Thinking · Newsletter
Is AppLovin actually incremental, or just taking credit from Meta?
Why AppLovin and Meta both claim the same purchase, what the published geo tests show (and who published them), and how to test it on your own account.
TL;DR. Sometimes, and you will only know for your own account by testing it. Both platforms can claim the same purchase. The published geo tests mostly come back positive, but almost all of them are published by AppLovin, a measurement vendor or an agency, and the independent results include one clean null. Platform ROAS cannot settle the question in either direction. A geo holdout read on your own sales data can, and this page explains how to run one.
Why both platforms claim the same purchase
Each platform counts conversions by its own rules, and neither one subtracts the other's credit.
Meta's default attribution setting is 7-day click and 1-day view, so it counts a purchase made within seven days of a click on a Meta ad, or within one day of someone seeing one without clicking (per Jon Loomer's guide to the setting). AppLovin says its in-platform reporting "currently captures click-through attribution only."
Picture one customer who scrolls past a Meta ad in the morning, taps an AppLovin ad in a game at lunch and buys that afternoon. Meta counts a view-through, AppLovin counts a click, and your two dashboards report two purchases against one order in Shopify. That is where the suspicion comes from, and it is fair. Adding platform-reported revenue across channels overstates the total.
The overlap is measurable in individual accounts. A case study from Metaply on one glass brand reports that Triple Whale "shows a 61% order overlap between AppLovin and Meta", meaning "roughly four in ten AppLovin-attributed orders had no Meta touch at all." That is one brand, and overlap tells you the channels reached some of the same buyers. It does not tell you whether those buyers would have bought anyway. Only a holdout answers that.
What the published tests say
What we could find and verify, with each source's own caveats.
| Source | Who published it | Design | What it reports |
|---|---|---|---|
| MaryRuth's | AppLovin | Haus GeoLift, 70% exposed, 30% holdout, 4 weeks plus a 2 week post-treatment window | 4.89% lift in direct sales, iROAS 49% above target; 1.84% Amazon lift; 4.03x combined iROAS |
| Ridge | AppLovin | Haus geo holdout, 20% of US regions held out, 3 weeks (Apr 21 to May 11, 2026), power banks only | 14.61% Amazon revenue lift, $1.24 Amazon iROAS against a $0.80 target |
| Four brands | Haus (Dec 2024) | 2-cell tests with a holdout, 2 to 3 weeks each | Lift in all four; two beat their efficiency target, two missed it |
| Cann | BlueAlpha (Feb 2026) | Geo holdout against a synthetic control, 19 days, about $10K a week | No statistically significant lift |
| Eight client tests | Common Thread Collective (Jan 2026) | Not described | All eight positive, averaging "slightly above 151%" of platform-reported impact |
MaryRuth's. The page title says the brand "beats incremental ROAS target by 40%". The body says the iROAS was "49% higher than their target". We use the body figure. The page gives no confidence interval.
Ridge. The title says "Ridge beats iROAS goal by 128%". The only iROAS in the body is $1.24 on Amazon against an assumed $0.80 target, which is about 55% above the goal. The Shopify lift is described only as "tens of thousands", so we could not reconcile the 128% from what the page reports.
The Haus four-brand study. This is the most useful one, because it shows the spread. "In all four cases, AppLovin drove incremental lift to these businesses," but "efficiency of the media, on the other hand, varied considerably between brands." Twillory came in 14.18% better than its iROAS target and Flux Footwear 37.22% better. Jones Road Beauty was 169.78% worse than its CPA target, and Fresh Clean Threads 9.8% worse than its iROAS target (the text says 10%). Haus's own conclusion: "results vary greatly between businesses."
Cann. BlueAlpha's geo holdout "detected zero statistically significant lift at Cann's ~$10K/week spend level". The point estimate was +$90K, with a 95% interval of minus $152K to plus $310K, which spans zero. BlueAlpha's own caveat is that "spend level is the single biggest variable that changes incrementality outcomes," and that "a different brand with different scale, creative, and audience might see a different result."
Common Thread Collective. In a January 2026 episode they report eight completed tests, say "every single one of the tests has come back positive," and compare the result to Meta 7-day click tests that land around "120%, 115%, 125%". A separate May 2026 post reports AppLovin discovery campaigns "averaging 284% incrementality", or 190% "after excluding outliers". BlueAlpha notes that, to its knowledge, CTC "did not publish the test designs, confidence intervals, holdout structures, or brand names."
