Customers at risk this week
Each row is one active customer, highest risk first:
Show the Top 25, Top 50 or All, and use Find a customer to search by name or id. A customer with nothing pushing their risk up shows “No single strong reason”.
How good is this model
Next to the list, the model’s results on the most recent weeks, which it never saw while training:- AUC on unseen weeks, with a verdict: good, fair, weak, not enough history, or too good to be true. See quality numbers.
- Top 10% flagged and Top 50 flagged: of the customers ranked highest, how many really left within the window.
- Learned from: how many customers, how many of them left, over how many days.
- Tested on: the customer-weeks and dates the numbers come from, and the date training stopped.
What drives risk here
The features that mattered most to this workspace’s model, across all customers.Export and retrain
- Export CSV downloads the whole list with every column.
- Retrain trains the model again on the data in the workspace.

