Read a Brand-Lift Report Before Revising the Treatment

The deck says brand lift came in weak. By the time that sentence reaches a director, it has usually shed the question it came from, the people it described and the comparison it rested on. What's left is a mood, and the mood is asking you to change something — tighten the opening, drop the music, hold the reveal a beat longer. You have a commercial to protect and no clear picture of what the number was measuring.

Read the report before you argue about the film. A lift label only becomes useful once you know what it is a difference between.

1. Recover the study behind the headline

"Brand lift" names a family of measurements, not one number.

The provider documentation behind this kind of report — Google's support page on Brand Lift measurement data, checked on 13 September 2026 — distinguishes a baseline positive-response rate from an exposed positive-response rate, and offers more than one way of expressing the difference between them. The difference is the lift. Everything that makes it mean anything lives outside the sentence "lift was weak": which question produced the positive response, which people answered, what counted as exposure, over what window, and against whom.

So ask for five things before you interpret anything.

• The question. Recall of the ad? Unaided awareness of the brand (naming it without being shown a list)? Consideration? Favorability? Any of these can be reported as "lift," and they are not interchangeable.
• The measured population. Not your intended audience — the people the survey actually reached, in the market, language and age band the study covered. If the brief was for one market and the fielding happened in another, the number still means something, but not the thing you assumed.
• The exposure definition. Who counted as exposed, on which platform or placement, verified how, within what window.
• The comparison group. Usually people who were not exposed — which includes people who saw the ad somewhere the study didn't track. That group is a practical approximation of a control, not a pristine one, and it's worth knowing how loose it is.
• The window and the study status. Measured when, relative to the campaign, and whether the report is a finished read or an interim one.

Ask too which creative and which versions were in the field. Campaigns often run a hero film, two cutdowns and a vertical crop. If the report covers the campaign, that is a campaign-level measurement, and your two-minute hero sits inside it.

If what you have is a photo of a bar chart with the axis label cropped and a slide title reading "weak lift," you don't have enough to change anything. Ask the study owner for the questionnaire and the context page. That request isn't obstruction; it's the minimum for a reading.

2. Put the two rates beside the label

Once you have them, put the baseline and the exposed rate on the same line, and only then look at the word "lift."

Here is an invented pair to work with. These are made-up rates, not a campaign. In this fiction, 40% of the comparison group gave the positive response, and 46% of the exposed group gave it.

Those two numbers can be described three ways.

All three describe the same pair of rates. They are not three effects, or three pieces of evidence, or a small one and a big one and a middling one. The provider's page documents more than one of these summaries, and a cropped chart will generally show whichever one travels best in a deck. Your job is to put the denominator back on it.

State the absolute figure first, in percentage points, because it's the only one that can't be mistaken for a percentage increase. Six percentage points is not "six percent more." It's a six-point gap between two rates.

The other two framings depend on where the baseline sits, which is why the baseline is part of the number. Move the baseline down to 10% and the same six-point gap becomes a 60% relative lift. Move it up to 70% and the same six points takes a fifth of the remaining headroom and reads as under nine percent relative. Same evidence, three very different impressions. None of the framings is dishonest on its own. The pair of rates is the fact; the framing is a choice about emphasis, and someone made that choice before the slide reached you.

This is also where "weak" has to be pinned down. Weak compared with what — another campaign, another market, a threshold the provider applies, a number someone in the room remembers from last year? A relative lift isn't obviously weak or strong without the category, the question and the design. Ask what the label was measured against. If the answer is "the usual benchmark," ask to see it.

3. Keep the uncertainty and the status attached

The estimate travels with qualifications that are easy to lose and expensive to lose.

Find the response counts. A six-point gap between groups of four hundred and a six-point gap between groups of forty thousand are not the same object, even when the arithmetic is identical.

Find whatever the report says about uncertainty and keep it in the same sentence as the estimate. Google's page carries its own wording for confidence and for how much data it considers usable. Treat that wording as a status label belonging to that report and take it to the study owner, rather than importing a convention from some other report or reading a simplified phrase as a general lesson in statistics. The page's suggestions about sample size and spending belong to the provider's conventions too, not to your reading of this result.

Find the study status. Interim, final, partial, still collecting. An interim read on a campaign that's still running is a different object from a closed one.

And read the insufficient-data lines carefully, because they're the ones most often translated into something they don't say. "Not enough data" is not a measured zero. If a version, a segment or a question comes back insufficient, that's the absence of a measurement — not a measurement of absence. Don't let it become "the cutdown with the dog scored nothing." In the same way, a study-level result describes the study-level population. It doesn't automatically describe a small subgroup inside it, and it certainly doesn't describe a scene inside the film. Small subgroups are where study-level numbers most often get borrowed for claims they can't carry.

4. Separate the measured response from the proposed cause

Now compare the object the study measured with the claim the room is making.

The study measured a survey-response difference between two groups of people. It did not measure sales, and it did not measure the quality of the film. A brand-response question is not a commercial outcome, and a study-level comparison between exposed and unexposed people doesn't isolate a line, a reveal, a performance or a music cue.

