Brand Lift Surveys measure what a campaign changed in people's heads rather than what it changed in their behaviour. Where Incremental Lift observes whether exposed users visited a site, opened an app, or walked into a store, a brand lift study asks exposed users what they think of the brand and compares their answers to those of an unexposed control group.

The two products share the front half of the process and diverge at the outcome. Both identify who was exposed by joining location data to OOH viewsheds. From there, one watches where people go; the other asks them a question.

How a Study Runs

1. Exposure identification. Location data is joined to the campaign's OOH inventory to produce the set of devices with an opportunity to see the media. This is the same process used for all Mira measurement - see Sources and Exposure.

2. Control construction. A matched group of unexposed users is assembled to answer the same questions, providing the baseline against which the exposed group's answers are compared.

3. Fielding. A short survey - typically four or five questions - is served to both groups through our survey delivery partner while the campaign is live. Responses accrue over the course of the campaign.

4. Measurement. For each question, the share of respondents giving a positive answer is calculated within each group, and the difference between the two is reported.

What Can Be Asked

Each question maps to a brand KPI. The common ones are:

• Unaided awareness - can the respondent name the brand in its category unprompted?
• Aided awareness - do they recognise the brand when shown it?
• Ad recall - do they remember seeing advertising for it?
• Favorability - how positively do they regard it?
• Consideration - would they consider it for their next purchase?
• Purchase intent - do they intend to buy?
• Message association - do they connect a specific claim or attribute to the brand?

Questions take one of four forms: single-select, multi-select, a frequency scale, or a rating scale.

📘

Fewer, Better Questions

Every question added divides the same pool of respondents further, and each question is read on its own. Four or five questions covering the KPIs that matter most will produce clearer reading than a longer survey covering everything.

Configuration Considerations

🚧

Pre-Campaign Setup Required

A brand lift study must be configured before the campaign begins. Unlike footfall measurement, there is no way to run one retroactively - the responses simply do not exist unless they were collected while the campaign was live.

Two properties of survey measurement make timing more demanding than it is for other conversion environments.

Responses are actively solicited. Web pixels and SDK postbacks passively record events that happen anyway. A survey has to reach a person and get an answer out of them, which takes time and a live campaign.

Exposure data has to stay current. Because the survey is fielded to people who have already been exposed, the exposure dataset feeding it needs to be kept up to date throughout the campaign, rather than assembled once at the end.

Reach out to your account team as early as possible to get a study set up. Question wording, the number of questions, and the KPIs being tracked all need to be agreed before fielding starts, and none of them can be changed once responses are coming in.

Reading the Results

Structurally, a brand lift study is an Incremental Lift analysis with no pre period. Each question is a single comparison between an exposed group and a matched control group during the campaign, and the reported lift is the difference between them. There is no before-and-after measurement, because there is no way to survey someone about a campaign before it has run.

That has one consequence worth carrying into how results are read: without a pre period there is no baseline check confirming the two groups were comparable to begin with. The control has to carry that weight on its own, which is why how it is built and kept balanced matters more here than it does for behavioural measurement.

For each question, results are reported as:

• the percentage of positive responses in the exposed group and in the control group
• the difference between them, in percentage points
• the relative lift - that difference as a proportion of the control group's rate

A question where 12% of the control and 15% of the exposed group recall the advertising shows a 3 percentage point difference and a 25% relative lift.

Each question carries its own respondent count. Partial responses are retained rather than discarded, so someone who answers the first two questions and abandons the survey still contributes to those two. This preserves as much sample as possible, and it means later questions in a survey are typically read on fewer respondents than earlier ones.

Attrition Is Kept Balanced Across the Two Groups

People abandon surveys partway through, and the ones who stay are not a random subset of the ones who started - respondents who persist to the last question tend to be more engaged, and engagement is exactly the thing a brand lift study is trying to measure.

That drop-off is only a problem for the comparison when it differs between the two groups. If the exposed group finishes at a higher rate than the control, then by the final question the two cells are no longer alike, and part of the measured lift is just the difference in who stuck around. This is differential attrition, and it is the specific failure mode that turns ordinary survey drop-off into bias.

Mira checks that attrition is balanced between the exposed and control groups from the first question to the last. Equal drop-off from both arms costs sample size and therefore precision, but it does not distort the comparison, because whatever selection happened applied to both cells alike. Unequal drop-off is the case that has to be caught, and it is caught by comparing completion across the two groups rather than by looking at the overall completion rate.

📘

Why Not Simply Discard Partial Responses

Dropping every incomplete response would guarantee identical respondent counts across questions, at the cost of discarding real answers and shrinking every cell to the size of the smallest one. Retaining partials and verifying that attrition is balanced keeps the sample while addressing the reason the imbalance would have mattered.

Status

A survey moves through three states:

Feasibility

Brand lift studies have their own scale requirements, which differ by region. See Feasibility for the impression guidelines.