Retargeting ROI Calculator

The calculation itself

Reported ROAS is attributed revenue divided by spend, exactly as on any other campaign. Incremental ROAS subtracts, from the numerator, the revenue the same audience would have produced with no advertising at all.

That subtracted amount is the baseline: the share of the reached audience who would have converted unaided. It is the only input the ad platform cannot supply, because it requires knowing what happened to people who were not shown the ad.

Incremental profit is then incremental revenue times contribution margin, less the spend. Break-even ROAS is one divided by contribution margin, unchanged from any other channel.

In practice

A 200,000-person pool with 60% reached is 120,000 people. Eight impressions each is 960,000 impressions, and at a $9 CPM that is $8,640 of spend.

A 0.7% click-through rate gives 6,720 clicks; a 5% conversion rate gives 336 orders. At $65 each that is $21,840 of attributed revenue and a reported ROAS of 2.53.

Now the baseline. If 0.1% of the reached audience would have bought anyway, that is 120 orders, leaving 216 the advertising actually caused. Incremental revenue is $14,040 and incremental ROAS is 1.63.

Break-even at a 45% contribution margin is 2.22. The campaign reports 2.53 and looks profitable; on the conversions it actually caused it is 1.63 and loses $2,322. Nothing about the campaign changed between those two sentences except which conversions it is allowed to claim.

Why retargeting always looks best

Retargeting selects its audience on a signal — a site visit — that already predicts purchase. It then shows ads to people on their way to a decision they had largely made, and records a conversion when they arrive.

The selection is doing the work that the advertising is credited with. This is the same mechanism that makes branded search report spectacular returns, and it is not fraud or bad measurement on the platform's part; the platform genuinely cannot see the counterfactual.

The practical consequence is that retargeting reports the highest ROAS in most accounts and is the channel most likely to be over-funded. Budget moves toward it precisely because the measurement flatters it.

Where the figure deceives

A baseline of zero is an assumption, not a default. Assuming it makes incremental ROAS equal reported ROAS by construction, which is why the number needs a holdout behind it rather than a guess.

Incremental conversions can come out negative. That is a real result: it means the campaign is claiming fewer conversions than the audience produces unaided, so what the report measures is the audience rather than the advertising.

Some campaigns lose money at any baseline. When the contribution on every claimed order is below the spend, incrementality is not the problem and no holdout will rescue it — the media cost is simply too high for the margin.

Frequency compounds the error. The same person seen eight times is eight impressions of cost, and past the point where repetition stops changing behaviour, the extra impressions add spend to the denominator and nothing to the numerator.

Running the holdout

Withhold ads from a random slice of the retargeting audience — 10% is usually enough — and compare conversion rates between the held-out and exposed groups over the full purchase window.

Randomise the split rather than using a geography or a device type. Any split that correlates with intent gives you a comparison of two different audiences instead of a measurement of the advertising.

Run it long enough to cover the actual consideration period. A three-day test on a product people take two weeks to buy measures timing rather than incrementality.

Re-run it when the audience definition, the creative or the window changes. The baseline is a property of that specific configuration, not a constant for the account.

Acting on it

Judge retargeting on incremental ROAS against break-even ROAS, and treat the reported figure as an upper bound rather than a measurement.

Watch the pool size. Membership windows expire continuously, and if new visitors arrive more slowly than old ones age out the pool drains, frequency climbs on fewer people, and performance decays for a reason the campaign report never shows.

Cap frequency where marginal lift stops paying for the impression, not at a round number someone recommended.

If there has been no holdout, say so. Incremental ROAS without one is a scenario, and presenting it as a measurement is how a channel keeps its budget for another quarter.