Why Your YouTube ROAS Is Fake (and the Number to Trust Instead)

An 89× ROAS on YouTube with more conversions than clicks isn't a win — it's engaged-view attribution inflation. The mechanism, the one-afternoon proof, and the metric a CMO will actually believe.

By Andrej Ruckij · · 9 min read

Why Your YouTube ROAS Is Fake (and the Number to Trust Instead)

By Andrej Ruckij · July 22, 2026

TL;DR: If a YouTube campaign reports a huge ROAS and more conversions than clicks, the number is inflated by engaged-view conversions (EVC) — Google crediting a sale when someone watched the ad without clicking and later bought anyway. Because EVCs sit inside the main Conversions column, the report counts purchases the ad stood near, not ones it caused. It’s worst on high-traffic sites with strong organic demand. You can disprove it in an afternoon by reconciling Google’s claimed conversions against the store’s own order backend, and the only number worth putting in front of a CMO is an incrementality lift from a geo-holdout — not the platform’s self-graded ROAS.

A team walks in with a headline: 89× ROAS. About €1,000 in YouTube spend, ~€60 average order value, and revenue in the tens of thousands. It looks like the best campaign anyone’s ever run.

It isn’t. And you can prove it’s not without a single meeting, because the report contradicts arithmetic.

The one detail that settles it: conversions > clicks

Buried in the same report: the number of conversions is higher than the number of clicks.

For a single purchase on a ~€60 product, one click can produce at most about one order. So if credited conversions exceed clicks, it is arithmetically certain that most of those conversions had no click at all. They came from Google’s view-based crediting — and on YouTube, that means engaged-view conversions.

An engaged-view conversion is counted when someone watches ~10 seconds of a skippable ad (or the whole thing if it’s shorter), doesn’t click, and then buys within a few days. Crucially, for YouTube video-action and Demand Gen campaigns, EVCs are folded into the main “Conversions” column — right next to the ones that came from real clicks. That single design choice is the entire illusion. (Full mechanics: glossary/engaged-view-conversions.)

Restate it as CPA and the impossibility jumps out

89× ROAS at €60 AOV means a cost-per-acquisition of about €0.67 — you’re being told you bought a €60 order for 67 cents. A healthy CPA at that price point is €10–30. So the claim is a 15–45× better-than-reality acquisition cost, on cold-ish video traffic. ROAS hides the absurdity; CPA exposes it. Whenever a ROAS looks magical, convert it to CPA and ask whether you’d believe that number.

Why a high-traffic store makes it worse, not better

Here’s the counterintuitive part: the more successful and well-known the store, the more inflated the YouTube report gets.

Attribution inflation scales with organic demand. A store doing millions of visits a month has a huge pool of people who were going to buy anyway — branded search, direct, word-of-mouth. Show them a YouTube ad, and Google credits every watch-then-buy as a conversion the ad “drove.” The ad didn’t create the sale; it stood in front of a river of existing demand and claimed the water.

This is the same failure the wiki documents for Google’s glossary/performance-max (it absorbs branded searches that would have converted anyway) and names in general in glossary/incrementality-testing: when attribution looks too good, it’s usually crediting the channel for conversions that would have happened without it.

Two problems hiding in one report

It’s worth separating them, because they have different fixes:

  1. Attribution inflation — the big one: view-based credit (EVC/VTC) counting non-caused sales.
  2. Double counting — a config bug where the same order is counted twice, e.g. a Google Ads conversion tag and a GA4 conversion import both firing, or the purchase action set to count “Every” instead of “One.”

And if you see fractional, decimal conversion counts, that’s a third layer: modeled conversions — Google statistically estimating conversions it couldn’t observe and filling them in. Part of the headline isn’t measured at all; it’s inferred.

The one-afternoon proof

You don’t need a formal study to kill the 89× internally. In order of speed and persuasiveness:

  1. Reconcile against the store’s own order backend. The store knows exactly how many orders shipped in the window. Pull the UTM- or coupon-tagged orders attributable to the campaign from the merchant’s system and compare to Google’s claimed count. If Google says ~1,100 orders and the backend shows ~40, that gap is the inflation — measured against ground truth Google cannot touch. This is the single most convincing number for a skeptical audience.
  2. Segment the report by conversion type. Add click-through, engaged-view, and view-through columns in Google Ads. Watching engaged-view and view-through dwarf click-through confirms the diagnosis in five minutes, from Google’s own data.
  3. Rebuild it click-through-only, on a verified purchase event, with a short (1-day-click) window and view credit off. That’s the believable floor.

One honesty caveat: even click-through-only over-credits on a high-demand site, because last-click still hands the win to an ad that merely sat on the path of organic demand (see glossary/multi-touch-attribution). Treat it as a ceiling on the believable — not as truth.

The only number a CMO will respect

Truth comes from a counterfactual: expose some regions or audiences to the campaign, hold others out, and measure the difference in sales. That difference is the incremental lift — the sales that wouldn’t have happened without the ad. It’s un-fakeable.

  • A geo-holdout you run yourself is the gold standard — and, unlike a platform study, it isn’t Google grading its own homework.
  • A Google Conversion Lift study uses the same logic done formally. More credible than attribution, but it carries a self-measurement discount with sophisticated buyers.

The 2026 standard (per glossary/incrementality-testing): a 10–20% holdout, stable conditions during the test, and lift reported as “the % of conversions that wouldn’t have happened without the campaign.” A defensible 15–25% lift is worth infinitely more than 89×, because it survives scrutiny.

Why this matters beyond one annoying report

If the YouTube test was meant to be a proof asset — something to put in front of prospects or a CMO — then the 89× headline is not just wrong, it’s actively harmful. No one senior believes 89×. It reads as either naivety or spin, and it poisons trust in every other number in the deck.

Two honest paths forward:

  • Lead with the un-inflatable efficiency metrics — CPM, view rate, CTR, reach and frequency, cost-per-qualified-landing — clearly labeled as media efficiency, not incrementality.
  • Run a clean round two with a geo-holdout and produce one real lift number.

The reframe worth internalizing: a buyer’s real question isn’t “how big is your number” — it’s “can I trust your measurement.” A team that caught its own attribution inflation and re-tested cleanly is a far stronger pitch than one waving an impossible ROAS. The honest version is more sellable than the fake one.

Key takeaways

  • Conversions > clicks is the decisive tell — it proves most credited conversions never clicked. They’re engaged-view credit, folded into the main Conversions column.
  • Convert ROAS to CPA — a 60–89× ROAS at €60 AOV implies a sub-€1 CPA, which is obviously impossible.
  • Inflation scales with organic demand — the mechanism is worst on big, well-known stores.
  • Prove it in an afternoon by reconciling Google’s claimed conversions against the store’s own order backend.
  • The only trustworthy number is incremental lift from a geo-holdout — ideally one you run yourself, not a platform-graded study.
  • For outreach, honesty sells better — a defensible 20% lift beats a fake 89× because the buyer is really assessing whether they can trust you.

Sources