Meta Conversions API: Event Match Quality Is Dropping
This document describes a specific diagnostic path when in Event Manager Meta you observe the downgrading of Event Match Quality (EMQ) after switching some traffic to Conversions API (CAPI) or after changes in implementation. EMQ depends on the quality and completeness of user identification data (e.g. email, phone, client ip address, fbp, fbc) and how these fields are to be hashed and transmitted. The quality decline is usually due to one or a combination: missing/incompatible identification field, wrong hashing (PHP/JS vs SHA256 raw bytes), non-compliance time stamp (differences >24h), duplication of events without proper deduplication (pixel + CAPI without event id) or filters and privacy rules in the advertising panel that cut or map fields. Below you will find an orderly differential diagnosis with step-by-step tests, specific commands in Meta logs and tools, and ways of verifying effect after corrections. We do not assume the cause — every point must be checked and eliminated sequentially.
How to Recognize Cause
Compare Pixel and CAPI paydays: check that EMQ compliant fields are present in server requests. Usually missing email hash, phone hash, client ip address or user agent. If the fields are sent from the client via the server, note which ones come from the cookie and which from the payment form.
Check the hashing process: Meta expects SHA256 in hex format for personal IDentifiers. Typical errors: double hashing, base64 instead of hex, hashing with extra salt or wrong character encoding (UTF-8 vs Latin-1). Log in examples of inputs and compare with sha256 online.
Identify problems with time stamp and event id: if event id or event time vary between Pixel and CAPI, the deduplication will not work and Meta may treat events as separate, which affects EMQ. Also check transmission delays (>24h) and buffering, which reduce the accuracy of the match.
Step by step solution
1. Pull out 50 random events from the Event Manager from the problem period; download the example of Payload Pixel (from the browser) and the corresponding payload CAPI (from the server) and save both in the JSON file for comparisons.
2 . For each identification field compare the values: are the email/phone/fbp/fbc/client ip address present in Pixel also in CAPI? If not, open the server code and check whether middleware passes the headers and body requests from the client.
3 . Implement the hashing test: take the original data (e.g. raw email), count SHA256 in hex on the server and compare to the value sent to Meta; if different, repair the hash function and ensure uniform UTF-8 encoding.
4 . Verify event id and event time: Force the sending of an identical event id from Pixel and CAPI in the test example; if pixel does not generate event id, generate a compatible identifier at the client and send it to both sources.
5 . Check delays: Measure the time between the event in the browser and the CAPI request; if >5–10s check queues, batching and retry; if >24h, check Joby cron and ETL, because such delays reduce EMQ.
6 . Browse the privacy configurations and field mapping in Event Manager: make sure that the fields are not mapped to custom labels or filtered by privacy rules that could cut the email/phone.
7 . Enter one amendment at a time (hashing or sending event id) and observe EMQ changes within 24–72 hours; document each deploy and implementation time.
How to see if the problem is gone
Test 1: Send 20 controlled test events with unique event id and complete hashed IDs; in Event Manager check that at least 80% has 'High' or 'Good' EMQ within 24–48 hours.
Test 2: Compare the matches indicator between Pixel and CAPI for the same event id; expect the number of properly matched matches to increase after the hashing/accession amendment by a minimum of 30% from baseline.
What to do
Do not simultaneously implement many changes (hashing, event id and delays). This makes it impossible to indicate which amendment has improved EMQ; introduce incremental changes.
Do not assume that the lack of email means a server problem — check the consent policy and frontend management, because the user can block the transfer of data.
Do not log full PIIs into production logs. Log only shortcuts/test identifiers and use masking to avoid breaking privacy rules.