Marketing test sample size: how many before you decide

Short answer: Decide the threshold before you look at the result: 50-100 units per channel, 300+ per message, three weeks per cadence.

Most marketing tests are decided after the data arrives, which means the decision is really an argument. Pre-registering the threshold is what makes a small sample say anything at all.

Sample-size floors that survive contact with reality

What you are judgingFloorWhy
Channel50-100 sends per variantBelow that, 3% vs 6% is inside the noise band
Message / copy300+ sendsCopy effects are small; small effects need volume
Cadence / sequenceThree weeksDay 1 vs day 5 shifts the most early on

The pre-registered kill rule

Write one sentence before the test: the number of units, the number of results, and what happens if it is missed. This removes the two most expensive failure modes in marketing:

  1. Killing a channel that works, because week one was small.
  2. Keeping a channel that never worked, because "the numbers are small, let's wait".

How to apply it this week

Pick the one channel you are unsure about. Write the kill rule and the keep rule with real numbers, and the date you will read the result. Do not touch the channel before that date. That single file is worth more than another dashboard.

Source

The rules above come from marketing-mindset - an open-source (MIT) skill that gives AI agents decision rules for marketing instead of templates. Full skill file, install instructions and the test-limits section are on the landing page.

Need this done for you? This skill is the method. The done-for-you version — offer, pages, tool, deliverability, tests with a written kill rule — is a fixed-scope package: marketing engineer, packages and prices (Sprint $900 · Engine $1,900/month · Full build $2,900).

Free browser tool from the same engine: two-proportion test size calculator · what the scope covers · how it is priced.