Plain English
Evals
Tests that check how well an AI does your task, using real examples and a clear bar for “good.”
Short for evaluations: a set of real test cases with known good answers that you run an AI tool against, before launch and after every change. For a small business, 20 real examples in a spreadsheet is a good start.
Why it matters
Without evals, “it seems fine” is the only test, and quality can slip unnoticed when a model or prompt changes. With them, you can compare tools and catch problems early.
At work
Before switching on AI-drafted replies, a support lead runs 20 of last month’s emails through it and marks each draft: send as is, fix or rewrite.
Think of it like a driving test for your tool: the same route and the same checks, every time, before it’s allowed on the road.
Related terms
Where it appears