Conversion Lift

A Conversion Lift study is a randomised controlled experiment that measures the causal, incremental impact of an advertising campaign on downstream conversions — purchases, sign-ups, or other goal actions. It is the closest available approximation to a "ground truth" for whether an ad campaign caused additional conversions to happen, rather than simply reaching people who would have converted anyway.

How it works

A Conversion Lift study randomly splits an eligible audience into two groups before ad serving begins:

Test group — exposed to the advertiser's ads as normal
Control (holdout) group — deliberately withheld from seeing the ads

The difference in conversion rates between the two groups over the study period is the measured lift — the conversions directly caused by the advertising. (Source: Registry — Conversion Lift — Google Meta TikTok Pinterest IAB — 2026-08-10, Google Ads Help, retrieved 2026-08-10)

Because both groups are randomly assigned before any exposure, they are statistically comparable on demographics, geography, prior behaviour, and device usage. Conversions that pre-date ad exposure are excluded. (Source: Web — Conversion Lift 2026-08-10, Triple Whale, 2026-03-02)

The fundamental distinction a Conversion Lift study (CLS) provides is causal, not correlational — unlike Multi-Touch Attribution (MTA) and Media Mix Modelling (MMM), which are rooted in historical correlation and credit allocation, CLS observes what actually changes when ads are shown versus withheld. (Source: Web — Conversion Lift 2026-08-10, Triple Whale, 2026-03-02)

Two study designs

Design	Experimental unit	Use cases
User-based	Individual users (randomised by aggregated attributes)	Demographic segmentation (age, gender), upper-funnel campaign types
Geography-based	Geographic regions (treated vs control geos)	Offline data integration; multiple campaign types in one study; when user-level randomisation isn't available

(Source: Registry — Conversion Lift — Google Meta TikTok Pinterest IAB — 2026-08-10, Google Ads Help, retrieved 2026-08-10)

Key metrics

All major platforms surface a consistent set of incrementality metrics from CLS results:

Metric	Definition
Incremental Conversions (Absolute Lift)	Additional conversions caused by ads — test group conversions minus control group baseline
Relative Conversion Lift %	Percentage increase in conversion rate vs the control group
Incremental Conversion Value	Revenue value attributed to incremental conversions
iCPA (Incremental Cost Per Action)	Total ad spend ÷ Incremental Conversions
iROAS (Incremental Return on Ad Spend)	Incremental Conversion Value ÷ Total Ad Spend
CPIC	Cost Per Incremental Conversion (TikTok terminology; equivalent to iCPA)

(Sources: Google Ads Help, retrieved 2026-08-10; TikTok For Business Help Center, April 2025 — via Registry — Conversion Lift — Google Meta TikTok Pinterest IAB — 2026-08-10)

Meta's API additionally surfaces conversions_CPiC (cost per incremental conversion) and buyers_incremental (incremental unique buyers) — separating conversion count from buyer count. (Source: Meta for Developers Marketing API, retrieved 2026-08-10)

Platform implementations (as-of 2026-08-10)

Google Ads

Two study types: user-based and geography-based (see above)
Multiple campaign types can be combined in a single geography-based study (Source: YouTube — Conversion Lift 2026-08-10, Google Ads channel, 2023-10-12)

In 2025, Google moved Lift Studies (Brand Lift, Search Lift, Conversion Lift) from a standalone section into the Experiments section of Google Ads, centralising lift measurement alongside campaign experiments (Source: Web — Conversion Lift 2026-08-10, ALM Corp, 2026-04-25) (as-of 2026-04-25)
Google lowered spend and conversion-volume thresholds in 2025, making the tool accessible at lower budget levels (Source: ALM Corp, 2026-04-25) (as-of 2026-04-25)