The claim usually arrives in causal clothing. The music cue is why recall stayed flat. The reveal is why consideration didn't move. Those are hypotheses about a mechanism inside the creative. The report's structure tells you how far it could support one.

The useful question: what would it take to tell that claim apart from the other explanations? If only one version of the film was in the field, the study contains no comparison between a cut with the cue and a cut without it. Nobody in the measured population saw both and told you which worked. Other explanations are available and unexcluded — placement, frequency, the market's existing awareness of the brand, the timing of the fielding, the wording of the question, or ordinary variation between two samples. The report as written doesn't rank them. Scene-level attribution needs a design that varies the scene, and most brand-lift studies don't.

That doesn't make the film irrelevant or the report worthless. It means the report is a measurement of a survey response in a defined population, and your scene is a hypothesis you're holding in the same hand. Keep them labelled. Don't invent what viewers thought. "People found the ending confusing" is not something the survey said unless the survey asked it, and even then it's self-report, not the reason the lift landed where it did.

5. Turn the reading into one answerable next question

Bring the creative back in here, because you're the person in the room who can describe what the film actually does. Use that to form the question, not the verdict.

Two follow-ups earn their keep. The first is about the versions: which cuts were measured, and is the version you're being asked to change the version that was in the field? If the hero film and the cutdowns ran together, ask whether the report can separate them at all, or whether the study was only ever designed to read the campaign as a whole.

The second is about the question. Does the survey question match the proposition the revision is meant to improve? If the revision is supposed to make the film feel warmer or more human, and the survey asked about unaided awareness of the brand name, then improving the film and improving the answer are not obviously the same job. If the revision is meant to raise consideration, find out whether consideration was asked, and whether enough people answered it to say anything.

Then hand the investigation to the measurement owner and agree who does what. Who pulls the version list? Who reads the questionnaire? Who decides whether a further read is warranted? That's a bounded, answerable piece of work, and it's not what the room was about to do.

What it isn't: a redesign of the experiment, a certification of the result, an optimization plan for the campaign, or a promise that your preferred alternative will raise lift. You're choosing an investigation, not a winner.

The sentence to take into the room

You can end the conversation with one sentence, provided you know which parts of it are named and which still have to be filled in. Here is that sentence built from the invented rates above — the same fictional 40% and 46%, still attached to a study nobody here has seen.

"On the recall question, among the measured population, the exposed rate was 46% against a 40% baseline — a difference of six percentage points, or 15% relative, in a report whose design I haven't checked and which says nothing yet about which scene carried it. Can you tell me which versions were in the field, and whether the survey asked about consideration at all?"

Two slots in that sentence are still empty, and neither can be filled from memory: the response counts behind each rate, and the report's own wording for confidence and usable data. Section 3's rule doesn't stop applying because you're about to speak instead of write. Say the counts out loud in the same breath as the estimate, quote the report's own status wording rather than a paraphrase of it, and if you don't have either yet, say that too — a sentence that announces what it's missing is still doing its job.

That is a quantity, a population, an honest limit and a next question, with two blanks left visible for the report to fill. It doesn't diagnose the film and it doesn't accept the label. The number's job is to sharpen the question you put to the creative, not to answer it for you.

Frequently asked questions

What should be recovered before a 'weak brand lift' headline drives changes to a treatment?

Ask for the question that produced the positive response, the measured population, the exposure definition, the comparison group, the window and study status, and which creative and versions were in the field. A cropped chart and a slide saying 'weak lift' are not enough to change anything.

Why put baseline and exposed rates on the same line before interpreting 'lift'?

They are the fact; the framing is emphasis. The same 40% and 46% example can be described as 6 percentage points absolute, 15% relative or 10% headroom. State absolute points first because six percentage points is not six percent more, and pin down what 'weak' is compared with.

How should uncertainty, response counts and insufficient-data lines be handled?

Keep response counts and the report's own uncertainty wording in the same sentence as the estimate. Note whether the study is interim, final or partial. 'Not enough data' is not a measured zero; it is the absence of a measurement. Study-level results do not automatically describe small subgroups or scenes inside the film.

Can a brand-lift report identify the music cue or reveal as the cause of the result?

No. The study measures a survey-response difference between exposed and comparison groups; it does not measure sales or film quality and does not isolate a line, reveal, performance or music cue. Those are hypotheses about a mechanism, and most brand-lift designs do not vary the scene to test them.

What is a useful next question instead of accepting or rejecting the label?

Ask which versions were in the field, whether the version being revised was measured, and whether the survey question matches the proposition the revision is meant to improve. Then hand a bounded investigation to the measurement owner. The sentence should state quantity, population, honest limit and next question, and leave blanks for response counts and confidence wording.

Further reading

• Is Retention Improving—or Are You Comparing Cohorts at Different Ages? — Read this when a blank cell or an insufficient segment is about to be read as a zero: it separates observed, incomplete and unknown cells before any comparison.
• Choose Evidence That Answers the Claim on the Slide — Read this when the room uses a survey response to argue about the film: it checks whether the evidence type actually answers the claim being made.