How Do I Know If My Meta Ads Actually Work?
You find out by turning them off for part of your audience and watching what happens to total revenue. Meta's own dashboard cannot answer this question, because it only reports on people who saw a Meta ad and then bought, never on the people who would have bought anyway. The difference between those two groups is the only thing that tells you whether the spend is doing work.
Three platforms, three answers, one budget to defend. Your GA4 export already holds the answer, and the first finding is free and never leaves your browser. Get your first finding free
By Joris van Huët, Founder & CEOPublished
The short version
- A high ROAS in Ads Manager means Meta found people likely to buy. It does not mean Meta caused them to buy.
- Retargeting is the clearest case: those people already visited your site. Much of that revenue was arriving regardless.
- The real question is incremental: how much revenue exists with the ads that would not exist without them.
- The only reliable answers come from holding something back: a holdout audience, a geo split, or a clean on/off test.
- Airbnb cut roughly $541M of performance marketing in 2020 and traffic returned to about 95% of the prior year's level anyway (see below).
Why this happens
Ad platforms are optimisation engines. You ask Meta to find people who will convert, and it does exactly that: it finds the people most likely to buy. Many of them were already going to buy. When those orders land, they land with a Meta ad in the path, and Meta reports them as its own. The better the targeting, the more of your existing demand it captures and re-labels as new.
The word for what you actually want to measure is incrementality: the revenue that exists because of the ad and would not exist without it. Reported ROAS and incremental ROAS can point in opposite directions, and the channels that look best in the dashboard are frequently the ones best at intercepting demand you already had.
The most-cited real-world example is Airbnb. In 2020 it cut performance-marketing spend by about $541 million (total marketing fell from roughly $1.14bn to $482m), and its traffic still returned to approximately 95% of 2019 levels without that spend, which told them the brand, not the bidding, was doing the work. Figures are from Airbnb's FY2020 reporting as covered by Marketing Week and Campaign in 2021.
What you can do about it today
Run a holdout
Exclude 10-20% of your audience from Meta ads for two to four weeks. Compare revenue per person in the held-out group against everyone else. The gap is your real lift. This is free and it is the single most useful thing on this page.
Separate prospecting from retargeting
Report them as two different things. Retargeting ROAS is almost always flattered; prospecting is where genuine new demand is created. Judging them on one blended number hides both.
Watch MER, not just ROAS
Total revenue divided by total ad spend across all channels. If platform ROAS climbs while MER stays flat, the platforms are trading credit between themselves and nothing new is being created.
Try a geo test if you cannot hold out an audience
Pause Meta in one comparable region and keep it running in another for two weeks. Compare total regional revenue. Less precise than a holdout, but it works when audience splits are impractical.
How Causality Engine helps
If you want to know where to test before you pause anything, Causality Engine reads your GA4 export and shows what each channel caused, next to last-click, in 1–2 minutes for €99. Each channel gets a next step and how to test it: change spend a little in one region, then compare revenue.
Go deeper: How incrementality testing works
Questions people ask next
My Meta ROAS is 4x. Isn't that good?
It might be, but 4x reported ROAS does not mean four euros arrived because of Meta. It means Meta was present in the path for four euros of revenue inside its attribution window. Some of that would have arrived anyway. The only way to know how much is to withhold the ads from part of your audience and compare.How long does a holdout test need to run?
Two to four weeks for most DTC brands, and it must cover at least one full purchase cycle. Shorter than that and normal week-to-week noise will swamp the effect you are trying to measure.Will pausing ads hurt my revenue?
You are not pausing everything; you are withholding ads from a small slice of the audience while the rest of the account runs normally. If revenue in that slice does not fall, you have learned something valuable and lost almost nothing. If it does fall, you have proof the spend is working and you can scale it with confidence.