A fraud model's AUC is stable but losses are up. How do you investigate concept drift versus a threshold or product change?
PICTURE THIS: RAG CHATBOT
Simple meaning
I would freeze the score distribution, check whether business thresholds or coverage changed, and look at precision at the operating point on delayed labels.
WHY — Concept Drift instead of guessing?
Why interviewers care about Concept Drift:
contrast on Concept Drift, not two memorised paragraphs.
the developer, then one case where picking wrong hurts.
Name the idea, why it exists, then one short example.
End with when you use it and one common pitfall.
STEPS — What happens step by step?
Before you speak the answer, walk the interviewer through these steps:
- 1I would freeze the
score distribution, check whether business thresholds or coverage changed, and look at precision at the operating point on delayed labels.
- 2If scores still rank
well but the cost of false negatives changed, that is a decision-policy issue.
- 3If calibration collapsed on
new fraud types, that is concept drift and needs new labels and features.
- 4Give an example
One tiny concrete case you can say aloud.
- 5Common mistake
What juniors usually get wrong.
- 6Close
When you pick this over the alternative.
EXAMPLE — See it in action
Here's a short line you can speak, broken into clear beats:
Note: Adapt this scaffold to your own project — keep it under 60–90 seconds.
Key takeaway
I would freeze the score distribution, check whether business thresholds or coverage changed, and look at precision at the operating point on delayed labels. If scores still rank well but the cost of false negatives changed, that is a decision-policy issue.