Standard engagement
Churn Prediction Model
Know which customers are about to leave — while you can still save them.
A churn-risk model trained on your usage data + billing data + support history. Scores every customer weekly, surfaces the top-N at-risk accounts to your CSM team, and explains the score so they know what to actually do about it. No black box — every prediction is explainable.
from $3,800
4 weeksPythonXGBoostSHAPSnowflakeBigQuerydbt
What ships during the engagement.
Feature pipeline (usage + billing + support + tenure)
Trained model with hold-out evaluation
SHAP-based explanations for every prediction
What you walk away with.
- Weekly churn score for every active customer
- Top-N at-risk list delivered to CSM Slack channel
- Explanation per score (what is driving the risk)
“They scoped, shipped, and operated our RAG pipeline in twelve days. Citation accuracy on our eval set landed at 92%, and ongoing tuning costs us less than a Slack seat.”
- How accurate is it?
- Depends on signal quality. Typical: 75–88% precision on top-decile risk accounts. We benchmark against a 30-day hold-out before go-live.
- How much data do we need?
- Minimum: 12 months of customer data and at least 50 churn events. Less than that and the model overfits.
Want to scope Churn Prediction?
A short call to confirm fit and timeline.