Conversion lift is the increase in conversions - purchases, sign-ups, app installs, leads - that a campaign causes, measured by comparing a group of people eligible to see the ads with a randomly chosen control group that is kept from seeing them. The difference between the two groups is the lift. The measurement exists because attribution reports credit ads with conversions that follow them, whether or not the ads made any difference.

The arithmetic

A study produces three core numbers. Absolute lift, which Google also labels incremental conversions, is treatment conversions minus control conversions after the control group is scaled to the same size. Relative lift divides that difference by the control group's conversions. Cost per incremental conversion, called iCPA by Google and CPIC by TikTok, divides total spend by the incremental conversions. When conversions carry values, the same logic yields incremental return on ad spend, or iROAS.

A retailer spends 45,000 dollars on a campaign aimed at one million people and holds back 10% as a control. The 900,000 people in the treatment group produce 9,900 purchases. The 100,000 controls produce 1,000, which scales to 9,000 for a group of 900,000. Absolute lift is 900 purchases, relative lift is 10%, and each incremental purchase cost 50 dollars. If the platform's attribution report credited the campaign with 6,000 purchases, its reported cost per acquisition was 7.50 dollars - nearly seven times lower than what the test found.

Most studies are intent-to-treat designs. Randomisation happens before delivery, so the treatment group includes people who were eligible but never saw an ad because they lost the auction or did not log in. That dilutes the measured effect, as PPC Land's explainer on incrementality notes, but it keeps the comparison clean.

How platforms build the control group

Three designs dominate. The oldest, the PSA test, shows the control group a public service announcement instead of the brand's ad. It keeps exposure symmetrical but costs money: a 10% holdout spends roughly a tenth of the budget on charity messages, according to PPC Land's explainer on the holdout study. Ghost ads, published by Garrett Johnson, Randall Lewis and Elmar Nubbemeyer in the Journal of Marketing Research in 2017, run the normal auction for control users, log the ad that would have won and show the next-best ad instead. The log identifies which controls would have been exposed, at no media cost.

Geo experiments split regions rather than people. Spend rises or stops in treatment areas while comparable control areas carry on, and a regression on pre-test data predicts what the treatment areas would have done. Jon Vaver and Jim Koehler published Google's framework in 2011. Google's help centre describes its geography-based Conversion Lift as more transparent, cross-channel and able to count offline conversions, but with a noisier group split, no demographic breakdown and generally higher budgets than user-based tests.

Platform implementations

Meta. Facebook introduced Conversion Lift in January 2015, and on September 4, 2015 added studies with multiple test and control cells so advertisers could compare strategies, according to Social Media Today. Meta's documentation blocks automated access, so these details are sourced indirectly. In Meta's lift tests control users pass through the same delivery pipeline and see the next-best ad, and a result is labelled highly conclusive when more than 90% of the posterior distribution sits above zero, according to PPC Land's holdout study explainer. Meta's "suite of truth" white paper drew on 307 Conversion Lift studies for 54 advertisers between March 2022 and November 2024, and found a median 31% undervaluation under rules-based attribution.

Google. Google offers Conversion Lift based on users and based on geography, alongside Brand Lift and Search Lift, inside the Experiment Center. The user-based version requires at least 1,000 observed conversions, excluding modelled supplementary conversions, and a campaign budget of 5,000 dollars, according to Google's setup guide. Holdbacks run from 1% to 50%, fixed at 30% when combined with Brand and Search Lift. Studies can last seven days, though 14 are recommended. Google announced the 5,000 dollar floor and a move to Bayesian statistics at Google Marketing Live on May 22, 2025. Its Bayesian help page lists a minimum of 150 treatment and 65 control conversions and reports 80% credible intervals built partly from priors drawn from previous studies in the same vertical.

TikTok runs its Conversion Lift Study as a managed service for eligible accounts, measuring web purchases, app installs and offline sales through its Events API, pixel, mobile measurement partners or TikTok Shop, according to its help centre. Its help page defines absolute lift as a percentage, which departs from Google's usage. TikTok also sells geo lift tests that raise spend in selected regions for a few weeks.

Pinterest lists conversion lift among its incrementality products. A seven-week study across 14 European markets for the Danish retailer Sostrene Grene recorded a 22% increase in incremental checkout conversions and an incremental ROAS of 1.3, according to Pinterest. Snap ran a month-long Conversion Lift Test for the lingerie brand Adore Me, measuring events from the Snap Pixel. Amazon has taken a different route: its self-serve catalogue of more than 50 third-party measurement products, opened on May 15, 2026, centres on brand lift and offline sales lift from vendors such as NCS.

Power, thresholds and certainty

Lift experiments are statistically expensive. Randall Lewis and Justin Rao analysed 25 field experiments with about 2.8 million dollars of spend in the Quarterly Journal of Economics in 2015. The median confidence interval on return on investment was more than 100 percentage points wide, and informative tests could require more than 10 million person-weeks.

Platforms manage this with minimums and graded language. Google shows study power between 50% and 95% and offers budget advice below 90%. Its certainty-of-lift bands treat 90% or more as a very good chance the ads caused the result, 70% to 90% as good, and 50% to 70% as directional. Below 50%, the interface displays "No lift", which Google says does not prove the ads failed. Google also introduced feasibility ratings on November 11, 2025, where High means a 60% to 90% probability of a conclusive result, and reported that studies shorter than 14 days showed drops of up to 17% in absolute lift for advertisers with long conversion lags.

