PredictIQ: Churn Prediction ML Pipeline
Machine Learning
Project Overview
An end-to-end ML pipeline that predicts customer churn 30 days in advance for a subscription SaaS business — with automated retraining, drift detection, and a CRM integration that surfaces at-risk accounts directly to the sales team.
Client: Concept Build — CloudServ Analytics
Duration: 7 weeks
The Challenge
The client was losing roughly 8% of customers monthly with no early warning system. Customer success teams were reactive — only noticing churn after cancellation. They needed a proactive, data-driven approach integrated into their existing CRM workflow.
Our Solution
Built a feature engineering pipeline in Python that pulls 90 days of product usage, billing, and support history from PostgreSQL. XGBoost model trained on 18 months of historical churn data — MLflow tracks all experiments, model versions, and metrics. Apache Airflow orchestrates weekly automated retraining and data drift detection. The FastAPI prediction service pushes a 30-day churn probability score to HubSpot via API, creating at-risk tasks for the customer success team on any account above threshold. Full pipeline containerised with Docker on Kubernetes.
Results
- Model achieved 87% precision and 82% recall on holdout test data
- Monthly churn reduced by an estimated 23% within the first quarter of deployment
- Customer success team received automated HubSpot tasks for at-risk accounts daily
- Automated retraining with drift detection ensures model accuracy is maintained over time
Client Testimonial
"Alicorn built the LiteCloud practice management platform for us — it handles our entire client workflow, task tracking and reporting. It has been running in production for over 8 months and the team has been responsive throughout. The co-founders are directly involved, which makes a real difference."
Lee Phillips
Digital Data Lead, Twinings Ovaltine
Ready to Start Your Project?
Let's discuss how we can bring your ideas to life and drive your business forward.
Get in Touch