Incrementality: the conversions that would have happened anyway
Incrementality is the share of conversions that would not have happened if you had not run the ads. Everything else is people who were going to buy anyway, and the ad just got there first with a click ID.
Attribution cannot answer this. Attribution answers "which click do we have on file." Those are different questions. Treating them as the same question is how a branded-search campaign looks like a printing press.
Incrementality, in one sentence
Lift: conversions in the exposed group minus conversions in a comparable unexposed group.
If you cannot name the unexposed group, you are not measuring incrementality. You are measuring correlation and calling it science.
Why attributed ROAS is not incrementality
Attributed ROAS counts every conversion the platform can attach to an ad. That includes:
People who searched your brand name and would have signed up from organic.
People who clicked a retargeting ad on their way to a bookmark.
People whose last click was paid and whose first intent was a friend.
None of that is fake. It is just not incremental. The platform is not lying; it is answering a different question than the one finance asked.
A $93 CPC on the keyword "incrementality" tells you what the market already knows: this number is expensive because it is the one that decides budgets.
What you can do before a geo-lift
Most vibe-coded apps and small SaaS teams do not have the traffic to run a clean geo-experiment. That is fine. You can still stop lying to yourself.
Split brand vs non-brand. Brand search is mostly harvested demand. Report it separately. Never let it subsidise prospecting.
Watch payback, not ROAS. Payback asks when the cohort's revenue crossed CAC. Harvested demand still has a CAC; it just pays back faster. See ROAS vs payback period.
Keep unattributed revenue unattributed. If you smear it across paid channels, you manufacture incrementality in a spreadsheet.
Hold out a channel for two weeks if you can stand it. Even a sloppy on/off is more honest than a 6x ROAS tile.
When a real test is worth it
Run a geo-lift or a time-series holdout when:
One channel is more than ~30% of spend.
The attributed ROAS is the reason you will not cut it.
You have enough conversions in both cells to see a difference that is not noise.
Until then, treat platform ROAS as a delivery signal and payback as the decision number. Incrementality is the question you graduate to, not the one you fake on day one.
FunnelKeeper does not pretend to be a geo-lift vendor. It joins spend to revenue, plots the payback curve, and leaves unattributed revenue on its own line so you can see the hole attribution cannot fill. Start free.
FAQ
Is incrementality the same as attribution? No. Attribution assigns credit among the clicks you captured. Incrementality asks whether the conversion exists because of the ads.
Can I get incrementality from Meta's reporting? Meta's incrementality products exist for large advertisers. For a $4–$8/day product they are not the tool. Your holdout is turning the campaign off.
Where does FunnelKeeper sit? It makes the payback join honest so you have a decision number while you are still too small for a lift test. It will not invent causal lift from last-click data.