Conversion Lift is available for Display and Video 360 (DV360) as well as Google Ads proper (Source: Google Ads Help, retrieved 2026-08-10)
Bayesian statistical methods are used in at least some study types (Source: Google Ads Help topic index, retrieved 2026-08-10)
For Demand Gen-only studies, Google surfaces delayed incremental conversions — modelled estimates for additional conversions expected after the study ends (Source: Google Ads Help, retrieved 2026-08-10)
Access: not available to all accounts; advertisers must contact a Google account representative (as-of 2026-08-10)
Certainty of lift is shown as a percentage range (50–95%) in 5% increments; practitioners are advised to target 90% certainty for reliable results (Source: ALM Corp, 2026-04-25)
Google recommends conversion-based bidding, data-driven attribution, and action-oriented creatives to optimise for lift, and warns against: running studies shorter than two conversion windows, or stacking multiple studies simultaneously (Source: YouTube — Conversion Lift 2026-08-10, Google Ads channel, 2023-10-12)
A feasibility status indicator is available before a study launches, allowing advertisers to assess likelihood of statistically precise results at their campaign volume (Source: YouTube — Conversion Lift 2026-08-10, Google Ads, 2023-10-12)
CLS is particularly suited to upper- and mid-funnel campaign types (YouTube, Demand Gen, Performance Max) whose effects do not appear cleanly in last-click reporting (Source: ALM Corp, 2026-04-25)

TikTok

TikTok's product: Conversion Lift Study (CLS) — RCT design (test group sees ads, control does not); metrics: Incremental Conversions, Absolute Lift, Relative Lift, CPIC, iROAS (Incremental Return on Ad Spend), Confidence Level (Source: TikTok For Business Help Center, April 2025) (as-of April 2025)
Supported data sources: Events API (recommended), TikTok Pixel, Mobile Measurement Partners (MMPs), TikTok Shop (Source: TikTok For Business Help Center, April 2025)
Scope: single campaign or all TikTok ads from an account simultaneously
Access: exclusive managed service; account representatives design and implement studies (Source: TikTok For Business Help Center, April 2025)
Over 1,400 global advertisers have run CLS studies (as-of search retrieval 2026-08-10)

Published CLS results (as-of 2026-08-10):

Torrid (fashion): +4.03% lift in purchases, iROAS $2.41
Aerie (Search Ads + In-Feed Ads): >3% lift in conversions, 3,800 incremental purchases, iROAS $2.64

Study design: 7-week Conversion Lift Study combining always-on and flighted campaigns
Søstrene Grene (European lifestyle retailer, 14 EU markets): 7-week CLS using Catalogue Sales ads (prospecting + dynamic product retargeting) + Lead ads → 22% incremental increase in checkout conversions, up to 46% lift in checkouts from Catalogue Sales exposure, 21% lift in sales, iROAS 1.3x (Source: Pinterest Business case study, retrieved 2026-08-10)

Snapchat

Adore Me: +12% lift in new customer acquisition
Kiehl's (MENA): +145% incremental lift in purchases (full-funnel multi-product strategy)
Neustar meta-analysis (10 brands, retail/CPG/entertainment): 3.7× fewer conversions and 3.2× less revenue attributed to Snapchat under last-touch attribution vs multi-touch — demonstrating systematic under-measurement of Snapchat's incremental contribution (Source: Snapchat For Business / Neustar, retrieved 2026-08-10)
2025 Snapchat/Nielsen/NCSolutions meta-analysis (10 MMMs with NCS lift result priors): for brands where ROAS increased, average ROAS and effectiveness more than doubled → additional $17.1M revenue across those campaigns (Source: Snapchat For Business, 2025) (as-of 2025)

Amazon

Amazon Brand Lift specifics (not a CLS): survey-based, available for Amazon DSP (US, Canada, DE, ES, FR, MX, JP, UK) and Sponsored Ads (US only); free at qualifying spend/impression thresholds; results in ~10 business days; primary metric is absolute lift (difference in response rate between exposed and control survey groups).

Amazon meta-analysis (Q4 2022 – Q3 2024): combining display + video ads produces 2.2× higher brand awareness lift than video alone, and 2.5× higher lift vs display alone. (Source: Amazon Ads library, retrieved 2026-08-10) (as-of Q3 2024)

CLS vs MMM vs MTA

Dimension	CLS (Conversion Lift Study)	Media Mix Modelling (MMM)	Multi-Touch Attribution (MTA)
Causal strength	Strong (experiment-based RCT)	Moderate–weak (econometric)	Weak (correlation / credit rules)
Holistic scope	Low (confined to one platform/channel)	High (full cross-channel)	Medium (within tracked touchpoints)
Time horizon	Tactical (campaign/tactic level)	Strategic (quarterly+)	Continuous but retroactive
Cookie dependency	None — operates on cohort outcomes	None	High (especially MTA)
IAB classification	Experiment-based	Econometric	Hybrid Proxy

