Why your ad platform, GA4 and your bank account never agree — and how to fix it

Ad platforms count conversions against the click or view they believe caused them, inside their own lookback window, and they model the conversions they cannot observe. GA4 counts sessions it can stitch together, with a different attribution model and a different window. Your finance system counts money that actually cleared, net of refunds, chargebacks and failed payments.

These are three honest answers to three different questions. The failure is not that they disagree — it is that most teams never decide which one is allowed to drive a budget decision.

Before touching any tag, write down the source that wins every argument. For ecommerce it is the order table. For lead generation it is the CRM opportunity, not the form submission. Everything else becomes a signal used to allocate, never to report.

Publish this decision to the whole team. Most reporting disputes are really governance disputes wearing a technical costume.

Attach an order or opportunity ID at the moment the record is created, and pass that same ID into every downstream event: the browser event, the server-side event, the offline upload, and the finance export. Deduplication only works when both sides speak the same identifier.

Browser-only tracking loses events to ad blockers, consent rejections, ITP truncation and slow pages abandoned mid-fire. Sending purchase and qualified-lead events from your server recovers a meaningful share of that loss and makes the payload trustworthy, because it is built from your database rather than from a data layer that a theme update can silently break.

Keep the browser event as well. With a shared event ID, the platform deduplicates and you get the best of both signals.

Consent mode changes what platforms observe and how heavily they model. If your banner defaults are inconsistent across pages, your data will fluctuate for reasons that have nothing to do with marketing performance. Audit the banner on every template, confirm the default state, and log consent rates as a metric you actually watch.

Build one small table: revenue per channel from the system of record, revenue per channel as reported by each platform, and the percentage variance between them. Review it weekly for four weeks and the pattern becomes obvious — one channel over-reports consistently, another under-reports, and the rest are noise.

Once you know each platform's bias, you can keep optimising inside the platform while reporting the reconciled number upward. That is the whole point: platforms are for steering, the system of record is for deciding.

Teams that run this model stop arguing about dashboards and start arguing about strategy, which is a far better use of a Monday. Budget moves are defensible, channel comparisons are fair, and the finance team stops treating marketing reporting as fiction.