How Many Responses Do We Need for Our Brand Lift Study?

Richard Harless, Founder & CEO of Feather · May 2026

It is the single most common question I get. And most industry answers fall into one of two camps, oversimplified ("you need 1,000") or buried in statistical jargon nobody outside a research team understands.

Here is the context most people do not have. Platform brand lift studies can require six-figure minimum spends. To detect a one-point absolute lift reliably, you may need 20,000+ survey responses.

Compare that to 2011, when the IAB's Best Practices for Online Ad Effectiveness Research recommended aiming for 200 respondents, or 100 for smaller sites. That was the standard for measuring whether digital advertising worked.

The statistics have not changed. What changed is the size of the lift being measured, and who is paying to measure it.

The maths is the same. The context is not.

At 95% confidence, margin of error (MOE) still works the same way it always has.

• n = 100 → ±10% MOE (directional only, flag the uncertainty)
• n = 300–400 → ±5% MOE (sufficient for most product and marketing decisions)
• n = 1,000 → ±3% MOE (tight enough for high-stakes, high-confidence decisions)

A note on what that margin covers. ±5% is the precision of a single percentage, such as the share of exposed users who recall the ad. Lift compares two groups, so its margin is wider. With about 400 responses in each group, a lift is measured to within roughly ±7 percentage points. That is why we size studies per group, and why we label every result Significant or Directional rather than reporting a single headline number.

Platform BLS is built for advertisers where a one-point lift is worth proving with extreme confidence. The minimums make sense in that context. But the industry has quietly drifted into a place where the default precision is calibrated for billion-dollar media decisions, even when the actual question in front of a marketer is much simpler.

What are you actually deciding?

Most marketers just need to know. Did this creative outperform the other one? Did the message land? Is it worth running again?

For those questions, a 300–400 response study with a ±5% margin of error is not a worse version of a 20,000-response study. It is a different tool for a different decision.

The question worth asking is not "how big is our sample." It is "what business decision does this sample need to support, and is the margin of error tight enough to support it?"

A practical framework

If you are deciding whether to kill a creative or scale it, ±5% at 300–400 responses gives you a clear signal. If you are defending a $5M media buy to a CFO, ±3% at 1,000+ responses is the right standard. If you are testing a new message on a small campaign, even 100 responses can tell you whether you are directionally on track, as long as you are honest about the uncertainty.

The problem is not sample size. The problem is using the wrong precision standard for the decision at hand.

What this means for how Feather designs studies

We start with the decision, not the methodology. If you need to defend a major buy to leadership, we will field enough responses to give you confidence at that level. If you need fast feedback on creative performance, we will design a lighter study that gives you a clear directional read in days, not weeks.

The cost difference is significant. A 400-response deterministic study costs a fraction of what a platform-mandated 20,000-response study costs. And because it is tied directly to your media delivery, exposed responses come from people who actually saw your campaign, and control responses from the same environment, not a recruited panel guessing at exposure.

Precision should match the decision. Not the platform's default.

If you are not sure what sample size you need, start with the decision. We will work backwards from there.

Feather runs brand lift studies across mobile, CTV, sponsorship and AI answers in Europe. feathermetrics.com