FromRetention deskReHargrave, renewal of relationship
No draft yet. The page is intentionally quiet until you ask for one. Sending the prompt to the model takes a moment and costs a small amount of tokens.
Each model gets the same inputs but reaches its own probability. The spread tells you how confident the consensus really is.
XGBoost typically anchors the high end of this committee; the simpler classifiers pull the average back toward the centre.
Feature engineering and class-rebalancing tend to sharpen these estimates, for better or for worse on any given customer.
Move any slider, flip any switch. The probability and the attribution chart update live. Use it to ask, “what would it take to bring this customer back into the green?”
Pressing “Draft retention email” sends the prediction and a short explanation to a small language model and asks for a letter to keep this customer.
Generated by Llama 4 Scout via Groq. Treat as a draft.