How many emails before you can trust a channel?

Short answer: 50-100 sends per variant before you judge a channel; 300+ before you judge a message; three weeks before you judge a cadence.

Twenty sends tells you that the plumbing works. It does not tell you whether the channel works.

The short answer

• 50-100 sends per variant before you judge a channel.
• 300+ sends before you judge a message.
• Three weeks before you judge a cadence.

Why 20 sends is not data

At 20 sends, a channel with a true 3% reply rate shows 0, 1 or 2 replies depending on the day. A channel with a true 6% reply rate shows 1, 2 or 3. The two ranges overlap almost completely, so the "winner" you pick is a coin flip dressed up as a result.

The same arithmetic runs the other way: a good channel killed at 20 sends usually gets killed in week one, when reply rates are at their most volatile anyway.

What to write down before you send

Pre-register the decision: "If fewer than 3 replies after 100 sends, the channel is out." Once it is written, a 20-send result cannot talk you into panic or hope. The thresholds are not magic - they are the point where the difference you care about becomes bigger than day-to-day noise.

Common objections

"We do not have that volume." Then you cannot judge the channel yet - test one variable at a time on low volume and expect to learn slower, not to learn the same things faster.

"Our deal sizes are huge." Volume floors scale down with deal size in principle, but small-n judgement stays unreliable; you compensate with more qualified data (calls, replies), not with confidence.

"The channel is obviously good." Then the threshold costs you nothing, because it will pass it.

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.