Hypothesis
If performance ads in a set of matched regions carry the brand's distinctive cues early and prominently, with the idea, offer and hook unchanged, total conversions and branded search in those regions will rise over the test and the weeks after it, because people who recognise the name respond more readily to later ads and searches, unless the cues only raise cost per result. The decision behind it: carry the brand's distinctive cues early and prominently in performance ads, or keep them light.
The guide to branding in performance ads sets up the question. It asks for branding to be tested as a variable, with the idea, offer and hook held constant and only the cues varied, and for the result to be read on two clocks: cost per result, which the test reports directly, and memory, which shows up later as branded search and other results becoming easier. A platform split test can read the first clock by arm. It cannot read the second, because a branded search or an order from another channel days later belongs to no arm of a split run inside the ad platform. This design moves the split to regions, so both clocks are read per arm.
The reason to expect an effect comes from the distinctive brand assets framework, which reports the Ehrenberg-Bass Institute's argument that a cue does its job only when it is widely recognised and uniquely linked to one brand, and that fame grows through repeated exposure. The brand versus performance guide adds that a conversion-optimised ad works better on a person who recognises the name. Neither claims that branded performance ads win in any particular account, and this page claims nothing about effect sizes. The test is how an account finds out.
Metrics and guardrails
The primary metric is conversions per region from the business's own records: orders, leads or sign-ups counted where they are recorded, from every channel, not only those the ad platform attributes. Read them across the test period and a follow-on window fixed before launch, and compare the branded regions with what the light-cue regions predict they would have been, as the incrementality testing guide describes for regional designs. Using the business's own records matters because the effect the hypothesis predicts arrives partly through branded search, direct visits and later ads, which no single platform credits to the original exposure. The answer is incrementalConversions caused by marketing. in the sense the guide uses: what the branded version adds over the light version, region against region.
Three guardrails read the rest of the picture:
- Platform-reported cost per result in each region group. The short clock. The branding guide expects a branded variant may cost a little more per result; the guardrail records how much, so the decision can weigh it against the long clock. Read it from each platform's own reporting for each region group, which the creative testing guide treats as the short-clock result.
- Branded search impressions and clicks by region. The long clock. Paid branded search can be read by region where the business runs brand campaigns. For organic branded search, the Search Console Performance report can be filtered by country, as the Google Search Console guide explains, so it reads branded clicks by arm directly only when the regions are countries; within one country, read branded organic visits by location from site analytics and use Search Console for the overall trend, as the brand search holdout design does.
- Frequency per person in each region group. If one group sees the ads more often, the comparison is between exposure levels, not cues. Read FrequencyAverage times a person saw an ad. by region group from the platform throughout the run.
Where a brand tracker or a lift study exists, the brand awareness and lift guide describes how to read recognition of the assets alongside. Treat it as supporting evidence; the decision rests on conversions and branded search.
Design
Randomise at region level. Choose regions that can be targeted separately, as the incrementality guide sets out from the platforms' regional targeting, and whose conversions and branded searches moved together over a pre-period. Pair or match them on that history, then assign one region of each pair at random to the branded group. Every region group runs only one version of every ad.
| Design element | Choice | Reason |
|---|---|---|
| Unit of randomisation | Region, matched on pre-period conversions and branded search | Lets branded search and later conversions be assigned to one arm |
| Branded arm | The brand's committed cues early and prominently: in the first frames of video, in every crop of a feed image | The branding guide's "brand early without opening on the logo" |
| Light arm | The same ads with cues late or small, as the account runs today | The current practice, so the test answers the real decision |
| Held constant | Idea, offer, hook, budget per region, audiences, bidding and every other ad element | The creative testing rule: change one variable |
| Primary metric | Conversions per region from the business's own records, over the test and follow-on window | Captures effects that arrive outside the ad platform |
| Reading | Against a counterfactual built from the light-cue regions and the pre-period | The geo method the incrementality guide describes |
The light-cue regions are a comparison group, not a Holdout GroupA randomly chosen part of an audience kept from exposure to measure what exposure caused.: they see the same advertising at the same weight, so the test isolates the cues. Build both versions fresh for the test rather than pairing a new branded cut with ads that have run for months, because new creative can enjoy a Novelty EffectA temporary change in behaviour caused by a change being new, which fades as users get used to it. that would be credited to the cues.
The isolation rules come from the creative testing guide: one variable, the same body, variants in parallel. The ad creative hook test applies them to the opening on a platform split, which is the right design when the question is the short clock alone. Here the hook is the thing held constant, and the cue is placed in the same frame without delaying it, which is how the branding guide reconciles branding early with a hook that opens on the person's situation.
