"Nobody pays" is three different problems wearing one sentence, and they have three different cures. Before changing anything, separate them: almost nobody arrives, or they arrive and leave, or they sign up and never pay. Then read the published report below on a product of exactly this shape — a narrow workflow tool for freelancers — because the two areas it is least sure about are the two that decide the answer, and neither is a feature.
Separate the three before you change anything
An app builder collapsed the build from months to a weekend. It did not collapse anything else, and the parts it did not touch are where the money is. So the first move is not a redesign; it is deciding which of these you are actually looking at.
| What you see | First thing to suspect | Also fits the same symptom | What separates them |
|---|---|---|---|
| Almost nobody arrives | Distribution, not demand — nothing has been tested yet | People are arriving and you cannot see them: a blocked analytics script, a redirect, or a channel counted somewhere you are not looking | Load the page yourself from a phone on mobile data and check you show up in your own numbers. Only once the counter is known to work is a small number evidence of anything |
| They arrive and leave | The promise does not match what they came for | The promise is fine and the people are wrong — a channel that sends browsers rather than buyers will bounce off an accurate page | Split the numbers by where they came from. One bad source with a good rest is an audience problem; an even bounce across every source points back at the page |
| They sign up and never pay | The problem is real but not expensive enough to buy a fix for | It is expensive enough and you never asked — a free tier that solves it well enough removes the reason to upgrade, and sometimes the user is not the person with the budget | Put a price on the page and ask for it early. A pitch that is refused is a demand finding; a pitch that was never made is not |
A split, not a measurement — argue with the boundaries if you like. Read the last column first: each symptom fits more than one cause, so the middle two are hypotheses to separate rather than readings to accept. What the split is for is that only the bottom row would be evidence about the idea; the top two would be evidence about the marketing, and the three lead to three different weeks of work.
The top row is the common one and the one people skip past, because it feels like an admission that nothing has been tested. It is exactly that, and it is good news: a product nobody has seen has not been rejected.
What a report on this kind of product actually says
Here is a real single-idea report — InvoiceRescue, a tool for chasing late invoices. Its verdict, scores and confidence columns are exactly what the run produced; the one edit made to it is disclosed further down. It came back PILOT_FIRST at 73. Read the card weakest-first, which is how a diagnosis is read:
The report's own scorecard, weakest area first, with the confidence it attached to each rating. The two areas it marked low confidence are the two at the bottom — and confidence is a separate statement from the score.
Two things in that chart matter more than the total.
The first is the floor: Pain + Demand Evidence at 5 out of 10, while 5 other areas tie at 8. Everything the report likes about this product is an execution property — margin, support load, whether the buyer trusts it enough to self-serve. Those are the things a competent build gets right, and a competent build is what you already have.
The second is the confidence column. The areas rated low confidence are Pain + Demand Evidence and Distribution Fit — which is to say, the report is least sure about exactly the two areas that decide whether money arrives. A low score with high confidence is a problem you can plan around. A middling score with low confidence means nobody yet knows, including the thing that wrote the report.
Both of the fixes it names are evidence, not features
The report carries its own list of what would move its verdict upward. There are 2 entries, taken from it rather than summarised:
From the published report's own "what would change my score" list. One disclosure, because the page's argument is that the list is the engine's: on that public sample, two improvement lines that recommended interviewing freelancers were rewritten into the async wording the product itself requires, and one of them is quoted here. The substance is the engine's; the channel was ours to correct. Notice what is absent either way — not one entry asks for a feature, a redesign, or more of the thing that is already built.
One is a channel you have to prove works. One is evidence from people who are not you. Both are work, and both are cheaper than the month of building that usually happens instead — because building is the thing you already know how to do, which is precisely why it is the wrong instinct here.
The report's recommended move says the same thing in a harder way: Charge 3–5 named buyers for an async paid pilot before broader build. Charging before building more is uncomfortable, and it is the only step that returns an answer rather than an opinion.
What to actually do this week
- Count arrivals before drawing conclusions. If the number of relevant people who have seen the offer is small, "nobody pays" is not a finding yet. Get the offer in front of people who already have the problem, then look again.
- Go and find the complaint in writing. Not your own experience of the problem — public posts, threads, reviews from people who are not you. If they exist, quote them on the page. If they do not, that is the finding, and it is better to have it now than after another month of features.
- Put a price on the page and ask for it early. A free tier converts a demand question into a usage question, and usage is not the thing you are missing.
- Change the promise before you change the product. The headline is the cheapest variable you own, which makes it the first one to move on the top two rows — not the only one, and the table above says what to check before you decide it was the promise at all.
What this report cannot tell you
It is one report on one idea, and the areas it is confident about are confident relative to the description it was given — not to a running business with users. Your product has something this report did not: real behaviour. Read your own arrival and signup numbers first; they beat any rating, including ours.
And a low score on demand evidence is not proof there is no demand. It means the evidence was not found, which is the same distinction our own pipeline draws when it drops a candidate. The cure for both readings is identical, and it starts with looking for the complaint rather than for a feature.
The short version
- "Nobody pays" splits at least three ways, and each way has a rival explanation that fits the same symptom. Only one of the three would be evidence about the idea; the other two are about distribution and the promise, and the table says what separates them.
- On a published report for this exact shape of product, the weakest area is Pain + Demand Evidence and the two low-confidence areas are the two demand-side ones.
- Both of the actions that report says would raise its verdict are evidence. Neither is a feature.
- Count arrivals, find the complaint in writing, put a price on the page, and change the headline before you change the product.