Data quality
Delivering an event is the easy part. Getting it matched, counted once, and accepted by the platform is where server side tracking succeeds or quietly fails.
Almost every data quality problem in Signals falls into one of four buckets. Each has its own page.
The four problems
Section titled “The four problems”| Problem | What goes wrong | Page |
|---|---|---|
| Matching | The event arrives but the platform cannot tie it to a user, so it cannot be used for attribution or optimization. | Match quality & EMQ |
| Formatting | Identifiers are hashed inconsistently, so the hashes never match even though the underlying data is correct. | Identifier normalization & hashing |
| Double counting | A browser copy and a server copy of the same conversion are counted twice, inflating results. | Deduplication |
| Rejection | The event is malformed or incomplete, and it is dropped rather than delivered. | Event validation & schema |
The failures are usually silent
Section titled “The failures are usually silent”This is the part worth internalizing. Very few data quality problems announce themselves.
- Meta treats an event as invalid where the customer information is only geography, gender and date of birth. Entire event streams get discarded with no error surfaced to you.
- A phone number hashed without its country code produces a valid hash that simply never matches.
- An event that fails validation is dropped and the rest of the batch continues, so nothing looks broken.
- A source column renamed or cased differently stops being recognized, and the rows carrying it are dropped.
In each case the system reports success. The numbers are just lower than they should be, and nothing tells you why.
What to check, in order
Section titled “What to check, in order”- Are events arriving at all? Check the platform’s own event tool, not Studio.
- Are they being matched? Check the platform’s match quality metric.
- Are they being counted once? Check deduplication overlap.
- Are required fields present and correctly typed? Check validation.
Work in that order. There is no point tuning match quality on events that are being dropped.
What Signals reports on
Section titled “What Signals reports on”Volume and delivery counts are recorded for every destination and for batch sources: how many events came in, how many went out, and how many failed or were skipped.
Deeper per-identifier quality reporting, including Event Match Quality, is scoped to web conversions through the Meta Conversions API Gateway.
See Monitoring for what that covers and what it does not.