Spillover is the regional design's weakness. People commute, travel and see ads near borders, and national media reaches every region. Prefer regions with natural boundaries, keep the cues in every other channel identical across groups during the run, and note anything national that lands mid-test. The academy exercise on designing a brand cue test for a bed linen brand's performance ads practises this regional design for a brand also planning its first radio and podcast spend, with the metrics, guardrails, duration and decision rule set in advance.
Sample size reasoning
Power comes from the number of matched regions and how steadily their conversions and branded searches moved together before the test. The brand effect builds slowly, so the smallest difference worth acting on is usually small against the regions' week-to-week noise. The incrementality guide explains the region logic: regions are few, their history must move together for the comparison to mean anything, and the number of regions and the length of the window decide what the test can see. Use the pre-period to check, before launch, whether a difference of the size that would change the creative policy would stand out from the noise between matched regions.
If it would not, change the plan rather than running a test that cannot answer. Add regions where the account can target more of them, lengthen the test and follow-on windows, or concentrate the test on the formats where cues matter most, such as short video, so that the difference between arms is larger.
Duration
The run has three parts, fixed before launch. A pre-period long enough to show the regions moving together. A test period long enough for the cues to be seen repeatedly, since a cue builds memory only through repeated exposure. And a follow-on window, after the ads in both groups return to the same version, in which branded search and conversions are still read by region. No product launch, promotion or other campaign may land in one region group only; if something national lands, it lands on both groups and is noted.
Reading the results and the decision rule
Read the test at the end of the follow-on window, not at the end of the flight. Compare conversions per region in the branded group with the counterfactual from the light-cue regions, with the range around the estimate. Then read the guardrails: cost per result by group, branded search by region and frequency by group. Write the rule into the plan:
| Result | Decision |
|---|---|
| Conversions per region rise reliably over the test and follow-on window and branded search moves with them | Adopt the branded variant, even at a slightly higher cost per result |
| Conversions do not move and cost per result rises in the branded group | Keep the lighter cues |
| Conversions do not move and cost per result is level | Either version will do; keep the cues the brand guidelines require and stop testing this variable |
| Frequency differed materially between groups | Treat the result as unreadable and rerun with frequency matched |
Pitfalls
A different concept tested as branding is the commonest flaw. If the branded version also changes the scene, the script or the offer, the test measures the concept. The branding guide says this directly; hold everything but the cues.
Branded search read nationally loses the long clock. A national branded search trend cannot say which region group produced the change. Read it by region, from paid brand campaigns and site analytics where Search Console cannot separate the regions. The brand search holdout test faces the same reading problem from the other side, and its handling of paid and organic brand clicks by region applies here.
A follow-on window cut short throws away the reason for the design. If the business reads at the end of the flight because a decision is due, it gets a platform split's answer at a regional test's cost.
Frequency differing by region group turns a cue test into a weight test. Budgets per region must match the regions' size, and frequency must be checked weekly; if delivery drifts, correct it early rather than adjusting the reading afterwards.
Example: an invented sofa brand tests its brand cues by region (illustrative)
Zoplanck Sofas tests early brand cues by region
Zoplanck Sofas, an invented online sofa brand, runs performance video across 20 regions it can target separately. It matches them in 10 pairs on 16 weeks of orders and branded searches, and assigns one region of each pair at random to the branded group. Both groups run the same 30-second video with the same hook, a family arguing over the remote, and the same offer. In the branded group the brand's teal colour, its three-note sound and its name appear in the opening 3 seconds; in the light group they appear only at the end. The test runs for 8 weeks, followed by a 6-week follow-on window with both groups back on the light version.
| Measure | Branded regions | Light regions | Difference against the counterfactual |
|---|---|---|---|
| Orders, test plus follow-on window | 2,460 | 2,280 | +4% (range roughly +1% to +7%) |
| Platform cost per result | £41.20 | £39.80 | +3.5% |
| Branded searches per week, follow-on window | 1,930 | 1,700 | +9% |
| Average weekly frequency | 2.1 | 2.1 | Level |
The branded regions also sold a little more than their pairs in the pre-period, so the counterfactual built from the light regions and the pre-period is higher than the light regions' raw count. Against it, orders in the branded regions run about 4% above prediction, with a range that stays above zero, and branded searches move with them through the follow-on window. Cost per result is a little higher in the branded group, as the plan expected. Zoplanck adopts the branded opening for its performance video and plans to read the next tracker wave for recognition of the teal and the sound.
Common mistakes
Brand cue tests fail on what they hold constant and when they read. The experiment library holds the other creative and regional designs.
- Testing a new concept as branding. Only the cues may differ.
- Using a platform split for a two-clock question. It reads cost per result and nothing that happens outside the ad.
- Reading at the end of the flight. The long clock needs the follow-on window.
- Comparing new creative with tired creative. Build both versions fresh, or novelty is credited to the cues.
- Letting frequency drift. Unequal exposure turns the test into a weight test.
- Reading branded search nationally. Only a reading by region belongs to an arm.
