Work
Five problems, and what closed them.
Client names are left out on purpose. Where a number would identify an account, I've used a proportion instead.
Two days of business review prep, done in thirty minutes
Executive business reviews were costing one to two days each, so the team could only run about three a week. The bottleneck was never the slides.
Prep time cut by roughly 95%The account that was ready to leave, twice
Close to a third of my book, sitting at 41% adoption, with a renewal coming. Part of the resistance turned out to be correct.
Adoption 41% to 75% · ~7x ROI · saved twiceOur metrics weren't showing leadership the value of customer success
Sales were strong and revenue was still going backwards. Rebuilding the function changed what the business chose to fund.
94% CSAT · team of 11 across three marketsA churn play every CSM would actually run
Good playbooks existed. Nobody ran them, because under pressure people fall back on instinct.
Churn down 18% in two quarters · 107% NRRThe handover between sales and customer success
The most expensive gap in most SaaS businesses, and the one nobody owns. Still being worked on, which is the honest answer.
Gap analysis · in progressAnalysis projects
Working with the data directly
Separate from the work above. These are modelling exercises rather than company outcomes, and I include them because being able to build the model matters as much as being able to read one. Both notebooks are published in full, including where the results were not good enough.
Predicting user churn on Waze
Nearly 15,000 users, twelve variables. Logistic regression, then random forest and XGBoost. Activity days turned out to be by far the strongest signal. The honest conclusion in the notebook is that recall was too poor for the model to be trusted, and that accuracy is a misleading measure on an imbalanced dataset.
Full notebook, PDFPredicting employee attrition
An HR dataset, asking what makes someone leave. Satisfaction scores clustered around 0.4 for leavers, a strange anomaly at four years of tenure, and most of the company working well over 167 hours a month. Same discipline as churn work, pointed at people instead of accounts.
Full notebook, PDF