Most churn tools only answer "who is likely to leave" — but knowing who leaves doesn't stop them. I wanted to catch the moment of hesitation itself, so I built ShopLite, an e-commerce platform with a retention copilot. It watches real behavioral signals — carts left un-checked-out, products abandoned, purchase frequency dropping — and turns them into a risk profile so a business understands why a customer is drifting. The heart of it is an AI assistant powered by Google Gemini that jumps in at the exact moment a customer hesitates: when they log in with a full cart or add an item and don't check out, it reads their actual cart and the live catalog, greets them in friendly, personalized language (in RM), and offers a "Proceed to checkout" button to guide them back. An admin-facing Monitoring Dashboard ties it together with KPIs, a churn trend chart, and a retention table showing each customer's risk and the reasons behind it. I learned that the strongest churn signals come from what customers do rather than what they report, and that a helpful, specific conversation beats a generic discount. It wasn't without bumps — "abandoned" turned out to be a fuzzy signal, a slow AI response once leaked between accounts (fixed with a session guard), and the model gave me a retired endpoint and a rate limit mid-demo — but making the bot resilient and the data believable turned it from a prediction tool into a real intervention: not just who leaves, but a reason to stay.
Built With
- java
- postgresql
- render
- spring