Skip to content
ML Academy

Experimentation and causality

Interview question

  • stafflevel
  • 25 minsuggested time
  • 2Data scientist, ML engineer

A team wants to A/B test a new dynamic pricing algorithm by randomising users: half see fares from the new algorithm, half from the old one, on the same flights. What is wrong with that design, and how would you test the algorithm?

Tests: Interference (violations of the no-interference assumption) in shared-inventory systems, alternative designs (cluster, switchback, market-level), bias vs variance trade-offs, and practical constraints including fairness and customer perception.

day 4day 5day 16

Sign in to practise: answers are saved before model answers are shown.