cross-validation grouped by facility, so the score generalizes to accounts the model has never seen. Paired it with a review app and a label store so analysts correct the model where it is wrong, and the corrections become training data.
-
Designed the dashboards for the people reading them: the CEO, CFO, and VP Sales Operations each get the view that matches the decision they own.
- Model the data SQL over Snowflake and SQL Server joins CRM, CPQ, and service orders.
- Train LightGBM, cross-validated with folds grouped by facility.
- Score Every open opportunity gets a tier.
- Review Analysts correct the model where it is wrong, in a review app.
- Store the labels Corrections land in a label store, not a spreadsheet.
Corrections become training data on the next run
Outcome
Executive analytics are now a standing part of how pricing strategy, pipeline, and revenue planning are discussed, with one agreed set of numbers.
The classifier agrees with human labels on the large majority of unseen facilities, and the review loop keeps it honest as the business changes.
The same data layer powers program reporting such as the Technician Opportunity Program, feeds the inputs to the pricebook automation program, and is the history the revenue forecasting ensemble trains on.
In the stack
Where it sits
The highlighted nodes are the systems and people this project touched.
Skills used
- SQL
- Power BI
- Snowflake
- MS SQL Server
- Python (production automation, 6,000+ lines)
- Data modeling & analytics
- Econometrics & statistics (regression with robust and clustered errors, GLMs)
- Supervised ML with human-in-the-loop labeling (LightGBM, scikit-learn)
- Revenue operations & pipeline analytics
- Executive stakeholder management
- Pricing strategy & CPQ workflow design