Every conversion report in Google Ads has a quiet lie baked into it: it counts conversions, not incremental conversions. If someone was going to buy from you anyway and happened to click a branded ad or a remarketing banner on the way there, Google counts that as a win. It wasn't. That sale would've happened with zero spend. I've run enough accounts to know this isn't a rounding error — on branded search and remarketing specifically, a meaningful chunk of 'reported' revenue is revenue you'd have gotten for free.
Attribution models, even data-driven ones, can't fix this because they only see what happened with ads on. They have no idea what would've happened with ads off. The only way to find that out is to actually turn ads off for a slice of your account and measure the difference. That's incrementality testing, and it's the closest thing to ground truth you'll get in paid search.
Why reported conversions overstate reality
Branded campaigns are the biggest offender. Someone googles your company name after seeing a review, getting a referral, or remembering your last purchase — then clicks your ad instead of the organic listing right below it. Google credits the ad. In reality you probably would've gotten that click and that sale organically, for nothing. I've written before about why you still need a branded campaign and why you're probably overpaying for it — incrementality testing is how you find the actual overpayment, not just suspect it.
Remarketing has the same problem in a different shape. You're showing ads to people who already added to cart, already requested a demo, already know your brand. Some percentage of them were coming back regardless. The ad didn't change their decision — it just got credit for a decision already made. The same logic applies, to a lesser degree, to high-intent non-brand terms where your organic or direct presence is already strong.
Designing a geo holdout test
The cleanest incrementality test splits comparable geographies into test and control. You pause or significantly cut spend in control regions, keep it running normally in test regions, then compare the conversion rate (or revenue per capita) delta over a fixed period. A few things make or break this:
- Pick regions with similar size, demographics, and historical performance — don't compare your biggest metro to a rural market and call it a control
- Run it long enough to cover a full sales cycle, especially if you've already set up offline conversion imports for long sales cycles — a two-week test on a 45-day cycle tells you nothing
- Hold everything else constant during the test window — no new creative, no landing page changes, no seasonal promos that only hit one region
- Measure total conversions in the geography, not just platform-reported conversions, so organic and direct pickup in the control region actually counts
The output is a lift percentage: how much more revenue the test region generated versus what the control region's trend predicted. That lift percentage is your real incrementality rate for that campaign type. Apply it honestly and you'll usually find branded search landing somewhere far below 100% incremental, while cold-audience prospecting campaigns often land close to it — because those clicks genuinely wouldn't have converted without the ad in front of them.
Designing an audience holdout test
When geo splits aren't practical — small account, single market, B2B with a short list of target accounts — I use audience holdouts instead. Take your remarketing or customer match audience, randomly exclude a statistically meaningful slice from targeting, and track their conversion behavior against the audience that still sees ads. If your excluded group converts at nearly the same rate as the targeted group, your remarketing spend is mostly chasing sales that were already coming.
This only works if the audience is large enough to produce a clean split and your customer match list quality is solid — a low match rate quietly undermines this test the same way it sabotages smart bidding generally, because you're not actually testing the audience you think you're testing.
A conversion isn't a win just because it happened after a click — it's only a win if it wouldn't have happened without one.
Budget allocated by reported ROAS: branded and remarketing look like the top performers, so they keep getting funded while prospecting campaigns get cut for 'underperforming.'
Budget allocated by incremental lift: branded gets trimmed to defensive coverage only, remarketing gets scoped to recent high-intent windows, and the savings get redeployed into prospecting that's actually generating new demand.
Turning test results into a reallocation decision
The point of running this isn't an academic exercise — it's a budget decision. Once you have an incrementality rate per campaign type, multiply your reported conversion value by that rate to get a real, incrementality-adjusted ROAS. Rank campaigns on that number, not the platform number. In almost every account I've tested, this reshuffles the ranking meaningfully: branded and remarketing drop, cold prospecting and new-audience Performance Max segments rise.
That doesn't mean killing branded or remarketing. It means funding them at the level their real incrementality justifies — enough branded spend to block competitor conquesting and protect your top slot, enough remarketing to catch genuinely warm abandoners within a short window — and pushing the freed-up budget into channels proven to create demand rather than just capture it.
- ✓︎Pick geo splits if you have multi-region spend, audience holdouts if you don't
- ✓︎Match test and control groups on size and historical performance before you start
- ✓︎Run the test through at least one full sales cycle with nothing else changing
- ✓︎Calculate lift against total conversions in the group, not just platform-attributed ones
- ✓︎Apply the resulting incrementality rate to reported ROAS before making budget calls
- How long should an incrementality test run?
- Long enough to cover at least one full sales cycle and smooth out normal week-to-week variance — often four to eight weeks, longer for B2B with longer consideration periods.
- Can I run this on a small budget account?
- Audience holdouts work on smaller budgets better than geo splits do, since geo tests need enough volume per region to produce a statistically meaningful comparison.
- Should I stop branded and remarketing spend entirely if incrementality is low?
- No — low incrementality means less spend is justified, not zero. Competitor conquesting and brand protection still have real value that a lift test alone won't fully capture.