How to find a startup idea worth building

A startup idea worth building usually starts as a problem you noticed, not a product you dreamed up. Paul Graham opens his essay How to Get Startup Ideas with the rule, writing that "The way to get startup ideas is not to try to think of startup ideas. It's to look for problems, preferably problems you have yourself."

The failure data points the same way. In its report on why startups fail, CB Insights reviewed 431 VC-backed companies that shut down since 2023 and, among the 385 where failure reasons could be identified, found poor product-market fit cited in 43% of them. Those are funded companies, many cited several reasons at once, and running out of capital tops the list, so the figure is not a precise share of bad ideas. It still shows that a large share of serious attempts break on the market side, the part an idea is supposed to settle.

This page covers the step before validation: where to look, how to tell a real problem from a plausible one, and the filter an idea should pass before you spend a week testing it. Validation itself is in how to validate a startup idea.

Why do brainstormed startup ideas usually fail?

Brainstormed startup ideas fail because they are built to sound good, not to match a problem someone has. Graham's essay explains why trying to think up ideas is risky: "it doesn't merely yield few good ideas; it yields bad ideas that sound plausible enough to fool you into working on them." At YC, he writes, these are called made-up or sitcom ideas, the kind a TV writer invents for a character's startup.

His own example is a social network for pet owners. Lots of people would say it sounds nice. Almost nobody would rearrange their week to use it. A brainstorm rewards that kind of idea, because the test in the room is whether the idea sounds reasonable, and the test in the market is whether anyone wants it badly.

Graham admits making the mistake himself. He describes starting a company in 1995 to put art galleries online and spending 6 months on it because, in his words in the same essay, "I didn't pay attention to users."

Where should you look to notice a startup idea?

The best place to notice a startup idea is inside work you already know well, where your experience lets you see what is missing. Graham borrows a phrase from Paul Buchheit about people at the leading edge of a fast-changing field and condenses the method into one line in the essay: "Live in the future, then build what's missing."

YC partner Jared Friedman turns that into a concrete exercise in his Startup School talk, published in the YC Startup Library. His best recipe, the one he says about half of YC's most successful companies can be traced back to, starts with what your team is especially good at. Go through every company you have worked at, internships included, and ask three questions: what did you learn there that other people do not know, what seemed broken or clumsy, and what did the company build in-house that other companies might need.

Three more places tend to produce real problems:

• Your own unmet needs. Graham's trick is to ask what you wish someone else would build so that you could use it, and what you would pay for right now. A clear answer to that is evidence of demand.
• Other people's complaints. Talk to people about what is tedious or annoying in their work, without pitching anything, and listen for a problem they have stopped noticing.
• Recent change. Friedman lists looking for things that changed recently, such as a new technology, platform or regulation, and asking what is now possible because of it.

Friedman is blunt about how available these are. In the same talk he says: "Once you learn how to notice good startup ideas, you'll see them all over the place."

How do you tell a real problem from a plausible one?

A real problem has someone who wants the fix now and badly, and Graham's sharpest test asks for exactly that person. In the essay he writes: "Who wants this so much that they'll use it even when it's a crappy version one made by a two-person startup they've never heard of?" If you cannot answer, he says, the idea is probably bad.

Graham describes good demand as a well rather than a shallow hole: a small group that wants something a lot, instead of a large group that is mildly interested. Facebook started with Harvard students, a few thousand people who wanted it a lot. The pet owners' social network is the shallow version, broad and lukewarm.

The practical version of the test is to name ten specific people who have the problem, not a segment. If you can name them, you can talk to them this week. If all you can name is a demographic, you have a theory about a market, not a problem.

What filter should a startup idea pass before you validate it?

A startup idea should pass four questions before you spend time validating it, and an idea that fails one of them is cheaper to drop now than after a prototype. The questions combine Graham's demand test with the idea quality score Friedman describes in his YC talk, which rates how big the idea is, founder and market fit, how sure you are that you are solving a big problem, and whether you have a new, important insight.

Question	What a passing answer sounds like	Warning sign
Who wants this right now?	Named people who have the problem today and have tried to fix it	A broad audience that would find it nice to have
How do they cope today?	A spreadsheet, a manual workaround, a paid tool they dislike	Nothing, because the problem barely bothers them
Why you?	You worked in the field, have the problem, or can reach the buyers	You read about the space last week
What changed recently?	A new technology, tool or rule makes a better fix possible now	Nothing changed, so you need a reason others failed

The "why you" row is founder-market fit, and the founder-market fit guide has a scoring exercise for it. The first row matters most. An idea with a strong answer to it and weak answers elsewhere can still be worth testing. An idea with a weak first answer is a solution looking for someone to use it.

YC has a name for that last case. Friedman calls an idea like an app that sends a plumber at the push of a button a solution in search of a problem, or SISP, and says founders who start there usually never find the problem.

Which reasons to drop an idea should you ignore?

Some reasons for dropping an idea are filters that hide good ideas, and Friedman names four in his YC talk: the idea seems hard to get started, the space is boring, the idea seems too ambitious, or competitors already exist. His example of the first is Stripe, which thousands of developers knew was needed and avoided because it meant dealing with banks and card networks.

Competitors are the filter most worth switching off. Friedman's view is that a market with no competitors often means nobody wants the product, and that the better position is a market with competitors where you have noticed something they all missed. Graham makes a similar point about boring problems in his essay: "Make something unsexy that people will pay you for."

Is there a case for brainstorming startup ideas anyway?

The case for brainstorming is real when you need an idea now: you left a job to build something, or your first idea failed and you cannot wait for inspiration. Graham concedes the point and spends the rest of his essay on tricks for finding ideas on demand, and Friedman's talk lists several recipes for generating ideas deliberately, ranked from most to least likely to work.

The catch is discipline. Graham's warning in the essay is that a deliberate search changes the error rate: "You'll see a lot more ideas, most of them bad, so you need to be able to filter them."

Idea lists work the same way. Pages like side project ideas for product managers, no-code startup ideas and weekend project ideas with AI tools are raw material, useful once you run each entry through the four questions above, and misleading if you pick the one that sounds most fun.

AI chat tools sit in the same place. They are fast at listing who might have a problem and how they cope today, and that makes them a good sparring partner for an idea you already noticed. They cannot tell you that anyone has the problem, because a model producing a plausible answer is the brainstorm failure in a new form.

What should you do once an idea passes the filter?

Once an idea passes the filter, test it with people who have the problem before you build. Talk to the ten people you named, using the method in how to run customer interviews, and look for a costly signal, such as money, time or a referral, rather than a polite yes.

Builders Camp's From Idea to Launch with AI bootcamp is a fast, practical path from an idea to a focused MVP and a real launch plan. It runs 1 week with 2 live sessions and 8 self-paced microlessons, and it covers validating the problem, defining the ICP and value proposition, scoping the MVP around assumptions, and choosing distribution moves that get first users even without an audience. One of its lessons asks what a proven way is to assess whether a product idea is worth building before writing any code. It is part of the Discovery Expert Track and the Vibe Coding Expert Track.

See the From Idea to Launch with AI bootcamp

If no idea passes yet, start a problem log today. For the next two weeks, write down every time you or someone you work with complains about a task, works around a tool, or says they would pay for a fix, with the date and who said it. At the end, the entries that show up more than once, from more than one person, are your shortlist.