Open your Meta dashboard. It reports 40 conversions. Open GA4 for the same campaign, same week. It reports 24. Ask your sales team how many enquiries they actually received, and you get a third number.
Nobody is lying. All three are measuring different things, and in 2026 the gaps between them have widened rather than closed.
If you have been treating one of those numbers as the truth and the others as broken, this article is about why that is the wrong frame, and what to do instead.
What changed
Several things happened in a short period, and the combined effect is larger than any one of them.
The privacy plumbing did not arrive as promised
Google retired ten Privacy Sandbox APIs in October 2025, including the Attribution Reporting API, and had already dropped the standalone cookie-choice prompt earlier that year. The replacement infrastructure the industry spent years preparing for is substantially not there.
Consent requirements tightened
Under 2026 consent standards, a meaningful share of visitors are never measured directly at all. Estimates commonly put the gap at up to 30% of data missing.
GA4 changed its defaults
Data-driven attribution is now on by default. If your dashboards were built around last-click and your numbers moved without explanation, the model changed underneath you. Plenty of reports broke quietly rather than loudly, which is worse.
Google also shipped a Conversion Attribution Analysis Report in February 2026, currently in beta, which changes how channel value is presented across the journey. Worth exploring, not yet worth building your reporting on.
The consequence: your numbers are modelled
This is the part that most reporting still hides. What GA4 shows you today is a blend of directly observed events, cookieless fallbacks, and machine-learning estimates filling the gaps.
That is not a scandal. Given the constraints, modelling is a reasonable response. But it changes what the number is. A modelled conversion count is an estimate with error bars, and treating it as a precise count leads to false confidence, particularly when you are comparing two channels whose figures were modelled differently.
Stop reconciling. Start comparing.
The instinct when dashboards disagree is to find the error and make them match. For most businesses this is unwinnable and a poor use of an analyst's month.
The more useful approach is to stop trying to produce one authoritative number and instead show the disagreement deliberately. Build a reporting layer that puts, for each campaign, side by side:
- Platform-reported conversions (what Meta or Google Ads claims)
- GA4 conversions
- CRM or actual sales records
Then make the gap itself a metric you watch. Stable gaps are fine, and you learn each platform's characteristic bias. A gap that suddenly changes is a genuine signal: tracking has broken, consent rates have shifted, or something real has changed in the funnel.
This is a modest engineering task and it produces more insight than another month of reconciliation.
The measurement marketers actually trust
In a January 2026 survey of 500 senior decision-makers, the method that earned the most trust was not platform reporting or media mix modelling. It was independent incrementality testing.
The idea is simple, and it sidesteps the attribution problem rather than solving it. Instead of asking "which touchpoint deserves credit," you ask "what happens to sales if we turn this off?"
Turn a channel off in one region and leave it running in a comparable one. Hold back a segment from a campaign. Compare outcomes. It is not elegant, it requires patience, and it answers the question that attribution models only approximate: is this spend actually causing anything?
Most businesses never run a single incrementality test, then argue about attribution models for years. The test is usually cheaper than the argument.
A dashboard worth having
If you are rebuilding reporting this year, a few principles hold up:
- Show the source of every number. A figure with no stated origin invites false precision.
- Report ranges where the number is modelled. "Between 20 and 28" is more honest and more useful than "24".
- Anchor on the metric closest to money. Qualified enquiries and closed revenue are measured in your own systems and are not subject to anyone's attribution model.
- Track consent rate as a first-class metric. If it drops, every downstream number moves, and you want to know that before you conclude a campaign failed.
- Keep a change log. Model changes, tracking updates and platform defaults all move your data. Without a log, you will spend next quarter explaining a step change nobody can date.
What this means for smaller budgets
If you are spending modestly across two or three channels, do not build an attribution science programme. The honest answer at that scale is:
- Ask every enquiry how they found you, and record it. Self-reported attribution is imperfect and surprisingly informative.
- Watch total enquiries against total spend over months, not campaign-level ROAS week to week.
- Run one deliberate on-off test a year on your largest channel.
That is less sophisticated than a multi-touch model and it will mislead you less often.
The summary
Attribution has not been solved and is not about to be. The dashboards will keep disagreeing. The businesses that make good decisions in this environment are not the ones with the most elaborate model. They are the ones that know which of their numbers are observed and which are estimated, and that occasionally turn something off to find out what it was really doing.
We build reporting that shows platform, analytics and CRM figures side by side, with the gaps made visible rather than smoothed over, as part of our reporting and analysis service. If your dashboards are currently disagreeing and nobody can say why, that is a good place for us to start.
