Platform ROAS Is Fiction. Here Is What to Measure Instead.

Add up the revenue your ad platforms claimed last month, then pull actual orders from your store. At most brands running three or more paid channels, the combined claim lands somewhere between 150 and 250 percent of reality. Every platform grades its own homework, every platform passes, and the sum of the report cards describes a business twice the size of yours. Platform ROAS is fiction in the strict sense: an internally consistent story that never has to reconcile with your bank account.

Why the number inflates

Four mechanisms do most of the work.

View-through credit. Meta's default attribution includes 1-day view, so someone who scrolled past your ad without clicking and bought the next morning through branded search counts as Meta revenue at full value. Nobody outside Meta can verify an impression happened, let alone that it mattered.

Double claiming across platforms. A buyer clicks your Meta ad on Tuesday and your Google Shopping ad on Thursday, then orders. Both platforms sit inside their attribution windows, so both book the full order. There is no referee deduplicating between them, which is how the industry gets away with claims that sum past 100 percent.

Modeled conversions. Where consent banners and iOS block tracking, platforms fill the gap statistically. The models are trained, tuned, and reported by the same party whose budget depends on the result. The gaps being modeled are real, as we show in our breakdown of why Shopify, GA4, and Meta disagree, but you are taking the size of the fill on faith.

Last-touch bias toward retargeting. Retargeting buys cheap impressions in front of people already close to buying, so inside the platform's own accounting it looks spectacular. A 12x ROAS on retargeting mostly measures how good the audience already was.

You can verify the inflation yourself in an afternoon. Export each platform's claimed conversion value for last month, sum the claims, and divide by actual store revenue. The ratio is your inflation factor. Most teams that run this exercise once never read a platform dashboard the same way again.

The number that cannot be gamed

Marketing efficiency ratio, or MER, also called blended ROAS, is total revenue divided by total marketing spend for a period, across all channels. Revenue comes from your order system. Spend comes from your invoices. No platform can inflate either side of the fraction, which makes MER the one performance number that survives contact with your P&L. If you did €400,000 last month on €100,000 of total spend, your MER is 4.0, and no attribution debate changes it.

Run it weekly against a target band derived from your gross margin and payback tolerance. A brand at 60 percent gross margin might set a floor of 3.0 and an ambition of 4.5. When MER sits inside the band, the machine is healthy. When it falls out, something real happened.

Two refinements sharpen it. Track new-customer MER separately from blended, since a strong repeat business can hide acquisition getting expensive underneath. And expect the number to compress as spend scales into colder audiences: a falling MER on rising spend is often the correct price of growth, so judge it against the plan rather than against last quarter.

MER has one blind spot, and it is a big one: it cannot tell you which channel did the work. That weakness defines the rest of the regime.

Demote platform ROAS, do not delete it

Platform ROAS still has a job. Within a single channel, the measurement bias is roughly constant, so comparing two Meta campaigns, two audiences, or two creatives against each other on Meta's own numbers is fair. Creative A at 3.8 beating creative B at 2.1 is a real signal even if both absolute numbers are inflated.

The line to hold: platform ROAS may move budget within a channel, never between channels. Comparing Meta's 4.2 against Google's 3.1 is comparing two different fictions written by two different authors.

Incrementality is the tiebreaker

When claims conflict, run an experiment. Holdout tests exclude a random audience slice from a campaign and compare purchase rates. Geo tests pause or boost spend in matched regions. The bluntest version is a spend-pause test: turn a channel off for two or three weeks and watch blended revenue.

The pause test settles arguments no dashboard can. When retargeting claims a 5.2 ROAS and two weeks of silence move revenue by nothing detectable, you have your answer, and it was worth far more than the paused spend. Experiments cost time and statistical patience, so reserve them for your biggest budget line and your most doubted one, roughly once a quarter each.

Attribution gives direction, experiments give proof

Between the always-on truth of MER and the occasional truth of experiments sits multi-touch attribution: models run over your own deduplicated touchpoint data, in your own warehouse, where no platform holds the pen. Used honestly, it tells you which channels open journeys, which ones close them, and where the next euro probably works hardest. Used dishonestly, it becomes one more single number to hide behind. The difference is running several models side by side and reading their disagreements, which we cover in Stop Choosing an Attribution Model. This is the layer where an attribution setup on your own data earns its keep.

What each metric can and cannot tell you

A weekly cadence that holds

Monday, thirty minutes: MER for the trailing week against the band, plus the trailing 13 weeks for trend. If it is in band, no cross-channel budget moves this week.

Daily, inside channels: media buyers optimize creative and audiences on platform numbers, within budgets that were set elsewhere.

Monthly: attribution review. Compare model outputs, flag channels whose credited revenue swings hard between models, and propose reallocations from that spread rather than from any single number.

Quarterly: one incrementality test, designed in advance, on the line item where the platforms' story and your skepticism diverge most.

The discipline underneath all of it is a single rule: no platform's claim about itself ever moves money between channels on its own. Write the decisions down. Six months of written decisions will teach you more about your channel mix than any dashboard.

Where Verity fits

MER, order-of-record revenue, and cross-channel spend only line up when they live in one place with one set of definitions. Verity lands your ad platform and analytics data in your own BigQuery, defines metrics like MER and net revenue once in a governed semantic layer, and keeps the weekly numbers on dashboards your whole team reads from. The platforms keep their opinions. You keep the ledger.