Brand lift study, explained: how out-of-home brand lift is actually measured

A brand lift study measures the difference in brand attitudes between people exposed to a campaign and a comparable group who were not, using a survey. For out-of-home, exposure has to be constructed, usually from device location, rather than read from an ad server, so the exposure definition and the control method come before the lift figure.

Out-of-home has no click, so its proof of effect comes from the brand lift study: a survey comparing people exposed to the campaign with people who were not. The idea is simple. The execution, in a medium where nobody logs the exposure, is where the useful questions are: what counts as exposed, where the control group comes from, how many respondents a lift needs, and what published results look like once the units are read.

What is a brand lift study?

A brand lift study is a survey-based comparison of two groups: people exposed to a campaign and a control group who were not. Both answer the same questions about the brand, and the gap between their answers is the lift. Cint's Lucid Measurement defines it exactly that way: a control percentage, an exposed percentage, and the difference between the two.

The questions cluster into five metrics; Kantar's LIFT Express lists awareness, favourability and purchase intent, Google's Brand Lift adds ad recall, association and consideration.

Two conventions for reporting the gap, and vendors mix them. Absolute lift is the difference in points; relative lift divides it by the control group's score. Cint's worked example: 52.39% of the control group aware, 55.05% of the exposed group, an absolute lift of 2.66 points, a relative lift of about 5%. A result quoted as +40% is almost always relative. Ask for both base rates; without them a percentage cannot be turned back into people.

How is exposure established for out-of-home?

Nothing in out-of-home records an exposure, so the exposed group has to be constructed. Three methods are in use: device location, where a phone observed inside a placement's viewshed marks its owner as exposed; claimed recall, where respondents who say they saw the ad form the exposed group; and modelled audiences, which estimate how many people were likely to notice but cannot name any of them.

Device location

The OAAA's measurement guide (March 2020) defines exposure as presence in a screen's exposure zone, or viewshed, while content is viewable, and notes that this does not require the content to have been seen. OUTFRONT Media's explainer (May 2024) gives the mechanics: measurement partners geofence the viewshed, a phone passing through is marked present by its mobile advertising identifier, and opted-in consumers from that set are surveyed.

Reveal Mobile, the exposure partner LoopMe named for out-of-home in May 2025, describes computing device-level exposure from media plans, ad rolls, experiential specs and moving waypoints, joined spatiotemporally to a persistent panel of 60 million US daily active users. LoopMe runs non-incentivised mobile surveys against that data. That is the device-based pattern in one sentence: a location panel supplies the exposed group, a survey vendor supplies the questions.

Claimed recall

Happydemics, which reports more than 1,300 DOOH studies since 2021, shows respondents the ad, asks whether they recall seeing it on a specific channel, and compares those who do with a control group. That measures the difference between people who remember the ad and people who do not. Memory correlates with attention and with existing affinity for the brand, so a recall-defined lift and a device-defined lift are not the same number, even for the same campaign. Both are legitimate. Label which one you have.

Modelled audience

Geopath impressions are the third thing, and the one most often confused with the first. Geopath defines impressions as the number of times people passing a display are likely to notice the ad; circulation only estimates opportunity to see, and Visibility Adjustment Index values "are not measures of audience". The model draws on aggregated data from more than 150 million mobile devices (Geopath's 2019 standards document), but its output is an audience estimate, not a roster of exposed people. It sizes the audience. It does not produce a survey sample. How the estimate is built, and why it changes for a moving unit, is on how out-of-home impressions are measured.

When the screen moves

On a fixed site the viewshed is a polygon drawn once. On a vehicle it travels with the unit, so exposure stops being a property of a place and becomes a distribution over the route, the hours and the dwell along it. Reveal's method acknowledges this by listing moving waypoints as an input. The consequence is that the exposed group is only as good as the route log behind it: two campaigns with identical impression counts can have entirely different exposed panels, one concentrated on a commuter corridor, the other spread thinly across a city. A lift study on inventory whose geography is a route should be briefed with the route data, not just the impression total. Firefly, whose LoopMe-measured taxi-top study is the third row of the results table below, lists brand study first among five offerings on its measurement page, with foot traffic, web conversion, app conversion and retargeting; the route log behind an exposed panel is the operator's to supply, so ask for it in the brief.

How is a control group built when nobody can be held out?

