We ran hundreds of startup ideas through 9 decision gates. Here is what keeps killing them

I have spent the last year watching founders make the same nine decisions in the same order, and I kept every answer.

ShipFit forces an idea through nine sequential gates. Worth building. Who pays. What hurts. How to win. What is v1. How to charge. Will they pay. How to launch. What to export. Nothing moves forward until the previous decision is made, which means the whole thing is a record of how founders actually think when they are made to choose.

So last week I stopped theorising and queried it. Hundreds of projects,, everything since go live. Here is what is in there.

Everyone is building the same product.

Start with what people actually submit. Of the ideas that got as far as picking a buyer, 43 percent were aimed at prosumers, individual people spending their own money on a tool that makes them better at something. Small business took 26 percent, straight consumer 26 percent, developers 3 percent, and enterprise 1 percent. One percent. Nobody is starting an enterprise company from a standing start any more, and honestly, good.

The shape of the thing being built is just as narrow. Of the ideas that reached the strategy gate, 51 percent resolved to a SaaS platform and 34 percent to a browser extension of some kind. Mobile was 5 percent. Desktop was under 1. At the pricing gate, 98 percent of products were classified as software. This is a monoculture, and if your idea is a prosumer SaaS tool you are not early, you are the median.

The more interesting number is which problems people choose to solve. The engine surfaces around ten distinct pain points per idea, and the ones founders gravitate to skew weekly rather than daily. Forty one percent of pain points occur weekly, 22 percent daily. Mean intensity was 70 out of 100, and only 38 percent were classed as must-solve rather than nice-to-have.

Three frameworks do ninety percent of the work.

I gave the engine a library of nineteen strategic frameworks to choose from at the "how to win" gate. Six have ever been picked. Three account for nine selections out of ten.

Jobs-to-be-Done took 39 percent of the ideas that reached that gate. Blue Ocean took 34 percent. 7 Powers took 17. Product-Led Growth got 8 percent. Playing to Win and Sales-Led Growth together got barely 2. That is the whole distribution. The other thirteen frameworks never came up once.

Before that, at the first gate, every single idea gets the same three treatments regardless of what it is: TAM/SAM/SOM, Porter's Five Forces, and a market timing analysis. Those are not chosen, they are compulsory, because there is no idea on earth that does not need sizing, competitive structure and a timing check.

What I did not expect is that framework choice tracks buyer type almost cleanly. Prosumer ideas pull Jobs-to-be-Done 45 percent of the time. Small business ideas and consumer ideas both flip to Blue Ocean, 48 percent and 44 percent respectively. And 7 Powers is overwhelmingly a prosumer instrument here, with 71 percent of its uses landing on prosumer ideas.

That makes sense once you see it. When a person buys a tool for themselves, the job they are hiring it for is legible, so Jobs-to-be-Done bites. When you are selling into a business or into a consumer market, the buyer is already drowning in options and the only useful question is where the uncontested space is.

The practical version: you do not need nineteen frameworks. You need to know whether your buyer is a person spending their own money or an organisation spending someone else's, and that determines which one of three frameworks is worth an afternoon.

The founders with the best ideas are the ones who keep going, and you can see it in the first five minutes.

This is the finding I keep chewing on.

The first gate returns a verdict on the idea. Founders who got a strong signal there went on to complete all nine gates 78 percent of the time. Founders told "let's find out" completed 2 percent of the time. Same product, same nine gates, same work required. Thirty nine times the completion rate, and the split happened before anyone had done anything.

Zoom out and the funnel is brutal. Of the projects with any completed work, 32 percent stopped dead after the first question and 15 percent went all the way through. But it does not fall away evenly. It drops off a cliff at gate one, thins through pricing and demand testing, then climbs again at gates eight and nine. Almost nobody quits after "will they pay". There is a point of no return in this process and it sits directly after the money conversation.

I do not think my model is clairvoyant about which ideas are good. I think conviction is the fuel, and a lukewarm answer on day one empties the tank. Founders rarely abandon an idea because the analysis defeated them. They abandon it the moment something hands them permission to stop.

There is a more generous reading, and it is probably also true: people are walking away from ideas the tool was unenthusiastic about, which is the entire point of validating anything. Either way the operational lesson is the same. Whatever you learn about your idea in week one determines whether there is a week twelve, so front-load the harshest question you have. If you cannot get excited about your own answer to it, the rest of the process is not going to rescue you.

