Attribution: what your ad dashboard is not telling you
Every ad platform reports the revenue it can attribute to itself, and every platform is generous about what counts. That is not dishonesty — it is the only measurement it can make. But attributed revenue and revenue you would not otherwise have had are different numbers, and only one of them pays wages.
What “attributed” actually means
Somebody saw or clicked an ad, then bought within a window. That is the whole claim. It does not establish that the ad changed the outcome.
The windows are wide — commonly seven to fourteen days for a click, sometimes one to seven for a view without a click. A customer who had already decided, searched for your brand and happened to pass an ad on the way is counted as an ad sale.
Run two platforms and both will claim the same order. The sum of attributed revenue across your channels routinely exceeds your actual revenue, which is the clearest possible sign that attribution is a claim rather than a measurement.
Brand traffic is where the illusion lives
Ads on your own brand name convert superbly and produce spectacular ROAS. They also largely buy customers who typed your name because they were already coming to you.
There are real reasons to bid on your brand — a competitor bidding on it, a marketplace listing you want to control, defending a launch. But it should be a deliberate decision, not the thing quietly propping up your account-level ROAS while the campaigns doing real acquisition look mediocre next to it.
Separate branded from non-branded before judging anything. They are different businesses wearing the same account.
The only test that settles it
Turn the campaign off and watch total sales — not attributed sales, total.
Pick a period long enough to see through normal variation, keep everything else stable, and compare against the equivalent period before. If total revenue drops by roughly what the campaign was claiming, the attribution was honest. If it barely moves, you were paying for demand you already had.
Sellers who run this test for the first time are usually unpleasantly surprised on brand campaigns and pleasantly surprised on upper-funnel ones that looked poor in the dashboard. Both findings are worth money.
A geographic split is cleaner where volume allows it: run the campaign in one region and not another, then compare. It removes the seasonality problem that before-and-after testing cannot.
Why marginal spend is what matters
Reported ROAS is an average over everything you spent. The decision in front of you is always about the next slice of budget — and that performs worse, because the cheapest and most relevant inventory is bought first.
A campaign averaging ROAS 5 against a break-even of 3.33 is not automatically safe to scale: the last portion of spend may already be at 3. The way to find out is not to read the average but to move the budget and watch what total sales do. The full reasoning is in how much to spend on ads.
If a 30% budget increase produces 10% more revenue, you have found the edge. That is a more useful fact than any figure in the interface.
TACoS is the honest long-run number
ACoS measures spend against attributed sales, so it inherits every problem above. TACoS measures spend against total sales, which no attribution model can inflate.
Watch them together. If ACoS holds steady while TACoS falls, organic sales are growing and advertising is building something. If both rise together, you are buying an increasing share of the same demand.
What to do with all this
- Treat reported ROAS as an upper bound, never as the number.
- Split branded and non-branded and judge them separately.
- Test incrementality on your largest campaigns at least once a year — pausing, or better, a regional holdout.
- Track TACoS alongside ACoS.
- Judge scaling decisions by what total sales do when you move the budget, not by the average in the panel.
None of this means advertising does not work. It means the dashboard measures the wrong thing well, and the right thing has to be measured by you.
Frequently asked questions
Because platforms claim orders independently and windows overlap. The same purchase appears in two or three dashboards. It is normal, and it is exactly why the totals cannot be added up.
Not necessarily, but test it. Pause for a fortnight and watch total sales. If they hold, you were paying for traffic you already had. If a competitor immediately takes the position, that is your answer too.
Long enough to cover a normal purchase cycle and see through weekly variation — two to four weeks for most e-commerce. Shorter than that and you are measuring noise.
Yes, and often more so. Sponsored placements on a listing you already rank for organically frequently cannibalise the free position. TACoS is the metric that exposes it.