Stratified sampling
Also called stratified random sampling, sampling by segment
Drawing a random sample from each group separately, such as each document type, so small groups get checked as well as big ones.
A plain random sample mirrors the whole pile, so a group that makes up 3% of it shows up about 3% of the time, and some weeks not at all. Stratified sampling splits the pile into groups first and draws from each one.
For an AI pipeline, it's how you keep measuring the output you no longer check by hand. Sampling every document type every week gives each one an error rate of its own, and catches a new kind of mistake in a small group before it has piled up.
For example
Instead of reviewing 100 records picked from everything, a team reviews 10 picked at random from each of its 10 document types, so the customs forms that make up 3% of volume are always checked.
Put it to use
Where you'll meet stratified sampling in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.