Half of all ideas are two ideas.

Now the big one. At the strategy gate the engine diagnoses the main structural flaw in each idea. Fifty one percent came back with the same diagnosis: you are trying to build two products at once.

Wrong buyer fit was a distant second at 18 percent. High risk approach was 15. Weak differentiation, the thing founders and investors talk about constantly, came in at 5 percent. Scope confusion outranked positioning by ten to one.

Here is what the shape looks like in practice. These are composites of patterns I see repeatedly, not anyone's actual submission.

A marketplace connecting freelance editors to publishers, plus the invoicing and project tooling the editors will need. Two products. One is a demand aggregation business that lives or dies on liquidity, the other is a workflow SaaS that lives or dies on daily active use, and they have almost nothing to do with each other.

A habit tracker for people managing a chronic condition, plus a dashboard so their clinic can monitor them. Two products. One is consumer retention, the other is a regulated B2B sale with a twelve month cycle. Different buyer, different pricing, different everything.

An AI writing tool for founders, plus a community where they share what they wrote. Two products. One you charge for, one you cannot charge for, and the second quietly eats every hour of the first.

Every one of these sounds like a strategy when you pitch it, because the second half explains why the first half will win. That is exactly what makes it so hard to see from inside. But watch what happens downstream: you cannot name one buyer, so you cannot pick one problem, so you cannot scope a v1, so you cannot price it, and every subsequent decision feels impossible for reasons you cannot articulate. Founders think they are stuck because their idea is hard. They are stuck because it is two ideas and each decision is being asked to serve both.

The test is cheap. Describe your product in one sentence with one "and" removed. If the sentence dies, you have two products and you need to shelve one this week, not after you build.

Observation one: pricing advice converges no matter what you feed it.

Whatever the product, the recommendation was freemium plus tiered 73 percent of the time, and 91 percent were positioned as mid market. Not premium, not budget. Mid market, freemium, tiered, over and over.

Some of that is genuinely correct, because 98 percent of these were software and that is what software does now. But if the answer is the same for almost every product, pricing is not something you should expect any tool or any advisor to hand you. It is the one decision you have to go and get from a real buyer, because the default is so strong that nobody's reasoning about it is doing much work.

Observation two: nobody is recommending paid acquisition any more.

Across every launch plan the engine produced, community took 45 percent of recommended channels, organic 35, partnerships 7, and paid advertising 12. Community and organic between them took 81 percent of all launch advice.

The median success target attached to those plans is 400 signups and 45 paying customers, which is an implied 11 percent signup-to-paid conversion. That is a demanding number if you got those signups from a community you spent six months in, and a fantasy if you bought them.

Whether this is the models absorbing bootstrapper consensus wholesale or a genuine reflection of what works pre-revenue, I cannot separate from this data. What I can say is that if you are budgeting for a paid launch, you are doing something that essentially nothing in this corpus recommends.

Observation three: Claude Code and Cursor are dead level.

At the final gate founders export config files for whichever AI coding tool they are actually going to build with. Claude Code was chosen by 92 percent of them. Cursor by 91. Tool-agnostic exports by 38 percent, and Replit by about one.

It is multi-select, and that is the finding. Most people took both. Nobody is picking a side in this fight, they are hedging, and they are hedging in almost exactly equal measure. I have not seen anyone else put a number on that race, so there it is.

Method, so you can hold me to it. Hundreds of projects from dozens of founders, everything since go live, one row per project and latest run only so power users cannot skew a bucket. Every gate past the first has a smaller sample because most projects stop early, and I have kept each figure to a single denominator rather than mixing them, so "39 percent of ideas get Jobs-to-be-Done" means 39 percent of the ideas that reached that gate, not 39 percent of everything submitted. Free text model output was normalised by hand, dozens of raw spellings collapsed down to six canonical frameworks. Buyer type is taken from the "who pays" gate only. This is a self-selected sample of people who chose to run an AI validation tool, not a random sample of founders. These are ideas, not businesses. I have no data on what got built or what sold, and no figure above should be read as an outcome.

Happy to hand the query pack, sample sizes and date range to anyone who wants to poke holes in it.

The frameworks behind the gates: https://shipfit.ai/frameworks/ How the nine gates work: https://shipfit.ai/how-it-works