Add Meta, Google, and TikTok's reported conversions and you will typically count 1.5-2.5x your actual sales. Every platform grades its own homework — their dashboards are sales documents with charts. Advanced measurement is not better dashboards; it is a reconciled system with an experiment layer. Here is the stack we implement for clients who report to CFOs.
Layer one: one source of truth
Establish a canonical record: orders or qualified leads from the business system of record (CRM, e-commerce backend), enriched with marketing source via UTMs enforced by convention and click IDs stored at capture. GA4 can serve as this layer for many teams — if configured with the discipline covered in our GA4 events guide. Platforms are then read as optimization instruments, never as accounting.
Layer two: a consistent taxonomy
- UTM convention, enforced: lowercase, fixed delimiters, a governed naming sheet. ("fb / cpc / remarketing / oct" beats whatever the media buyer typed that day.)
- One metric dictionary: CPA defined identically everywhere (which conversions count, what window, which dedupe rules).
- Channel groupings that match how budget decisions are made — not Google's defaults.
Layer three: the incrementality program
Attribution models apportion credit; only experiments estimate causality. The quarterly incrementality battery:
| Test | Method | Answers |
|---|---|---|
| Geo experiments | Hold out matched regions; difference-in-differences | "Does this spend drive sales at all?" |
| Ghost bids / PSA | Serve placebo ads to a control cell | True conversion lift of prospecting |
| Platform experiments | Meta's A/B, Google Drafts & Experiments | Structured in-platform reads |
| Switch-off tests | Brief budget suspension on stable cells | The uncomfortable truth about channel dependence |
Layer four: reporting that decisions flow through
One weekly scorecard: spend, platform-reported CPA, source-of-truth CPA, and the ratio between them per channel (the calibration factor), plus blended MER. When platform-to-truth ratios drift, something broke — tracking, dedupe, or the campaigns. Monthly: incrementality findings feed budget reallocation; quarterly: the whole framework is re-baselined. This is the measurement spine inside our tracking & analytics service and the media buying practice it feeds.
Where do teams usually start?
Layer one and two — reconciliation and taxonomy. They're unglamorous, take weeks not months, and make every subsequent decision cheaper. Experiments come after the baseline is trustworthy.
How often should incrementality tests run?
One meaningful test per quarter per major channel. More produces noise; less produces faith.
Measurement maturity is a competitive moat. Build yours.