How to Get and Evaluate Startup Ideas: YC's 10-Question Test

I started working through Y Combinator's podcast back catalogue recently, and the episode that has stayed with me is How to Get and Evaluate Startup Ideas with Jared Friedman. Its central claim is deliberately humble: nobody can reliably predict which startup will succeed. What you can do is materially improve your odds by starting with the right kind of problem and matching it to the right founding team. Friedman says the framework draws partly from analysing YC's top 100 companies by valuation, which is about as close to an evidence base as this field gets.

Some of it confirmed instincts I already had and had never managed to articulate cleanly. A good deal of it did not, and those parts are the reason I am writing this up rather than just recommending the episode. I have also turned the ten-question test into a free one-page scorecard, if you would rather run the exercise on a real idea than read about it.

The Direction of Travel: Problem First, Technology Second

If you take one thing from the episode, take the arrow direction.

Wrong: "AI is amazing, so what could I build with it?"

Right: "This group of people has a painful problem. Can AI now solve it substantially better than what they use today?"

YC calls the first one a Solution In Search of a Problem, and the trap is that it never feels like a mistake from the inside. Exciting technology makes it trivially easy to invent a plausible use case. You can generate twenty before lunch, and each one will sound reasonable when you describe it. None of that is evidence that anybody is in enough pain to change their behaviour, let alone pay.

This is the part that confirmed something I have argued at length elsewhere. The gap between a working model and a working product is where most AI projects quietly die, and the failure usually starts earlier than people think: not in the engineering, but in the direction the idea was generated from. A system built to justify a technology inherits no constraints from a real user. A system built to relieve a specific pain inherits all of them, which is precisely what makes it buildable.

The Four Mistakes Founders Keep Making

1. Solving something that is not actually painful. People will agree your idea sounds useful. Agreement is cheap. The question is whether they will change a habit or move budget, and most polite enthusiasm converts to neither.

2. Walking into a tarpit idea. Tarpit ideas attract founders repeatedly because the problem looks obvious and the solution looks straightforward, yet dozens or hundreds of startups have already failed at them, because there is a structural difficulty underneath that is invisible from the outside. Friedman's example is apps that make organising plans with friends easier. Everyone has felt that pain. Almost nobody has built a business from it. If an obvious problem has been obvious for two decades, the interesting research question is not "why has nobody solved this?" but "what killed everyone who tried?"

3. Running with the first idea without evaluating it. Enthusiasm is not validation. It is the feeling of not having looked yet.

4. Waiting for the perfect idea. The opposite failure, and just as fatal. There is no perfect starting idea. Successful startups commonly morph significantly after launch, which means your first idea is a hypothesis and a vehicle, not a destination.

The Ten-Question Test

This is the most immediately usable part of the episode: a checklist you can run against any idea in an hour.

1. Do you have founder-market fit?

Friedman treats this as arguably the most important question, and the reframe is what makes it powerful. Do not ask "is this a good startup idea?" Ask "is this a particularly good startup idea for us?"

Your experience, domain knowledge, relationships, technical ability or unique insight should give you an advantage that a generic smart team would not have. YC's example is PlanGrid: one founder understood construction deeply, another had exactly the software skills the product needed.

2. Is the market big enough?

You generally want one of two things: a market that is already large, or a small one growing extremely fast. Coinbase is the example. Cryptocurrency trading was tiny in 2012, but the founders could see what the market would become if Bitcoin worked at all.

So do not evaluate total addressable market today. Evaluate TAM at the point your company reaches scale.

3. How acute is the problem?

The distinction is between a customer thinking "that's quite convenient" and a customer thinking "I need this." Only the second one reliably becomes revenue.

Brex is the example: early-stage startups could not obtain normal corporate credit cards at all. That is categorically different from making an existing process marginally nicer.

The strongest signal that a problem is acute is what people currently do instead. Look for spreadsheets, manual processes, WhatsApp and email threads holding a workflow together, several products cobbled into a chain, expensive consultants, or an entire job role that exists only to work around the problem. People do not build elaborate workarounds for mild inconvenience.

4. Is there competition?

Counterintuitively, competition is usually a good sign. It proves customers exist and are willing to pay, which is the expensive thing to find out. Zero competition is not automatically exciting: sometimes nobody has built your idea because nobody wants it.