Why vendor case studies are not the answer
Every positive study above was published by someone with a stake in the result: the ad platform, the measurement company that ran the test, or an agency that runs the channel. That does not make them wrong. It means you are reading the tests someone chose to publish: a brand whose test came back flat has no reason to write a case study, and the platform has no reason to host one. The four-brand Haus study and the Cann test are worth more than their size suggests, because they publish the misses.
Three more limits apply. Only the Cann test above reports a confidence interval. Lift is measured at one spend level, and the scaling that follows (Ridge raised daily spend "nearly 4x" after its test) is not measured. And the windows are two to four weeks, which says little about month six.
Our read: the evidence is good enough to say AppLovin can be incremental for consumer brands. It is not good enough to say it will be for yours, at your spend, with your creative.
How to test it yourself
AppLovin's own measurement team says an incrementality test, "run well, is the only instrument that would tell you with statistical confidence." We agree. This is the method we would use.
- Run a geo holdout. Turn AppLovin off in a set of regions and leave it on everywhere else. AppLovin describes holdouts as "typically sized at 20-50% of the geo footprint." Use a matched market or synthetic control design so the holdout regions look like the exposed ones before the test starts.
- Measure your own sales, not either dashboard. Read total revenue or orders in Shopify, and Amazon if you sell there, since several of the tests above found part of the lift on Amazon. The useful output is an incrementality factor: incremental conversions divided by platform-reported conversions. Haus's example is 300 incremental against 500 reported, a factor of 0.6. Multiply platform ROAS by that factor and you have a number you can budget on.
- Hold Meta steady. Same budget, same campaigns, same creative during the test window. If Meta spend moves, you are measuring two things at once.
- Run it long enough. The published tests ran two to four weeks, and MaryRuth's added a 2 week post-treatment window to catch late purchases. AppLovin says campaigns "need to be out of the learning phase and delivering steadily before a test starts." Haus notes that Meta's own Conversion Lift studies need at least seven days and at least 10% of the audience in each cell.
- Size the spend honestly. A test at a spend level you will never run tells you little. Cann's null result came at about $10K a week.
- Test the creative too. On AppLovin the creative decides who you reach. If you only run cut-downs of your Meta ads, you are likely to reach the people Meta already reached, and overlap goes up. That is why we run a native versus repurposed bake-off before the holdout, as described in our guide to incrementality testing for creative.
For the full framework, see how to prove a new channel adds demand instead of re-counting it, and for why the business metric is the one to judge on, measure creative on purchases, not on the ad.
Where Flixr fits
We are a performance marketing agency that runs AppLovin, Meta and TikTok and makes the creative for them, so we have a stake here too. We are judged on the client's own numbers, the same standard we are asking you to hold AppLovin to. The channel differences are in AppLovin vs Meta Ads.
Questions people ask
Does AppLovin count view-through conversions? AppLovin says its in-platform reporting "currently captures click-through attribution only." Meta's default setting counts both 7-day clicks and 1-day views, which is one reason the two can both claim a purchase.
Is AppLovin incremental for ecommerce brands? In most published geo tests it showed lift, but efficiency varied widely and at least one independent test (Cann, about $10K a week) found no significant lift. Almost all positive results are vendor-published. Test it on your own account.
How long should an AppLovin incrementality test run? The published tests ran two to four weeks, MaryRuth's with a 2 week post-treatment window. Start only once campaigns are out of the learning phase.
What is the difference between an iROAS and the ROAS in the AppLovin dashboard? Dashboard ROAS is attributed revenue divided by spend. iROAS is revenue that would not have happened without the ads, measured against a holdout, divided by spend. Only the second tells you whether the channel adds demand.
Sources: AppLovin, MaryRuth's case study; AppLovin, Ridge case study; AppLovin, "Measurement lens", 16 Jun 2026; Haus, "Early AppLovin results are in", 10 Dec 2024; Haus, "Understanding Meta incrementality testing"; BlueAlpha, Cann geo holdout, Feb 2026; Common Thread Collective, "Is AppLovin actually incremental? 8 tests", 15 Jan 2026; Common Thread Collective, discovery campaigns, 26 May 2026; Metaply, glass brand case study; Jon Loomer, "Meta Ads Attribution Setting: A Complete Guide".