Every network's ad manager can tell you how many sales it thinks it drove. What it can't tell you is how many of those people would have bought anyway — from a Google search for your brand name, a friend's recommendation, or an email you sent the same week. A click or a view gets counted as credit even when the ad barely mattered, and two or three networks often claim the same sale at once, since each one grades its own homework. Ads and Pixels starts every network at a neutral estimate — "probably right, give or take half" — until a real incrementality testing result says otherwise. That's the discipline behind marketing incrementality: treat every platform's own number as a claim, not a fact, until a test proves it. Once a network is shown to be over-claiming, its numbers get less credit going forward, and on Meta and Google today cross-network budget allocation shifts spend to match.
Ads and Pixels / Lift Tests
Incrementality testing that shows which network actually caused the sale
Every network's dashboard is happy to take credit for a sale someone else already had planned. Incrementality testing settles it with a real experiment — a group that saw the ad, compared to one that didn't — and on Meta and Google today, a conclusive result moves your budget toward the network that's actually working.
Why a platform's own conversion count over-claims
What incrementality testing actually measures
Incrementality is the sales or leads an ad caused that wouldn't have happened without it — not the sales it happened to be near. A pixel can tell you someone converted after seeing an ad; it can't tell you whether they'd have converted anyway. The only way to answer that is an experiment: hold a comparable group back from the ad, let everyone else see it as normal, and measure the difference in real outcomes between the two groups. That difference is the incremental lift, and it's what incrementality testing hands back — real incrementality measurement, not a single network's own attributed total.
Four incrementality testing designs, and who each one fits
Which design fits depends on which network you're testing, how many regions you run in, and what your account is eligible for. Ads and Pixels supports four incrementality testing designs.
Meta Conversion Lift
Where your Meta account is eligible, Ads and Pixels orchestrates this study from your dashboard and pulls the result back in. Fits accounts already spending enough on Meta to qualify for its own lift tooling.
Google Conversion Lift
Google doesn't offer an API for this one, so it's set up the way Google intends: in Google Ads directly, or with your Google rep. You enter the result into Ads and Pixels once it's done, and that entry is audited. Fits Google Ads accounts with a rep relationship or the scale to self-serve it.
Geo holdout
Some regions keep running your campaign; matched control regions are held out entirely, and the difference is measured from your own region-by-region results. Needs at least 10 regions and 28 days of pre-period history to find a good match. Works on Meta, Google, TikTok, Snapchat, LinkedIn, Pinterest, X and Reddit. Fits multi-region advertisers, especially on networks without a lift product of their own.
Switchback
A single campaign alternates on and off across scheduled periods shown in advance; every toggle goes through the same approval as any other change, and an "off" period only counts if its spend actually hit zero. Fits smaller accounts, or ones in too few regions for a geo holdout, that still want a real answer without a full platform study.
Which design runs on which network
Not every incrementality testing design is available everywhere. A platform lift study only exists where the network itself offers one — Meta and Google, and only for that network's own campaigns. Geo holdout and switchback reach further, since neither depends on the network having a lift product at all.
| Network | Meta Conversion Lift | Google Conversion Lift | Geo holdout | Switchback |
|---|---|---|---|---|
| Meta | Yes | Not included | Yes | Yes |
| Google Ads | Not included | Yes | Yes | Yes |
| TikTok | Not included | Not included | Yes | Yes |
| Snapchat | Not included | Not included | Yes | Yes |
| Not included | Not included | Yes | Yes | |
| Not included | Not included | Yes | Yes | |
| X (Twitter) | Not included | Not included | Yes | Yes |
| Not included | Not included | Yes | Coming | |
| Microsoft Advertising | Not included | Not included | Coming | Coming |
Reddit supports geo holdout today but not switchback yet, since resuming a paused Reddit campaign on a schedule isn't switched on. Microsoft isn't wired into lift testing at all yet — coming, same as the AI optimizer.
The limits that keep incrementality testing honest
A test that's too short or too small hands back noise dressed up as an answer, so every incrementality testing design in the product has hard limits. Tests run 14 to 56 days: long enough to build a real sample, not so long the campaign changes underneath it. A test is only accepted when enough conversions are expected to detect a lift of 30% or smaller; if your volume can only reveal a bigger effect than that, the app tells you why and doesn't start it. The two platform studies add a spend floor: $5,000 over the run, which is Google's own published minimum for a conversion lift study and the stand-in minimum Ads and Pixels uses for Meta, since Meta doesn't publish one. Geo holdout adds two requirements of its own: at least 10 regions, and at least 28 days of pre-period history so a control region can be matched to a test region at a correlation of 0.8 or better and an error of 20% or less.
What a conclusive result changes: cross-network budget calibration
A conclusive incrementality testing result changes where your budget goes next. Every network starts on a neutral prior — its reported conversions treated as "probably right, give or take 50%" — until a lift test says otherwise. If a network's true lift comes in well under what it reported, its numbers get discounted from then on, and less budget moves its way. That discount feeds cross-network allocation directly: money moves from the lowest marginal return to the highest, only when the receiver beats the donor by at least 1.2x, and only up to 10% of a campaign's daily budget per week by default (15% at most). A network with no lift result isn't penalized, but it's marked "uncalibrated" in the app. Today that calibration moves money on Meta and Google, where the AI optimizer runs; elsewhere a result is still recorded and starts affecting budget once the optimizer reaches that network.
A running test freezes optimizer moves on the campaigns it's testing
If the optimizer quietly shifts a tested campaign's budget while an incrementality testing run is trying to measure that same campaign's effect, the result stops meaning anything. So any campaign inside a proposed or running lift test — any of the four designs — is off-limits to automated budget moves for as long as the test is active. You can still change it by hand.
Frequently asked questions
Real answers before you make a decision.
What's the difference between incrementality testing and normal conversion tracking?
Conversion tracking tells you someone converted after seeing an ad. Incrementality testing tells you whether they'd have converted anyway, by comparing a group that saw the ad to a matched group that didn't.
Do I need $5,000 to run any lift test?
Only for the platform studies: Meta and Google Conversion Lift need $5,000 in spend over the run. Geo holdout and switchback have no spend floor, but every design still needs enough conversions to detect a lift of 30% or smaller, so a switchback test is usually the realistic choice for a smaller account.
Can Ads and Pixels set up a Google Conversion Lift study for me automatically?
No — Google doesn't offer an API for it, so the study is set up in Google Ads or with your Google rep. Once it's done, you enter the result into Ads and Pixels and that entry is audited like any other change.
Is Meta Conversion Lift fully automated?
It's orchestrated from Ads and Pixels where your account is eligible for Meta's lift tooling, and results are pulled back in. Meta's own eligibility and API behavior are still being confirmed on a live account, so treat it as guided, connected setup rather than fully hands-off.
What happens to my numbers if I never run a lift test on a network?
Nothing gets penalized — that network's reported conversions stay at the neutral starting point, and it shows as "uncalibrated" in the app so you know its numbers haven't been checked against a real result.
Can the optimizer still move budget on a campaign I'm testing?
No. Any campaign inside a proposed or running test is frozen against automated budget moves until it concludes, so the result isn't measuring a moving target.
Find out which network is actually earning its budget
Connect your accounts and run incrementality testing on the network you trust least. A conclusive result recalibrates its numbers automatically from there.