The conversion tracking audit: fifteen checks before you spend more
Audit tracking by firing each conversion yourself and confirming it arrives exactly once in GA4, the ad platforms and the CRM. The common faults are events firing on page load, the same conversion counted twice, and no server-side events, which loses a meaningful share of conversions.
Key takeaways
- Fire every event manually and verify it in every destination. Assume nothing.
- Deduplicate server and browser events with a shared event id, or ROAS doubles.
- Reconcile against the CRM monthly, not once at setup.
- Fix the signal before increasing budget — it changes what the algorithm finds.
Almost every account we audit is optimising against a number that is wrong. Not slightly wrong — structurally wrong, because the conversion counts page views, or double-counts, or misses a fifth of purchases. Every decision on top of that is a guess with a chart attached.
The audit
Fifteen checks
- Every conversion fires on a completed action, not on page load or a thank-you page reached by refresh.
- No duplicates — one tag per conversion, not a Tag Manager tag *and* a hard-coded pixel.
- Server-side events exist for the main conversion, deduplicated against the browser by a shared event id.
- Event match quality checked in Meta Events Manager; low scores mean too few customer parameters.
- Enhanced Conversions enabled in Google Ads.
- Only real outcomes marked primary — micro-conversions should not train the bidding.
- Offline conversion import configured where the sales cycle is longer than a fortnight.
- Call conversions have a minimum duration, so a ten-second wrong number is not a lead.
- Consent mode configured, so tracking is lawful and modelling still works.
- Cross-domain tracking verified if the funnel crosses a domain — checkout, booking tool, subdomain.
- Attribution windows documented, so nobody compares two numbers built on different ones.
- Form submissions reach the CRM, verified by submitting one yourself.
- Source and campaign persist onto the CRM record, captured at submission.
- Bot and internal traffic filtered, including your own office IP.
- A monthly reconciliation between platform, analytics and CRM totals, with differences explainable.
Why this returns more than creative work
It changes what the algorithm optimises for. Send back qualified leads rather than raw form fills and the same budget starts finding a different population — reliably, and without touching a single ad. The mechanics are in server-side tracking setup.
How much traffic you need to test
Roughly two hundred conversions per variant within a month for an A/B test to conclude. Below that, fix the obviously broken things and watch ten session recordings instead — usually more informative than a test that never reaches significance. The full engagement is on Conversion & Analytics.