Subscription and SaaS businesses
In subscription businesses, churn is a silent revenue killer. This predictive module turns churn management from a reactive scramble into a proactive strategy: machine learning spots at-risk users early, and generative AI writes personalised save-offers so customer success teams can step in before the cancellation happens.
0.84
ROC-AUC on 1,409 customers the model never saw (0.5 would be a coin flip)
55%
of customers who really cancelled were caught, and 64% of the customers it flagged really left
3
models trained and compared: Logistic Regression, Random Forest and XGBoost
For SaaS and subscription providers, losing users after just a few months badly undercuts profitability. Keeping a customer is always cheaper than winning a new one, yet most businesses only react once the customer is already out the door.
After cleaning thousands of raw customer records, Logistic Regression, Random Forest and XGBoost models were trained and tested to find the most accurate way to separate loyal users from flight risks.
The winning model gives every active customer a churn probability score, pushed automatically into Google Sheets as a fast, lightweight operational backend.
The highest-risk profiles go to Gemini 2.5 Pro, which acts as a strategic advisor and proposes a personalised retention move, such as a targeted discount or a plan adjustment.
Everything is wrapped in a Streamlit app on Streamlit Community Cloud, where marketing and customer success filter high-risk cohorts and export action-ready CSVs in seconds.
The system protects monthly recurring revenue and raises customer lifetime value.
Portfolio project · Public dataset (Kaggle)