(Sources: IAB/IAB Europe Guidelines for Incremental Measurement in Commerce Media, November 2025; Web — Conversion Lift 2026-08-10, Stape.io, 2026-07-31; SegmentStream, 2026-02-23)

Incrementality testing (including CLS) is not affected by cookie restrictions, ITP, ad blockers, or cross-device tracking failures in the way traditional attribution is, because it operates on cohort-level outcomes rather than user-level tracking chains. (Source: SegmentStream, 2026-02-23)

The CLS + MMM combination

CLS gives "local truths" (what worked at a specific moment for a specific campaign); MMM gives "global structure" (how to deploy that same channel across the full mix over time). Using them together closes the gap between tactical proof and strategic planning. (Source: Stape.io, 2026-07-31)

CLS results are increasingly used as calibration inputs (priors or constraints) into open-source MMM tools such as Meta's Robyn, Google's Meridian, and PyMC-Marketing, so that MMM model coefficients reflect experimentally proven lift ranges rather than purely historical correlations. (Source: Stape.io, 2026-07-31) (as-of 2026-07-31)

As of a 2025 survey, 40% of marketers use incrementality results to calibrate their MMM models. (Source: Web — Conversion Lift 2026-08-10, cited via multiple 2025–2026 sources)

IAB/IAB Europe standards (2025–2026)

The IAB/IAB Europe Guidelines for Incremental Measurement in Commerce Media (November 2025) provide the industry-standard framework for evaluating incrementality methods. Key definitions and classifications:

IAB definition of incrementality (as-of November 2025):

"Incrementality measures the causal impact of marketing by identifying the additional business outcomes directly driven by a campaign or tactic, compared to what would have occurred in the absence of marketing activity."

Four-method taxonomy:

Category	Examples	Causal Strength	Holistic Scope
Experiment-based	RCTs, Holdouts, Ghost Ads, Matched Markets	Strong	Low
Model-based Counterfactual	Synthetic Control, ML propensity	Strong–Moderate	Medium
Econometric	MMM, Time-Series Regression	Moderate–Weak	High
Hybrid Proxies	New-to-brand %, platform-reported incrementality, simple MTA	Weak	Low

Platform-native Conversion Lift products (e.g. Google, Meta, TikTok holdout experiments) are classified as Experiment-based with "strong" causal strength — but explicitly noted as "usually confined to one platform" and prone to contamination in multi-platform scenarios. (Source: IAB/IAB Europe, November 2025)

Three requirements for causality (IAB):

A credible counterfactual or intervention
Control of bias (seasonality, pricing, promotions, competitor activity, omitted variable bias)
Separation of signal from noise (statistical robustness — confidence intervals excluding zero, bootstrapping, falsification tests)

Experiment-based = "gold standard" for validating ROI of commerce media investment and proving causal lift on specific products or promotions. (Source: IAB/IAB Europe, November 2025)

The IAB Europe Commerce Media Measurement Standards V2.1 (May 2026) states that a Retail Media|Commerce Media Network is not required to offer incrementality measurement to achieve IAB Europe certification, but if it does, it must use one of the methods specified in the November 2025 Guidelines. From August 2026 onwards, only V2 certification is available (V1 accepted until July 2026). (Source: IAB Europe, May 2026) (as-of May 2026)

Data quality and clean rooms

Browser-based setups can miss 10–30% of conversions due to ITP, short cookie lifetimes, ad blockers, and script errors; for CLS this is material because measurement bias between test and control groups distorts the lift estimate. (Source: Stape.io, 2026-07-31) (as-of 2026-07-31)

Where an advertiser cannot send conversion data directly to an ad platform (due to healthcare/financial regulation, internal data governance, or privacy policy), a data clean room allows exposure and conversion data to be matched without either party seeing the other's raw records; the clean room is torn down after the study. (Source: Decentriq, 2026-07-17)

Some walled-garden platforms (e.g. Amazon) run lift experiments entirely within their own ecosystem because they control both ad serving and, often, conversion data — a clean room is only necessary when exposure and conversion data sit with genuinely separate companies. (Source: Decentriq, 2026-07-17)

