1
3 Comments

Every founder can name their best channel. Almost none of us can prove it.

I have been having the same conversation with three different founders this month, and it always arrives at the same place.

Ask someone where their users come from and you get a confident answer in under two seconds. Ask how they know, and the answer gets softer. It usually turns out to be one of these:

  • the dashboard says direct, so they went with the thing they remember doing that week
  • a channel got credit because it was the last click, which mostly measures who reminded the person, not who convinced them
  • they ran two things at once and gave the win to the one they enjoyed more

I have done all three.

Here is what made me stop trusting my own read. A channel can look enormous while it is being paid for, and the moment you stop paying, the number does not decay. It falls off a cliff. That tells you something honest and slightly brutal: it was never a channel, it was a receipt. You were buying visits, not building a way to be found.

The opposite also happens. A place that sends you five people a month for a year is easy to write off as noise, right up until you notice those five people behave completely differently from the hundreds. Low volume is not the same as low value, and a short measurement window makes them look identical.

What I have landed on, and I am not sure it is right:

Judge a channel by what it does when you stop feeding it. Anything that goes to zero the day you stop paying was rented. Anything that keeps trickling was earned. Both can be worth having, but only one of them compounds, and I think most of us are counting them in the same column.

The part I cannot resolve is timing. Earned channels take months to say anything, and a bad month looks identical to a dead channel. So you are choosing between quitting too early and pouring effort into something that was never going to work.

How do you personally decide a slow channel is dead rather than early? Not the theory, the actual moment you called it. I would like a rule I can apply on a Tuesday when I am tired, and I do not have one.

on September 9, 2026
  1. 1

    That is the cleanest rule anyone has given me, thank you.

    The bit I am stuck on is that your own comment holds the tension. The compounding channel does not show up in analytics for months, and the test is thirty days with zero spend. If both are true, the thirty day test will kill channels that were going to work and I would never find out which ones.

    Where I have got to since posting: the survival test is right and the window is the wrong unit. A channel that sends five people a month is not measurable in thirty days at the volume most of us are at. What IS measurable in thirty days is whether it sends anything at all without being fed. Zero is a verdict. Two is not, and two is where the interesting ones live.

    Your Threads example is the one that proves it. If the user arrived two weeks after the reply and you had already written the channel off, you would have counted it as dead on the exact day it was working.

  2. 1

    Attribution is the hardest part because the dashboard lies. We burned €150 on Google Ads — 4 clicks, 0 installs. First real user came from replying in a live Threads conversation two weeks later. The channel that compounds doesn't show up in analytics for months. My rule now: if it survives 30 days with zero spend, it's real. If it dies the day you pause ads, it was rented.

    1. 1

      Rented versus real is a good line and I have the receipt for it. 146 euro on Google Ads got us 9 sign-ups and exactly 1 payer. 412 dollars on Meta got 0 payers, ever. Both went to zero the day the budget did, so your rule holds.

      Where it got interesting when I ran it against 90 days of real referrers: Google 65, a job board 61, Hacker News 32. The Hacker News number is one comment I left months ago. It is still sending people and it has never cost anything.

      The part the rule does not catch, and I only found this yesterday: ChatGPT sent 26 people in that same window. There is no click to attribute and no referrer for most of it. I only learned we were being recommended because I asked it the buying question myself and read the answer back.

      So I would add a third bucket to yours. Rented dies when you stop paying. Real survives 30 days unpaid. And there is a kind that never shows up in analytics at all, which you only find by asking the machine what it says about you.