Is the simplest way to measure how often two or more people labeling or reviewing the same data land on the exact same answer, expressed as a straightforward percentage of matching decisions out of the total compared. It's calculated by dividing the number of items where reviewers agreed by the total number of items reviewed.

‍

‍Why it matters

It's popular because it's easy to calculate and even easier to explain to someone outside a data team, a 90% agreement rate is intuitive in a way a more complex statistic isn't. The tradeoff is that percent agreement doesn't account for agreement that would happen purely by chance, which matters most when there are only two possible labels, since two random reviewers could agree a good portion of the time without applying any real judgment consistently at all.

‍

‍Why teams often pair it with something else

Because of that chance agreement issue, percent agreement is usually reported alongside a chance-corrected metric like Cohen's Kappa, rather than used entirely on its own for anything high stakes. It still has a real place though, quick daily QA checks, or communicating results to a non-technical stakeholder, where the simplicity is actually the point rather than a weakness.