Both failures present identically: you built something and nothing happened. They are distinguished by one question — of the people who genuinely understood what it does, what fraction tried it, and of those, what fraction came back? If almost nobody saw it, you have a distribution problem and rewriting the product will not help. If plenty saw it and did not return, more traffic makes the problem worse and more expensive.
It is live, it works, and nothing is happening. From the inside this is one feeling, and it is actually two completely different failures with fixes that point in opposite directions. Picking wrong costs a quarter.
A product problem means people saw it, understood it, and did not want it. A distribution problem means almost nobody with the problem ever found out it existed. The reason these get confused is that both produce the same dashboard: a flat line.
Why builders default to the wrong answer
Faced with an ambiguous signal, people reach for the explanation they know how to act on. If you build software, 'the product is not good enough' is a problem with a familiar shape — you can open the editor and start. 'Nobody has heard of it' requires doing something you have never done and probably do not want to do.
So the diagnosis follows the available treatment rather than the evidence. This is how a product that fifteen people have seen gets rewritten twice.
The one number that separates them
Qualified exposures. Not visits. Not impressions. The number of people who both have the problem you solve and understood what you built well enough to make a decision about it.
A thousand visitors from a launch aggregator is not a thousand qualified exposures. It is close to zero, because those people are browsing launches, not looking for your solution. Thirty conversations with people who have the problem is a far better sample than ten thousand pageviews, and it is the sample most founders skip because it is slower and involves talking to strangers.
The threshold is around one hundred. Below that, any conclusion about the product is unsupported. Zero conversions from twenty exposures and from two hundred are entirely different pieces of evidence, and only the second is evidence at all.
The diagnostic
Work through these in order. Stop at the first one that gives you a clear answer.
| Question | Answer | What it means |
|---|---|---|
| How many qualified exposures, honestly? | Under 100 | Distribution. Stop here; nothing below is interpretable yet. |
| Of those, what share tried it? | Under 10% | Positioning. They understood it and did not think it was for them. |
| Of those who tried it, what share came back a second time? | Under 20% | Product. They used it and it did not earn a return visit. |
| Do the ones who came back keep coming back? | Yes, a handful do | Product is fine for someone. Narrow the ICP to them and go back to distribution. |
| Can you name five users and what they use it for? | No | Distribution, and you are also flying blind on the product. |
The second row deserves a note, because it is the one people mislabel. Someone who saw it clearly and did not try it is usually not rejecting the product — they are failing to see themselves in it. That is a positioning failure, which is a distribution problem wearing product clothing, and the fix is words rather than code.
Return usage is the cleanest signal you have
If you only instrument one thing, instrument whether people come back.
Signups measure your landing page. A second unprompted session measures your product, because by then the copy has stopped helping and the thing has to be useful on its own. The gap between those two numbers is the most honest feedback available to you, and it does not require anyone to answer a survey.
Ten users where four return weekly is a better position than a thousand signups where nobody comes back twice. The first is a product with a small audience, which is a distribution problem and therefore solvable. The second is a landing page with a product attached.
The fixes, and why order matters
If it is distribution: stop building. Pick one channel and do it consistently for a month rather than four channels for a week each. Talk to people individually, at a volume that feels excessive, until you have your hundred. The instinct to improve the product while you do this is strong and worth resisting, because it makes the experiment unreadable.
If it is product: stop pouring traffic in. More distribution on a product people do not return to just raises the cost of the same information. Go to the handful who did come back, find out what they use it for, and consider whether that is a narrower and better product than the one you set out to build.
If it is both: fix distribution first regardless. Product changes cannot be evaluated without a flow of qualified users, so working on the product while distribution is broken is guessing with extra steps.
The uncomfortable part
For most technical founders it is distribution, and it is distribution for a reason that has nothing to do with the product: distribution is the work you avoided by becoming someone who builds things.
That is not a character flaw, it is a specialisation, and it stops being viable the moment you need someone other than you to use the thing. The hundred qualified exposures are not going to appear on their own, and there is no version of this where you skip them.
Operater exists for the specific case where the diagnosis is distribution and the reason it stays broken is that nobody on the team knows how to run outbound. The Sales agent works out who has the problem you solve, writes to each of them, handles the replies and books the call — which is the hundred exposures, arriving without you having to become a salesperson first.
It does not fix a product problem, and it is worth being plain about that. If people are seeing it clearly and not coming back, more of them seeing it will only get you the same answer faster.
Key takeaways
- Count qualified exposure, not traffic. A hundred people who understood the offer is a sample; ten thousand pageviews from a launch aggregator is not.
- The diagnostic threshold: below roughly 100 qualified exposures you cannot conclude anything about the product, because you have no sample.
- Return usage is the cleanest single signal. Signups measure your copy; a second session measures your product.
- The two fixes point in opposite directions. Distribution says ship less and talk to more people. Product says stop pouring traffic into a leaking bucket.
- Builders default to assuming a product problem because it is the failure they know how to fix, and that preference is why so many rewrite something nobody has seen.