The ROAS that lies: why your 6× return might be losing you money

Key takeaways

• Every platform grades its own homework. Each counts the sales it touched, under its own attribution window, so the same order is claimed more than once.
• Test 1, reconcile: add up platform-reported conversions and compare them with real orders in your back end. A gap of more than about 15% means double counting.
• Test 2, branded search: pause brand-name ads for two weeks. If revenue barely moves, you were paying for customers who were already looking for you.
• Test 3, geo holdout: switch paid media off in one comparable region for a few weeks and compare total revenue, not attributed revenue.
• Optimise for profit after ad spend, not platform ROAS. The number looks smaller and the business gets healthier.

Why a big ROAS can hide a loss

A 6× ROAS on a dashboard feels decisive. It’s reported with confidence, it goes in the board deck, and budget follows it. But ROAS as a platform reports it answers a narrow question: how much revenue did people who touched this ad go on to generate, within this platform’s attribution window?

“…returns from paid search are a fraction of conventional non-experimental estimates.”

Blake, Nosko & Tadelis, NBER Working Paper 20171 (eBay field experiments)

It doesn’t answer the question that matters: how much revenue would you have lost without the ad? A customer who searched your brand name, clicked your ad and bought would probably have clicked the organic link one line below. The ad gets the credit; your bank account sees no difference.

The attribution problem no one wants to name

Fewer clicks make the problem worse: Pew Research found users clicked a traditional search result on 8% of Google visits with an AI summary, against 15% without, so more journeys finish in places no tracking pixel sees.

Meta, Google and TikTok each use their own attribution model and window. A shopper who sees an Instagram ad, later clicks a Google ad and then opens a TikTok video before buying can be counted as a conversion by all three. Each platform is telling the truth by its own rules. Together, they describe sales that never happened.

Privacy changes made this worse. With less user-level data, platforms fill the gaps with modelled conversions. Modelling isn’t wrong, but it isn’t a receipt.

Three tests that expose broken tracking

Test 1: the dashboard reconciliation

For one month, add up the conversions every platform reports: Meta, Google, TikTok, email, affiliates. Compare the total with actual orders or qualified leads in your back end or CRM.

If the platforms claim more than about 15% above reality, they are double counting. Every channel’s ROAS is inflated by an unknown amount, and the channel that claims most aggressively wins budget it hasn’t earned.

Test 2: the branded search pause

Pause ads on your own brand name for two weeks, in a quiet period, and watch total revenue and organic clicks. For established brands with no competitor bidding on their name, revenue often barely moves, because the organic result catches the same customers. eBay’s published experiment in 2013 found exactly that for its brand terms.

If competitors bid on your name, the answer may differ. That’s the point of testing rather than assuming.

Test 3: the geo holdout

Pick two comparable regions. Keep paid media running in one and switch it off, or sharply reduce it, in the other for three to six weeks. Compare total revenue between them, adjusted for their normal difference.

If the region without ads earns nearly as much as the region with them, most of your attributed revenue would have arrived anyway. The full method is in incrementality testing with geo holdouts.

What replaces broken ROAS

The audience is reachable without inflated numbers: Facebook’s ad reach in the UAE was 9.70 million people (85.0% of the population) in late 2025 (DataReportal).

• Incrementality tests on a regular cycle, at least quarterly, for the channels that take most of the budget.
• Server-side conversion tracking (Meta Conversions API, Google Enhanced Conversions) with deduplication, so the signals platforms optimise on are accurate. Set it up under consent mode.
• Media mix modelling once spend and history are large enough to support it.
• One source of truth for revenue: the back end or CRM, not any platform’s dashboard.

The mindset shift

Stop asking which platform has the best ROAS. Ask what the business earns after ad spend, and which spend changes that number. Your reported ROAS will probably fall. Your decisions will be easier to defend to a finance team, and the budget you free up can go where it actually adds sales.

In-house at a Dubai interiors company, judging campaigns on booked measurements and orders rather than platform conversions is how budget moved. Those are previous-employer results, described in the 30-day attribution audit. The performance marketing service applies the same method.