How do you randomize users for a model A/B test without leaking?
PICTURE THIS: DATA SPLIT
Fit on train, tune on val, report on test once.
Simple meaning
Hash a stable user id with a salt into buckets so the same user always sees the same model.
WHY — A/B Testing instead of guessing?
Why interviewers care about A/B Testing:
question about A/B Testing.
trade-offs, and what you would actually do on a MLOps project - not buzzwords.
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:
- 1Hash a stable user
id with a salt into buckets so the same user always sees the same model.
- 2Do not randomize per
request or users will flicker and metrics get noisy.
- 3Keep assignment logs so
you can analyze intent-to-treat later.
- 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
Hash a stable user id with a salt into buckets so the same user always sees the same model. Do not randomize per request or users will flicker and metrics get noisy.