Practical technical writing from the work of building production models that turn existing business data into better decisions.
The modern tech stack is not a software list. It is data, talent, compute, and a sales team that works the list. What breaks when one is missing.
Most companies do not need a new platform to get value from ML. Train where the data lives, score on a schedule, and write results into the CRM.
A churn score alone changes nothing. How to separate drivers you can act on from ones you cannot, give each an owner, and prove the save play works.
Seasonality, concept drift, and quiet data breaks. Why forecasts that tested well go stale, and how to build ones your planners can keep using.
Beyond AUC: calibration, lift at the cutoff you will actually use, expected value in dollars, and drift monitoring. Judge a model by the decisions it drives.
More data is not the goal. Recording outcomes, the leads you skipped, and what you knew at decision time is what makes lead and churn models better.
What actually predicts in lead and customer datasets vs what looks good on a dashboard. How to shape CRM fields so scores change who gets called.
Owning data is not the win. Using it is. How fitted churn, lead, forecast, and targeting models change the weekly rhythm.
Spirit is shut down, and Google still agreed to pay $10M for its enterprise data. Why your operating history is the real moat.
In an AI world, your model is only as good as its training data. Why dirty CRM records ruin lead scoring and churn models.
Planned themes for future notes.
Explore our models or talk to us about fitting lead scoring, churn prediction, and forecasting to your data.