Limitations and criticisms

Seasonality and competitor contamination: external events, promotions, and competitor activity during the test period can contaminate test/control group outcomes; baseline tracking is recommended for several weeks before a study begins. (Source: Decentriq, 2026-07-17; SegmentStream, 2026-02-23)
Single tactic, single moment: a CLS covers only one specific channel or tactic at one specific time — it does not represent the full marketing mix. (Source: SegmentStream, 2026-02-23)
Setup cost and duration: significant time and resources required; ending a study early weakens statistical confidence and can produce misleading conclusions. (Source: Triple Whale, 2026-03-02)
Holdout revenue cost: the control group is withheld from advertising to generate the counterfactual; this creates an opportunity cost, especially for large holdouts. (Source: implied by practitioner literature on holdout sizing)
Single-platform confinement: IAB guidelines explicitly flag that platform-native CLS is "usually confined to one platform" and prone to contamination in multi-platform campaigns. (Source: IAB/IAB Europe, November 2025)
Audience overlap risk: significant overlap between the tested campaign and unrelated media running in parallel can blur the test/control comparison. (Source: Decentriq, 2026-07-17)

Contradictions

2026 emerging landscape

Three shifts are defining the 2026 landscape (Source: Web — Conversion Lift 2026-08-10, YouTube — Conversion Lift 2026-08-10, synthesis from multiple 2025–2026 sources):

Always-on lift — continuous synthetic-control tests (INCRMNTAL, Haus) vs fixed-window holdouts, trading statistical purity for eliminating the opportunity cost of withheld impressions
MMM calibration — CLS results used as priors or constraints in open-source MMM tools (Google's Meridian, Meta's Robyn, PyMC-Marketing) to ground econometric models in experimentally proven lift ranges; Google's open-source Meridian positions CLS outputs as the canonical calibration input
Privacy-preserving lift — multiparty computation (Meta's Private Lift Service; clean room providers) enabling CLS for regulated industries without exposing user-level data

Key terms

Term	Meaning
Conversion Lift	The incremental conversions caused by an ad campaign, measured via RCT holdout
Test group	Audience segment that sees ads as normal
Control/holdout group	Audience segment deliberately withheld from seeing ads
Absolute Lift	Total number of incremental conversions
Relative Lift %	Conversion rate difference between test and control, as a percentage of the control rate
iROAS	Incremental Return on Ad Spend — incremental conversion value ÷ total spend
iCPA / CPIC	Incremental Cost Per Action / Cost Per Incremental Conversion — total spend ÷ incremental conversions
Ghost Ads	Ads served only to test groups but not billed; control group sees a placeholder — a design variant of holdout testing
Synthetic Control	Statistical method creating a "synthetic" counterfactual from weighted historical data; used in always-on lift tools as an alternative to a clean holdout

What practitioners report

Roughly 60% of retargeting conversions are assessed as non-incremental across industry CLS studies — the gap between platform-attributed and truly causal conversions is widest on retargeting and brand campaigns. (Source: YouTube — Conversion Lift 2026-08-10, web synthesis, 2025–2026)

Ending a lift test early is a recurring practitioner mistake; running studies for the full recommended duration is consistently cited as critical. (Source: Triple Whale, 2026-03-02; Google Ads channel, 2023)
Over 50% of CPG marketers now treat incremental sales as a significant KPI for judging campaign success, per a Digiday/Circana 2025 report. (Source: Decentriq, 2026-07-17, citing Digiday/Circana 2025) (as-of 2025)

Next frontier (dangling links)

Concepts referenced within this page that do not yet have dedicated pages: miROAS · Geo Holdout Testing · iCMAM · Ghost Ads · Synthetic Control · INCRMNTAL · Haus · Meta Robyn · Google Meridian · PyMC-Marketing · Media Mix Modelling (MMM) · Multi-Touch Attribution (MTA) · Halo Attribution

Conversion Lift

Conversion Lift

How it works

Two study designs

Key metrics

Platform implementations (as-of 2026-08-10)

Google Ads

Meta

Meta Incremental Attribution (2025–2026)

TikTok

Pinterest

Snapchat

Amazon

CLS vs MMM vs MTA

The CLS + MMM combination

IAB/IAB Europe standards (2025–2026)

Data quality and clean rooms

Limitations and criticisms

Contradictions

2026 emerging landscape

Key terms

What practitioners report

Next frontier (dangling links)