A churn prediction model has been in production for 14 months. It was deployed at 0.84 AUC on the holdout set and held around 0.82 in production for the first year.
Over the last 10 weeks, production AUC has declined steadily to 0.71. There was no deploy, no code change, and no model retrain in that period. The prediction volume is unchanged. Infrastructure metrics are all normal.
The business is asking whether to retrain, and you have been asked for a diagnosis first.
Walk through how you would find out what actually happened, what you would check in what order, and what you would do about each possible cause. Be specific about the distinctions you are drawing and how you would test them.
Build an interview-ready technical plan: frame the problem, explain the ordered approach, name the trade-offs, and show how you would validate the outcome. The AI reviewer grades that reasoning against this problem’s rubric.
Four guided sections · works on any screen