- Tracking answers 'did the event fire correctly?' Measurement answers 'did this spend cause this outcome?' Most tracking audits stop at the first question.
- A tag health dashboard showing 100% event coverage tells you nothing about whether your attribution model is assigning credit correctly. Those are two different systems.
- If finance and marketing are reconciling numbers by adding a footnote that says 'platform reporting differences,' you don't have a measurement architecture, you have two spreadsheets that don't talk to each other.
- Build measurement architecture as a layer on top of clean tracking: a defined source of truth, a documented attribution logic, and an incrementality check that runs on a schedule, not just when a campaign gets challenged.
Here's a pattern I've seen often enough to consider it a category of its own: marketing's quarterly report shows a healthy attributed revenue number from paid social. Finance's revenue recognition for that same channel and period shows something meaningfully lower, sometimes by a wide margin.
Nobody on either team thinks their number is wrong. Marketing pulled from the ad platform's own dashboard. Finance pulled from the CRM's closed-deal report. Both systems tracked correctly. Both reported exactly what they were built to report.
The problem isn't a tracking bug in cases like this. There's no broken tag, no missing conversion event, no misfired pixel. Every event fired exactly as configured. The problem is that nobody ever decided which number was supposed to be true, or how the two systems should reconcile when they disagree. That's a measurement architecture problem, and it shows up more often than it should. It's the same root cause I walk through in why marketing and finance see different revenue numbers.
Two different jobs, one word
"Tracking" and "measurement" get used interchangeably in most marketing orgs, and that's the root of a lot of wasted debugging time.
Tracking is the plumbing. It's GA4 events, Meta Pixel fires, server-side GTM tags, the Conversions API payload. Tracking's job is narrow: did the event happen, and did we capture it. Did the purchase event fire with the right value. Did the event_id get passed so the ad platform doesn't double-count it. Did the hashed email make it into the Enhanced Conversions payload. These are yes/no, pass/fail questions. You can debug them in the Network tab. You can verify them in Tag Assistant. There's a right answer.
Measurement is a different discipline sitting on top of tracking. Measurement's job is to answer whether a marketing dollar caused an outcome. That's not a yes/no question. It's a modeling question, and it requires decisions: which touchpoints get credit, over what window, using what methodology, reconciled against what source of truth. This is the same gap I keep running into when I look at attribution models that don't survive contact with finance — the model works fine in isolation and falls apart the moment someone asks it to agree with another system.
You can have perfect tracking and still have no measurement architecture. This happens constantly. Every event fires cleanly, every tag passes QA, and the company still can't answer "did our paid social spend actually grow revenue, or did it just take credit for revenue that was going to happen anyway."
If someone asks "did this campaign cause incremental revenue," and your answer requires opening the ad platform's dashboard and reading the reported ROAS, you don't have measurement architecture. You have tracking with an attribution model bolted on, and the platform is grading its own homework.
Why clean tracking doesn't prevent this
I've walked into accounts with immaculate GTM containers. Every tag consent-gated correctly. Server-side container deduplicating server and browser events with matching event_ids. Enhanced Conversions hashing emails properly. By every tracking QA standard, the setup is excellent.
And the CMO still can't tell the CFO why the paid media dashboard says one thing and the P&L says another.
That's because tracking QA checks the wrong layer for that question. A tag health audit verifies the event pipeline. It doesn't verify:
Which platform's attribution model is the source of truth. Ad platforms tend to default to attribution windows and models that credit themselves generously. In my experience, GA4's data-driven attribution model distributes credit using a different logic, and it can deprioritize paid social relative to channels closer to conversion, though the exact behavior depends on your account's data and settings. Both systems can be "correctly tracking," and they'll still never agree, because they're not measuring the same thing. Check your own platform's current attribution settings directly rather than assuming a specific default, since these settings change and vary by account. If your numbers look consistently off, see how to tell if your GA4 attribution is wrong.
Whether last-click, multi-touch, or media mix modeling is the reporting standard for board-level numbers. If marketing reports MTM-attributed revenue and finance reports cash collected against closed-won deals in the CRM, the numbers will diverge by definition. Neither is wrong. They're answering different questions. If you're deciding between the two, multi-touch attribution vs MMM breaks down which one fits your spend and channel mix.
Whether attributed performance reflects causal lift or just correlation. A retargeting campaign might show a strong reported ROAS in a platform dashboard, say a rough estimate of 6x, while actually claiming credit for purchases that would have happened anyway. Tracking has no way to catch this. Only an incrementality test (a holdout group, a geo lift test, a matched-market design) can tell you whether the spend actually moved anything. This is the same blind spot I've written about in why last-click attribution keeps overstating retargeting.
None of these are tracking failures. They're architecture decisions that were never made, or were made implicitly by whichever platform's default settings won by inertia.
Where this actually breaks in practice
The pattern I see most: a company scales spend on a channel because the platform-reported ROAS looks strong. Some time later, finance flags that revenue didn't grow proportionally. Someone gets asked to "audit the tracking." The tracking is fine. What's missing is a documented answer to "which number do we trust, and why," decided before the campaign launched, not reverse-engineered after finance asks a hard question in a QBR. The three-layer breakdown in why your CAC numbers disagree covers the same disagreement from the cost side instead of the revenue side.
What a measurement architecture actually contains
Measurement architecture isn't a tool you buy. It's a set of decisions, documented, that sit above your tracking stack. At minimum it needs four pieces.
A designated source of truth for revenue. Pick one. Usually the CRM or the general ledger, not the ad platform. Every attribution number gets reconciled against this, not the other way around. If the ad platform reports a much higher number than the CRM, the CRM wins for any conversation with finance. Write this down and get sign-off from both marketing and finance leadership, because this decision is where most of the political friction lives.
A documented attribution methodology, with its limitations stated explicitly. Whether you use last-touch, a multi-touch model in GA4, or a media mix model run periodically, write down what it does and doesn't capture. If your MTM model can't see offline conversions or CRM-sourced pipeline, say so in the same doc that reports the numbers. The failure mode isn't using an imperfect model. It's presenting an imperfect model's output as if it were ground truth.
A recurring incrementality check. Not a one-time test you ran two years ago. A schedule. Periodic holdout tests on your top channels by spend, geo lift tests when you're evaluating a new channel, a standing process for validating that attributed performance corresponds to actual causal lift. This is the piece almost nobody has, and it's the piece that catches the retargeting-fallacy pattern before it costs real money. For more on how these tests actually run, see how incrementality testing works in practice.
A reconciliation process for when marketing and finance disagree. Not a footnote. An actual process: who looks at the gap, what gets checked first (attribution window mismatch, currency/timezone handling, refund and chargeback treatment, CRM data entry lag), and how the resolved number gets communicated. If this process doesn't exist, every quarterly review starts from zero credibility.
You don't need to build all four pieces before you get value. Pick your single biggest attributed-vs-actual gap this quarter, usually your top-spend paid channel, and run the reconciliation manually once. Document where the numbers diverge and why. That one exercise usually reveals which of the four pieces is missing.
What to check in your own stack this week
Pull your last quarterly marketing report. Find the total attributed revenue number. Now find the same period's revenue in your CRM or ledger for the same channels.
If they match closely, you likely already have an informal reconciliation process, even if nobody's written it down. Go write it down, because informal processes don't survive a team member leaving.
If they don't match, don't start by re-auditing your tags. Ask instead: does anyone in the company know why these two numbers are supposed to differ, and by how much? If the honest answer is no, that's your actual finding. The tracking isn't broken. The architecture was never built.