Out-of-home cannot serve a public-service message to a control group the way a digital campaign can, so the control is constructed after the fact. Four methods are in use, each answering a slightly different question, and the vendor should name which one it used.

The quiet problem is that an out-of-home control group is defined by absence of evidence. A device is in the control because it was not seen in the viewshed, and location panels are samples: 60 million daily active users is large and still not everyone. Some of the control was exposed and never logged. That pushes measured lift down, a conservative bias but a bias, and it is why the OAAA guide asks providers whether there is panel bias and whether data is normalised to census.

How large does the sample need to be?

Large enough that the margin of error is smaller than the lift. Cint's significance FAQ spells out the test: a two-tailed z-test on the difference in proportions, with the margin of error equal to the z-score for the chosen confidence level times the standard error. Margin smaller than absolute lift means significant. Cint defaults to 90% and calls 80% results directional.

Run that arithmetic on an ordinary study. Control at 40% aided awareness, exposed at 43%, 250 respondents per cell: the standard error of the difference is about 4.4 points and the 90% margin about 7.3 points, more than twice the lift. The same three points need roughly 1,500 per cell to pass 90% and just over 2,000 to pass 95%. A ten-point lift passes at 250. Small lifts on small samples are noise; large lifts on small samples are the ones worth checking for a broken control.

Three vendors, three thresholds. A lift reported without one is a number, not a finding.

What does a good out-of-home brand lift result look like?

Published out-of-home results range from a 13.3% average across thousands of studies to lifts above 40% in single-campaign case studies, and an average and a single campaign are not the same kind of number. The table lists the named, dated results this desk could verify against the publisher's own page.

Read the third row against the first. A +43% from one campaign and a 13.3% average across thousands are not comparable, and the case study, like most operator case studies, publishes lifts without base rates or sample per cell, so the figures cannot be converted into points or confirmed as relative. That is normal for the genre, and it is the reason to ask for base rates before a result goes into a deck. For scale, Cint's example of a typical US digital campaign is a 2.66-point absolute lift in awareness.

Brand lift, incrementality and attribution are three different questions

Brand lift measures a change in what people think. Incrementality measures a change in what they do that would not have happened without the campaign. Attribution assigns credit for an outcome to touchpoints. The OAAA guide draws the line explicitly: measurement summarises plays, impressions, reach and frequency, while attribution "utilizes in-flight and post-campaign analytics to demonstrate causality or correlation".

Meta's GeoLift documentation states the incrementality definition plainly: the difference between what was observed and what would have happened had the campaign not run. Programmatic buys report delivery by screen and hour, the exposure log an incrementality model needs; direct buys usually have to reconstruct one (see programmatic DOOH). Incrementality testing gets its own page on this site, incrementality testing; the short version is that it moves the question from attitudes to behaviour and inherits every exposure problem above.

How do you brief a brand lift study?

Brief the exposure definition, the control method, the sample per cell, the confidence level and who is paying, in writing, before the flight starts. A study designed after the campaign has no pre-period, no route log and nothing left to hold out.

1. Exposure definition. Device observed in a viewshed, claimed recall, or modelled? For moving inventory, what route data defines the viewshed, and who supplies it?
2. Control construction. Matched respondents, synthetic control, geographic holdout or pre/post, and which variables were matched on.
3. Sample per cell, not in total. Exposed and control separately, and again for any cut you intend to report.
4. Confidence level and test. 80, 90 or 95%, one- or two-sided, and whether results below the threshold will still be reported as lifts.
5. Base rates. Control and exposed percentages for every metric, so absolute and relative lift can both be computed.
6. Survey timing. When it runs relative to the flight, unaided before aided, and whether the creative is shown.
7. Who pays and who runs it. A study the media owner funds and staffs is still useful; label it as such, and hold the raw crosstabs.
8. Which benchmark. Norm set, channel, period, and whether the norm is a range or a point. Cint publishes a 95% interval around each benchmark and says to treat it as a range.

The OAAA guide's 2020 question list covers most of this in different words: is a control group utilised, how are baseline and lift determined, how are dwell time and exposure calculated, is there panel bias. Six years old and still the right list.

Where to go next

The category these studies measure is laid out in the guide to out-of-home advertising. The impression chain a lift study sits on top of, and why it changes when the screen moves, is on how out-of-home impressions are measured.