Google Analytics 4 and Meta Ads: Reconciling Different Numbers

Different totals can be correct answers to different questions. Start with settled business records, then check whether each platform received an event, then inspect its attribution rules. Do not add GA4 purchases to Meta purchases to calculate business sales.

A worked reconciliation: demo-1042

This is a hypothetical order, not customer data. Assume an eligible Meta ad view at 10:00, a tagged newsletter click at 10:10 and a $100 payment at 10:20. Assume the required consent, successful collection and identity matching. The following day a $20 partial refund settles. Use the same currency and time zone throughout.

The ledger still contains one order and $80 retained revenue. Two platform credits do not mean two sales. A Meta view does not require a Meta UTM on the later email visit. Conversely, seeing a UTM does not prove Meta received a Purchase or caused the sale. Google explains attribution models, purchase and refund event fields and reporting identity. Record the actual model and window rather than assuming account defaults.

Negative scenarios and expected results

What NovaStorm has actually tested locally

The documented September 15, 2026 engineering check covered email and mocked Google signup, first and last tagged touches, a direct return, and an immutable registration snapshot in an isolated PostgreSQL database. Its verification record reports 19 backend integration tests, 30 frontend tests and five browser checks passing. Deployment and production migration were explicitly not performed at that documented check; these results do not establish current production behavior.

On October 2, 2026, the existing web-core registration-attribution.test.ts was run with npm run test -- src/lib/__tests__/registration-attribution.test.ts: all 10 tests passed. That run checks browser-storage behavior with mocks, including tagged touch preservation across a direct return and module reload, expiry, consent/GPC denial, sanitization and bounded OAuth transport. It does not test production purchase emission, Meta CAPI delivery, refunds, Google attribution or advertising lift.

Acceptance before relying on the data

1. In a controlled environment, create one uniquely identified test order and a partial refund. Record ledger settlement, currency, event time and each authorized event receipt without exposing customer details.
2. Verify first and latest tagged touches, a direct return and denial of marketing consent. Verify returning-user updates cannot replace the original registration snapshot.
3. Check purchase and refund identifiers independently from registration attribution. Confirm the implemented browser/server path receives and deduplicates exactly the intended event.
4. Compare the selected GA4 model and Meta window after reporting delays. Leave unknown identities and unmatched rows visible; do not force totals to agree.
5. Repeat the agreed acceptance procedure after release with authorized test data and an owner. Treat production acceptance and any incrementality experiment as separate work.

Use a mature acquisition cohort with payment, refund, event and channel columns, then review disagreements individually. The attribution-window guide explains why changing a window changes credit. Budget decisions still need contribution margin and evidence beyond a preserved source snapshot.