Why three platforms all claimed the same sale

If you export conversions from Meta, Google and TikTok and add them up, the total is usually larger than the number of orders in your store. Not by a rounding error. By a third, sometimes more. Every platform sees one slice of the journey and books the whole sale to itself.

This is not fraud and it is not a bug. It is the predictable result of three ad networks each running their own attribution window over the same customer, with no shared ledger between them. The buyer is real. The sale is real. The double and triple counting is an artefact of how the platforms are built to report.

The stakes are local. Australian businesses spent A$18.4 billion on internet advertising in 2025, up 11.5% year on year, with A$8.0 billion in search and A$5.4 billion in video, video's strongest year on record at +19.8% (IAB Australia / PwC Australia, IARR 2025). Most of that flows through the same handful of platforms, each reporting on its own terms, so the reconciliation problem below is not a US import. It is sitting in the spend reports of nearly every Australian advertiser.

Measured, an incrementality vendor, documents cases where ad platforms collectively claim credit for up to 140% of actual revenue. PantoSource, walking through a single retailer's numbers, shows 650 real orders surfacing as more than 1,200 platform-reported conversions, roughly 85% inflation (PantoSource, Multi-Platform Attribution). Either way, the dashboards are describing a business that does not exist.

One sale, counted three times

Walk a single purchase through a normal week. A customer scrolls Instagram and sees a prospecting ad. Two days later they search the brand on Google and click. That evening a Meta retargeting ad reminds them, and they buy. One order. One payment. One line in your Shopify export.

Now look at what each platform records. Meta, working on a 7-day-click / 1-day-view window, sees both the prospecting impression and the retargeting touch and books the conversion. Google, on a 30-day-click window, sees the branded search click and books the same conversion. TikTok, if it was anywhere in the path on a 1-day-view default, may book it too. Three platforms, three plus-ones, for one sale.

The root cause is mismatched windows with no reconciliation. Meta counts a view up to a day later and a click up to a week later. Google counts a click up to 30 days later. TikTok defaults to a 1-day view. The same conversion falls inside more than one window, so it is booked more than once, and nobody nets it out because no platform can see the others.

The arithmetic of over-counting

The inflation is easy to quantify once you have both numbers: the orders your store actually recorded, and the sum of conversions the platforms claim.

The same buyer who passed through all three platforms generated three reported conversions for one real sale. Scaled across a month, that is how 650 orders become 1,200-plus on the combined dashboard, an 85% overstatement that no single platform is lying to produce.

The over-counting is not uniform

Some channels inflate far more than others, because some channels claim credit for demand that already existed. The right question is not how much a channel reports, but how much of it is incremental: sales that would not have happened without the spend.

Independent measurement puts numbers on it. Meta reports roughly 26% more conversions on average than neutral third-party analytics, driven by view-through credit and modelled conversions, and Google over-attributes by around 15-20% when Enhanced Conversions and Consent Mode V2 modelling kick in (AdAmigo.ai, citing Varos and EasyInsights, 2026). Branded search shows only 20-40% incrementality, meaning 60-80% of those buyers would have converted organically. Retargeting is often 40-70% non-incremental (Measured). In one Meta holdout test, true incremental return was 2.1x against a platform-reported 4.8x, an overstatement of roughly 2.3 times.

Branded search is the clearest case. A customer who already typed your name was usually going to find you. Crediting that click with a full conversion, and then cutting the budget that drove the unbranded demand upstream, is how brands quietly defund the spend that actually works.

The platforms are not measuring the same thing badly. They are each measuring a different thing, correctly, from inside their own walled garden.

Privacy made the gap permanent

It is tempting to wait for the platforms to fix this. They will not, because a large part of the gap is now structural. Apple's App Tracking Transparency opt-in plateaued near 35% of users shown the prompt (Adjust, Q2 2025), and iOS 14.5 cut Meta's tracking accuracy by an estimated 30-40%. Marketers commonly see only 40-60% of their actual conversions in platform reporting, with the rest filled by modelled conversions that now account for 20-35% of reported results. Modelling closes part of the gap. It never closes all of it, and it never will. The deterministic, user-level tracking that made last-click feel precise is gone for good.

What to trust instead: triangulate, do not pick

The mistake is hunting for the one true dashboard. There isn't one. Credible measurement triangulates independent methods, each answering a different question, and reconciles them against the business numbers that cannot be double-counted: orders, contribution margin, and cash.

• Last-click tells you which touch closed. Useful for execution, blind to demand creation.
• Data-driven attribution weights touches across the path. Closer to reality than first or last click, still blind to incrementality.
• Marketing-mix modellingworks top-down to separate base demand from incremental lift across every channel, including offline. It answers where the next dollar should go. Google's open-source Bayesian MMM, Meridian, went generally available on 29 January 2025, lowering the cost of doing this honestly.
• Incrementality testing, via geo holdouts, is the only method that proves causation: what additional sales the spend actually caused.

The closed loop runs in that order. MMM finds saturation and strategic gaps; geo experiments validate the causal claims; attribution optimises execution inside the boundaries MMM sets; and the experiment results feed back as priors. The governing rule is to optimise on marginal return, not average. The first A$100k into a channel can look healthy while the last A$50k is wasted, and only the marginal view shows it.

Brands that adopt this triangulated approach typically find 10-25% efficiency gains from reallocating to causally-validated channels, without spending another dollar (Measured). The stack scales with spend: under A$1M, attribution plus one or two incrementality tests a year; A$5-20M, all three methods integrated; above A$20M, continuous MMM with an ongoing experiment programme.

The number under the number

There is a deeper reason platform ROAS misleads, beyond the double-counting: it ignores margin. A reported 4x return on a 30% contribution margin is a different business from 4x on 70%. The breakeven rule is unforgiving:

This is the through-line for the work we publish: we judge channels on contribution margin and CAC payback, not on platform-native ROAS. We call the underlying measure Profit Velocity, the rate at which marketing and sales effort converts into durable contribution margin. The right scoreboard is the one the platforms cannot inflate, because it is built from your orders, your costs, and your cash, not from three dashboards each claiming the same sale.

Blufire client · Cheapest Liquor

10.2x · $1.85M

An APAC Search Award winner. Measured on margin and incremental return rather than summed platform ROAS, which is how the spend stayed profitable as it scaled.

So the next time the dashboards add up to more sales than your store recorded, do not reconcile them by picking a favourite. Net them against the one ledger that cannot lie, and measure what each channel causes, in margin, not what it claims, in revenue.