Match rate

You send a thousand customer records to a platform. It recognises some of them and cannot place the rest. The proportion it recognised is the match rate.

It is the most useful diagnostic in the identity stack because it fails loudly in a way almost nothing else does. A hashing normalisation mismatch, a field sent in the wrong format, a missing identifier — none of those raise an error. They show up here, as a number lower than it should be.

What actually moves it:

• Normalisation — the biggest single cause of an unexpectedly low rate, and the easiest to fix.
• Which identifiers you send. More signals give more chances to match, subject to what you are permitted to send.
• Overlap. A platform can only match people it knows. Audiences skewed away from that platform match worse, and no amount of engineering changes it.
• Consent. Records you may not send are not in the numerator or the denominator, and that alone can make two campaigns look different.

The two mistakes. First, treating it as a quality score for your data rather than a measure of overlap plus formatting — a low rate can mean your file is broken or simply that this platform does not know your customers. Second, comparing match rates across platforms, which compares their user bases more than your setup.

Where it stops being about data. A match is not a conversion and not a person seeing an ad. It only means two systems agree a record refers to the same someone.