Why it matters

Conversion lift is the main check on platform attribution. Brett Gordon, Florian Zettelmeyer, Neha Bhargava and Dan Chapsky compared observational methods with 15 Facebook experiments, covering 1.6 billion impressions, in Marketing Science in 2019 and found the methods often overstated effects. A 2023 follow-up across 663 experiments found median lift of 29%, 18% and 5% for upper, middle and lower-funnel outcomes, while stratified propensity score matching produced 173%, 176% and 64%, as PPC Land reported in September 2026. Lift results also calibrate marketing mix models and train automated bidding, including the incremental attribution setting Meta rolled out in June 2025.

Limitations and disputes

The sharpest criticism is structural. The company selling the media usually designs the test, holds the control group and computes the answer. The IAB and IAB Europe guidelines of November 3, 2025 place platform-reported incrementality in their weakest tier of evidence, "hybrid proxies". Karan Dhir, a measurement product lead at Genentech, proposed asking who holds the holdout and whether it can be audited. Vendor lift claims often arrive without method: a Lifesight report cited a 22.3% paid search conversion lift from connected TV without sample size, period or statistical approach.

Single-platform tests also see only one channel. A control user on Meta still sees the brand on YouTube and in search, so the test measures Meta's contribution on top of everything else, not in isolation.

Privacy rules cut the other way. Since Apple's App Tracking Transparency took effect in April 2021, conversions from iOS users who decline tracking reach platforms only in aggregated or modelled form. Google warns that user-based Conversion Lift "relies mostly on opt-in traffic" and does not support iOS app campaigns at all. TikTok counts app conversions from opted-in users only.

Google's guidance also reports that conversion-based bidding and data-driven attribution correlate with higher measured lift - advice that also steers spend towards its automated products.

Not the same as

Sales lift applies the same design to purchases, often offline and matched through panel or loyalty data. Conversion lift covers any tracked event.

Brand lift measures survey answers such as ad recall. Search Lift measures branded searches on Google and YouTube, and its US minimum is 10,000 dollars, according to Google. Neither counts transactions.

A/B testing compares two versions of an ad or page, with both groups exposed to advertising. It measures which treatment works better, not whether advertising works at all.

iROAS is an output of conversion lift when conversions carry value, not a separate method.

Recent developments

On October 7, 2026, Google made Search and Performance Max campaigns eligible for self-serve user-based Conversion Lift, previously available only through account representatives, according to Search Engine Land. Google Ads API v25.1, announced on August 19, 2026, exposed 24 Conversion Lift metrics to allowlisted accounts, though studies still cannot be created through it. Google also added a "cooldown" option measuring delayed incremental conversions for Demand Gen studies started on or after March 2, 2026, and in May 2026 announced Meridian GeoX, which feeds geo experiments into its mix model. The Trade Desk placed conversion lift enhancements in closed beta on August 27, 2026. Meta, in a March 3, 2026 announcement on click attribution, called Conversion Lift the "gold standard" of measurement while conceding that full adoption "can take time".

Timeline

  • December 2011: Jon Vaver and Jim Koehler publish Google's geo experiments methodology.
  • January 2015: Facebook introduces Conversion Lift.
  • September 4, 2015: Facebook adds multi-cell Conversion Lift studies.
  • November 2015: Randall Lewis and Justin Rao publish 25 field experiments in the Quarterly Journal of Economics.
  • 2017: Johnson, Lewis and Nubbemeyer publish ghost ads in the Journal of Marketing Research.
  • March 2019: Gordon, Zettelmeyer, Bhargava and Chapsky publish 15 Facebook experiments in Marketing Science.
  • April 2021: Apple's App Tracking Transparency takes effect.
  • 2023: Gordon, Moakler and Zettelmeyer publish 663 Facebook experiments in Marketing Science.
  • April 17, 2024: TikTok geo lift tests described by PPC Land.
  • May 22, 2025: Google announces a 5,000 dollar experiment minimum and Bayesian methods.
  • June 2025: Meta rolls out incremental attribution globally.
  • November 3, 2025: IAB and IAB Europe publish incrementality guidelines for commerce media.
  • November 11, 2025: Google adds feasibility ratings and confirms the 5,000 dollar threshold.
  • March 2, 2026: Google's cooldown measurement becomes available for new Demand Gen studies.
  • March 3, 2026: Meta announces click attribution changes, citing incrementality as the long-term goal.
  • May 5, 2026: Google announces Meridian GeoX.
  • May 15, 2026: Amazon DSP opens self-serve access to over 50 third-party measurement products.
  • August 19, 2026: Google Ads API v25.1 adds 24 Conversion Lift metrics.
  • August 27, 2026: The Trade Desk puts conversion lift enhancements in closed beta.
  • October 7, 2026: Google opens self-serve Conversion Lift to Search and Performance Max.

Summary

Who: Advertisers and agencies commission conversion lift studies; Meta, Google, TikTok, Pinterest, Snap and The Trade Desk run them on their own inventory, while Amazon routes lift measurement through third-party vendors.

What: A randomised comparison of conversions between a group eligible to see ads and a held-out control group, reported as absolute lift, relative lift, cost per incremental conversion and incremental ROAS.

When: Geo experiment methods date from 2011 and Facebook launched Conversion Lift in January 2015; Google lowered its minimum to 5,000 dollars in 2025 and opened Search and Performance Max to self-serve studies on October 7, 2026.

Where: Inside ad platforms' experiment tools, such as Google's Experiment Center, and in managed studies run by platform account teams, using pixel, server-side and offline conversion data.

Why: Attribution counts conversions that would have happened anyway. Conversion lift estimates what advertising caused, though seller-run tests, single-channel scope, statistical noise and privacy-driven signal loss limit what one study can prove.