Match rates dropped after a change
A match rate that falls after a release, a migration or a form redesign is usually telling you that a field stopped arriving or stopped being formatted the way it was. The integration itself rarely changes on its own.
Start by establishing the date it changed, because that is what tells you which change to look at.
The usual causes
Section titled “The usual causes”| What changed | How it shows up |
|---|---|
| A column was renamed or removed in the source | The mapped field stops arriving. Coverage for that identifier falls to zero while everything else looks normal. |
| A form field was made optional | Coverage falls gradually rather than all at once, because some people still fill it in. |
| A site release broke the dataLayer | Several identifiers stop arriving together, from the date of the release. |
| The tag manager container was republished | A tag can be dropped or reverted in the process. The user-provided data tag is the one that matters here. |
| The phone number format changed at source | The field still arrives and stops matching, which looks worse than it being absent because nothing appears to be missing. |
| A normalization step was dropped from the export | Same effect. The hash is present, well-formed and matches nothing. |
| The destination dataset or audience was switched | Deduplication and matching both reset, because they are scoped to the dataset. |
| A token was rotated or an account was reauthorized | Delivery may have paused before it resumed, leaving a gap that reads as a drop. |
| Consent collection changed on the site | Fewer events carry identifiable data, by design. The drop is real and the integration is behaving correctly. |
How to isolate it
Section titled “How to isolate it”- Fix the date. Find the last day at the old rate and the first day at the new one, then list what shipped in between.
- Compare coverage rather than the rate. Take the share of records carrying each identifier before and after. One identifier dropping to zero points straight at a schema or tag change; all of them falling together points at the source.
- Take one record and follow it end to end. Compare the value in your system with the value in the export, and check the normalization is still being applied in the right order.
- Run Test Connection on the instance. It checks that the mapped fields still contain data, so a mapping pointed at a field that no longer exists fails here rather than silently.
- Check the mapping in Studio. A field that disappeared from the source may have quietly unmapped itself, and the mapping list only offers fields that are actually arriving.
If nothing changed on your side
Section titled “If nothing changed on your side”Check whether the volume changed rather than the rate. A rate calculated over far fewer events moves around more, and on low-volume events a destination may stop reporting a score at all. Compare absolute numbers alongside the percentage before treating it as a regression.
Still stuck
Section titled “Still stuck”Contact support with the date the rate changed, the project and instance, and what you know shipped around that time. Include a sample of the current export if you can, since most of these are visible in one.