Customer Churn Prediction Model
Client: B2B SaaS - 50K users
The project
Development of a 90-day customer churn prediction model with 89% precision. The model analyzes more than 200 behavioral features: login frequency, feature usage, support tickets and account health.
Architecture & implementation
Implementation of a complete MLOps pipeline with MLflow for experiment tracking, model versioning and automatic monthly retraining. Integration with Salesforce CRM to trigger targeted marketing actions.
Monitoring & adoption
Predictive dashboard for the Customer Success team: prioritized list of at-risk customers, prediction drivers (SHAP values) and recommended actions. 100% team adoption within 2 weeks.
Results & production rollout
Results measured over 12 months: 35% lower churn rate, NRR up 18 points and an estimated €2.4M in recurring revenue retained. The model is retrained every month to maintain its performance.
Screenshots
