Base·Note Data

Blog · Incrementality · Meta · Google · TikTok · Amazon

Incrementality testing across ad platforms: an overview

How Meta, Google, TikTok, Amazon, Pinterest and Snapchat run incrementality tests, compared side by side, with what each one reports and where it falls short.

Incrementality testing by platform: Overview · Meta · Google · TikTok · Amazon · Pinterest · Snapchat

Every major ad platform runs incrementality tests the same basic way: hide ads from a random group, then compare its conversions with the group that could see them. They differ in who runs the test, how big it has to be, and what you get back.

The idea in one picture

A holdout test compares people who saw ads with people who did not Eligible audiencerandomly split Test groupsees your ads Control groupads are withheld Lifttest minus control Same window and same conversion tracking for both groups. The gap is what the ads caused.

Every platform below follows this pattern. They differ in who runs the split, how big the holdout is and what they report.

What the numbers look like

Say you spend $45,000 reaching 1 million people and hold back 10% as a control. The 900,000 people who could see ads make 9,900 purchases. The 100,000 held-out people make 1,000, which scales to 9,000 for a group of 900,000. The gap is 900 incremental purchases: a 10% relative lift, or $50 per incremental purchase. Platform attribution might credit 6,000 purchases, which implies $7.50 per purchase, nearly seven times lower than the test found (PPC Land).

How each platform tests

  • Meta runs randomized, people-based Conversion Lift studies in Ads Manager, offers free open-source GeoLift for regional tests, and added modeled Incremental Attribution in 2025.
  • Google offers user-based and geo-based Conversion Lift, with Bayesian analysis and a power estimate before you start. Eligible Search and Performance Max campaigns went self-serve in October 2026, and Google tightened its data requirements in November.
  • TikTok runs Conversion Lift Study as a managed service: you share conversion data, and its measurement team designs, runs and reads out the study. Unified Lift combines it with a brand lift survey.
  • Amazon has no self-serve conversion lift test. It offers Brand Lift surveys, Amazon Marketing Cloud (AMC) for holdouts you design, and a modeled Long-Term Sales metric that it expanded at unBoxed 2026.
  • Pinterest runs Conversion Lift Studies with a holdout (20% in one published case) and relies on partners for geo tests.
  • Snapchat runs Conversion Lift Tests measured through the Snap Pixel, over about a month in its published cases.

How each one runs

PlatformMain testWho sets it upControl groupLength and size guidance
MetaConversion Lift (people-based)Experiments in Ads Manager if your account qualifies; otherwise a Meta repRandom holdout, commonly 10–20%Meta recommends at least $5,000 and 28 days
GoogleConversion Lift, user- or geo-basedGoogle Ads; self-serve for eligible Search and Performance Max since Oct 7, 2026; not every account has accessHoldback of 1–50%; held-out users see the next ad in the auction7 days minimum, 14+ recommended; 1,000 observed conversions and a $5,000 budget
TikTokConversion Lift Study; Unified Lift adds brand liftTikTok account rep (managed service)Random test and control; both still see your other marketingMinimum spend depends on objective; amounts not published
AmazonBrand Lift and AMC analysesAds console (Brand Lift); you or a partner (AMC)Exposed versus held-out shoppers you definePractitioner guides suggest 3–4 week holdouts
PinterestConversion Lift StudyPinterest repHoldout, 20% in a published case7 weeks in a published case; requirements not published
SnapchatConversion Lift TestCampaign Manager Measurement section (per a third-party guide)Randomized control groupAbout a month in a published case

What you get and where it stops

PlatformMetrics reportedGeo or offline optionBiggest catch
MetaLift %, incremental conversions, confidence interval, iROAS, cost per incremental conversionGeoLift (free, open source)Measures Meta alone; stopping early or changing budget mid-test weakens the result
GoogleIncremental conversions, relative lift, incremental value, iCPA, iROAS; study power of 50–95%Geo-based Conversion Lift counts offline conversionsStrict data setup: GA4-imported conversions aren't supported; user-based results are a lower bound
TikTokIncremental conversions, lift, CPIC, iROAS, confidence levelGeo lift tests reported by PPC Land; not on TikTok's own pageNot self-serve; app conversions come from opted-in users only
AmazonBrand Lift survey results; AMC query output (you compute lift and iROAS); modeled Long-Term SalesGeo and product-group holdouts run by advertisersNo built-in conversion lift; Long-Term Sales is a model, not a test
PinterestIncremental checkouts, sales lift, iROASGeo pause-to-measure tests through partnersLittle public documentation; details come from case studies
SnapchatConversion lift, incremental conversionsThird-party geo holdoutsLittle public documentation; details come from case studies and third-party guides

Reading the results with care

  • Most platform tests are intent-to-treat: people eligible for ads who never saw one stay in the treatment group, which dilutes the measured lift.
  • The company selling the media usually designs the test, holds the control and computes the answer. PPC Land reports that IAB and IAB Europe guidance from November 3, 2025 puts platform-reported incrementality in its weakest tier of evidence.
  • Each test sees one platform. A Meta test still measures against exposure on YouTube and search.
  • Results are ranges. Lewis and Rao found the median ROI confidence interval across 25 field experiments was more than 100 percentage points wide.
  • A result showing no lift doesn't prove the ads failed. Google notes it can also mean the study was underpowered.
  • Attributed numbers keep shifting under you. From November 12, 2026, Meta's Pixel counts some conversions measured on customers' own devices by default, which can raise reported results without any change in sales (PPC Land). A lift test is how you check whether more credited conversions means more sales.

I design and read incrementality tests for brands, on the platforms and independently of them. See how a Measurement Audit works.

Sources

More posts Services and pricing

Work with me

Want a second set of eyes on your measurement?

I help marketing and growth teams figure out what their spend is actually driving, with MMM, incrementality testing and attribution that finance trusts too.

Book a 30-min call