Hypothesis
If a reach campaign caps weekly frequency at a lower level, it will reach more unique people for the same budget and produce at least the same lift, because exposures beyond the cap add little for most people. The decision behind it is concrete: set a tighter frequency cap on a reach or video campaign, or leave frequency to the platform.
The hypothesis rests on a planning argument rather than a law. Research associated with the Ehrenberg-Bass Institute describes brands growing mainly by reaching more category buyers, which favours spreading impressions across more people; the idea of Effective FrequencyThe number of exposures a media plan assumes a person needs before an ad has its intended effect. holds that some repetition is needed before a message registers. Both can be true for different brands, messages and formats, which is why the cap is worth testing rather than copying. A cap can also raise the cost of each impression, because delivery has to find fresh people rather than serve the cheapest ones again, so a gain in reach is not free. The frequency and creative fatigue guide explains how FrequencyAverage times a person saw an ad. builds and what controls exist; this page designs the comparison.
Metrics and guardrails
The primary metric is lift per unit of spend. For an awareness campaign, that is Brand LiftThe difference in brand measures between people exposed to a campaign and a comparable unexposed group.: the difference between exposed and unexposed respondents on recall, awareness or consideration, measured by a lift survey, divided by the cell's spend. For a campaign with a conversion goal, it is incremental conversions against a holdout, divided by spend. The brand awareness and lift guide explains how lift studies are run and read, and the incrementality testing guide covers holdouts for conversions.
Three guardrails describe the trade-off. Unique ReachUnique people who saw an ad. per cell shows whether the cap did what the hypothesis expects: if the capped cell reaches no more people, the cap was not binding. Cost per thousand impressions shows the price of that reach, because a cap can make each impression dearer. The share of impressions above the cap level in the uncapped cell shows how much repetition the platform would choose by itself; if it is small, the cap changes little and the test will show little.
Design
Randomise at user level through the platform's split or lift study tooling, so each person is in one cell and each cell has its own unexposed comparison group. Where cells cannot be kept apart at user level, use matched regions instead. Both cells run the same creative, audience definition, placements, budget and flight dates; one has the frequency cap, the other has none or the platform's default.
| Design element | Choice | Reason |
|---|---|---|
| Unit of randomisation | User, via split or lift study tooling; region if users cannot be separated | Cells must not share people or the cap leaks |
| Cells | Capped versus uncapped, same budget and creative | One change: the exposure limit |
| Measurement | Each cell against its own holdout or lift survey | Lift is relative to the unexposed |
| Primary metric | Lift per unit of spend | Rewards the efficient exposure level, not raw volume |
| Cap level | Chosen before launch from planning, written in the plan | Prevents choosing the cap after seeing data |
| Flight | Same dates, clear of other heavy campaigns | Outside activity raises baseline exposure unevenly |
Where the cap is set depends on the platform. On Meta, the frequency control is documented for ad sets that optimise for reach and for reach and frequency buying; on DV360, caps can be set at campaign, insertion order and line item level and all apply, as the DV360 bidding and pacing guide explains. Set the cap at one level only for the test and record where. To choose a cap level and predict the reach each cell should achieve, the reach and frequency calculator for awareness campaigns turns budget, cost and audience size into expected reach and average frequency.
Sample size reasoning
Lift is measured against unexposed people, so each cell needs its own comparison group and enough respondents or conversions to see the smallest difference in lift that would move the cap. A brand lift difference between two exposure levels is usually smaller than the lift of the campaign itself, so this test needs more respondents than a single lift study. The incrementality testing guide covers sizing cells when a lift study is the measure; decide the smallest difference worth acting on before launch.
If the budget cannot buy enough respondents for two well-powered cells, test a bigger contrast between cap levels, run the test on a larger campaign, or read only reach and cost and decide the cap on planning grounds, recording that lift was not measured.
Duration
Run for at least the flight's planning cycle, usually several weeks, so frequency can build in the uncapped cell; a test that ends after a few days never reaches the exposure levels the cap is meant to prevent. Avoid flights that overlap other heavy-weight campaigns for the same brand, which add exposure the cells do not control. Fix the end date and the survey timing before launch.
Reading the results and the decision rule
Read the test once, at the end of the flight. Compare lift per unit of spend between cells with the interval from each lift study, then check reach, cost and the uncapped cell's frequency distribution. Write the decision rule into the plan before launch:
| Result | Decision |
|---|---|
| Capped cell has equal or better lift per spend and higher reach | Adopt the cap for campaigns like this one |
| Uncapped cell earns clearly more lift per spend | Remove the cap; repetition is doing work for this message |
| Capped cell reaches no more people | The cap was not binding; leave frequency to the platform |
| No reliable difference in lift, cap costs more per impression | Leave frequency to the platform, or test a different cap level |
A cider maker tests a weekly cap on a video flight
Calloway Bay Cider, an invented craft cider brand, plans a six-week video flight before summer with £40,000 split evenly between two cells. One cell caps frequency at two impressions per person per week; the other is uncapped. Each runs a brand lift survey. The capped cell reaches 1.9 million people at an average frequency of 4.1 over the flight; the uncapped cell reaches 1.4 million at 6.3, with a third of its impressions above the cap level. Ad recall lift is 5.8 points in the capped cell and 6.1 in the uncapped one, with overlapping intervals, and the capped cell's cost per thousand is higher. Calloway Bay adopts the cap for future awareness flights.
Pitfalls
Cells sharing an audience are the pitfall that voids the test, because a person in both cells sees the uncapped weight whatever the capped cell says. Use the platform's split or lift tooling, or regions, and never two ad sets targeting the same people side by side. Cross-platform frequency unseen is the second: the cap applies within one platform, and the same people may see the brand elsewhere, which raises their real exposure in both cells equally but limits what the cap can achieve.
Judging by click-through rate is the third. A cap usually improves clicks per impression, which looks like a win and says nothing about lift. Read the lift studies; use clicks only as colour.
Common mistakes
Frequency tests fail on their setup and measure more than on their arithmetic. The remarketing holdout test asks whether remarketing works at all and shares the holdout logic, and the experiment library lists every design.
- Running the cells as two overlapping ad sets. The cap leaks and the test reads nothing.
- Comparing reach without lift. More people reached is only better if lift per spend holds.
- Choosing the cap after the data arrive. Fix the level before launch.
- Running too short a flight. Frequency never builds in the uncapped cell.
- Testing a cap on a conversion objective where none exists. Use a reach objective or a different design.
- Reading click-through rate as the result. Caps raise it by construction.
