The short answer
To read a brand lift report, find the primary KPI and check it was fixed in advance, check the sample size and the smallest lift it could detect, read the full-sample result before any sub-group, convert relative lift to points, check whether the 95% range crosses zero, see how exposure was proven, count the exclusions, and ask for the raw data.
Key takeaways
- Start with the primary KPI on the full sample, not the highlights page.
- Convert every relative lift into percentage points and check the 95% range.
- A study with 200 people per cell can’t reliably detect lifts under about 13 points.
- No raw data, no exclusions and no proof of exposure means the headline can’t be checked.
The steps
- 1
Find the primary KPI
Look for the one metric the study was designed to move, such as consideration. Ask when it was fixed. If the headline is a different metric, or the KPI wasn’t named before fieldwork, treat the headline with caution.
- 2
Check the sample and the detectable lift
Find the number of people in the exposed and control cells. At a 30% baseline, 200 per cell reliably detects only lifts of about 13 points; 300 per cell detects about 10.5. A smaller real effect will usually read as no measurable change.
- 3
Read the full-sample result first
Ignore the highlights page and find the result for the whole sample. Sub-groups are small and testing many of them throws up chance results; only trust sub-groups that were fixed in advance and adjusted for multiple tests.
- 4
Convert relative lift to points
If the report says ‘+29%’, find the two underlying percentages and subtract them. A move from 10% to 13% is +30% relative but only +3 points.
- 5
Check the 95% range
Look for the range around the lift. If it crosses zero, the study can’t rule out no effect, whatever the headline says.
- 6
See how exposure was proven
Ask how the report knows the exposed group saw the ad: recognition of the real creative, a decoy ad that never ran, media data. Being surveyed where the ad ran is not proof.
- 7
Count the exclusions
Find how many people were removed and why: fraud, speeders, duplicates, failed attention checks, false exposure claims. No exclusions at all is a warning sign, not a good sign.
- 8
Ask for the raw data
A credible vendor will share the respondent-level data and show how each headline number was calculated. ‘Available on request’ should mean ‘attached’.
A worked example.
One report we reviewed led with ‘+29% purchase intent’. That came from people aged 25 to 34 only: 39 exposed against 32 control, with p = 0.20. On the full sample of 200 against 200, purchase intent moved +3.1 points, with an approximate 95% range of −6.7 to +12.9. The honest reading is no measurable change.
| How it was reported | How it should be read | |
|---|---|---|
| Headline | +29% purchase intent | +3.1 points |
| Base | 39 vs 32 people, one age band | 200 vs 200, full sample |
| Confidence | Not shown on the highlights page | Range −6.7 to +12.9, crosses zero |
| Verdict | ‘Strong uplift’ | No measurable change |
A one-page checklist.
| Check | Good sign | Warning sign |
|---|---|---|
| Primary KPI | Named and fixed before fieldwork | Headline metric chosen afterwards |
| Sample | Detectable lift stated up front | No mention of what the study could detect |
| Headline | Full sample, absolute points | Sub-group or relative lift |
| Confidence | 95% range shown | ‘Significant uplift’ with no range |
| Exposure | Recognition, decoy ad, media data | Surveyed in the venue, or ‘ad recallers’ |
| Exclusions | Counted, with reasons | None reported |
| Data | Raw data and crosstabs attached | ‘Available on request’ |
Questions
What does it mean if a brand lift result isn’t significant?
It means the study can’t rule out that the campaign had no effect on that metric. It doesn’t prove the campaign failed, especially if the study was too small to detect the size of lift you would expect. That is why the detectable lift should be stated before fieldwork.
Should I ignore sub-group results?
Not entirely, but treat them as clues rather than headlines unless they were fixed before fieldwork and adjusted for multiple tests. Small groups produce large swings by chance.