Production AI evals at the world's largest used camera marketplace

The challenge

MPB is the world's largest platform for buying, selling, and trading used cameras and photography equipment — operating across the UK, US, and Europe. Their product team had built AI-powered listing tools, but without clear success criteria or evaluation frameworks it was difficult to measure whether model changes were actually improving outcomes — or silently introducing regressions into listings at scale.

What we did

We ran a custom 2-day AI evals program with MPB's product and ops team. The cohort learned how to define success criteria for AI features, build evaluation datasets, score model outputs systematically, and create feedback loops that continuously improve AI performance over time.

Sessions combined live instruction with hands-on work inside MPB's actual product environment — so everything learned was immediately applicable to real listings and workflows.

The outcome

The team came away with a working eval framework, a set of reusable evaluation datasets, and the confidence to own AI quality assessments without relying on external consultants. Listing automation accuracy improved as a direct result of better feedback loops.