The question therefore shifts from "is this space taken?" to "what do we understand that the incumbents do not?" Cloud storage was crowded when Dropbox launched. Dropbox had a sharper insight about seamless operating-system-level syncing, and that was enough.

5. Do you, or people you know, actually want it?

A blunt sanity check. If you would not use it, nobody around you wants it, and you have never met someone desperate for it, then ask honestly where your belief in this problem is coming from.

Founders solving their own problems also get a feedback loop that is very hard to buy: they are users, so they notice when the product is wrong before a customer has to tell them.

6. Why is now the right time?

One of the most powerful questions in the talk. What changed recently that makes this possible now? Candidates include AI capability, smartphones, new APIs, cheaper hardware, regulation, demographic shifts, consumer behaviour, remote work, new infrastructure, or falling compute costs.

YC's example is Checkr: gig platforms like DoorDash and Uber suddenly needed fast, high-volume background checks that traditional processes were never designed to handle. The demand did not exist in that shape a few years earlier.

The corollary is the useful bit: if your idea could have been built identically ten years ago, you owe yourself an explanation of why nobody did it successfully.

7. Is there a proxy proving the model?

Suppose nobody does exactly what you propose. Perhaps somebody has done it successfully in another country, another industry, or for another customer segment. That is real evidence, and it is cheap to find.

YC points to Rappi against DoorDash, and Nuvocargo against Flexport: proven models adapted to Latin America.

8. Can you imagine working on this for years?

A startup can consume a decade. But YC adds a qualification I found genuinely freeing: you do not need to arrive pre-obsessed with the industry. Plenty of excellent businesses operate in unglamorous ones. Traction tends to manufacture passion, rather than the other way round.

9. Is it actually scalable?

Be suspicious of companies that present as software but require substantial human labour per additional customer. If 2× customers means roughly 2× employees, you have built a services business wearing a software costume.

That is not a bad business. Services companies can be excellent, profitable and durable. It simply behaves nothing like a venture-scale startup, and confusing the two leads you to raise the wrong money against the wrong expectations.

10. Are you in a good idea space?

Subtly different from evaluating the idea itself, and the question I now think is the most under-rated of the ten. If idea A fails, can you pivot to B, C or D while keeping your customer knowledge, technology, distribution, relationships and industry expertise?

Fivetran is the example: several pivots inside a fertile data space before landing on the product that worked. The space carried the company while the idea changed.

Three Things That Look Bad and Are Actually Good

This section is the best part of the episode, and the part I had most thoroughly wrong.

1. It looks difficult to start. Paul Graham named this schlep blindness: everyone can see the opportunity, but nobody wants the tedious work of starting. Stripe is the canonical case. Developers universally hated payment integrations, and building the alternative meant banks, financial infrastructure and regulation. That difficulty was not a cost of the business. It was the moat, because it kept better-funded people away.

Hard does not mean bad. Hard often means fewer competitors.

2. It is boring. Payroll, compliance, accounting, infrastructure, logistics, procurement. None of that sounds like "AI social network for creators", and that is an advantage, not a handicap. Gusto built a very large company in payroll. YC's argument is that six months in, every startup looks the same from the inside: build, sell, fix, recruit, talk to customers, repeat. Whether the premise sounded glamorous at the start has almost no bearing on that daily loop. Whether customers desperately want the thing has all of it.

3. There are already lots of competitors. Which, again, means there is definitely a market. The opportunity becomes finding the thing incumbents are fundamentally wrong about.

Where Good Ideas Actually Come From

The most surprising statistic in the episode: Friedman's analysis suggested roughly 70% of YC's top 100 companies found their ideas organically rather than through deliberate brainstorming.

Which is to say the founder ran into something and thought "why is this still so terrible?", rather than sitting down on a Saturday and writing "give me 50 startup ideas".

The actionable version of that statistic is not "wait for inspiration". It is: put yourself in environments where problems become visible to you. Organic discovery is not luck; it is a consequence of proximity. If you are not near real work, you will not see real problems, and you will end up brainstorming, which is the 30% path.

Seven Ways to Find One

YC's methods, in roughly descending order of quality:

1. Start with what you or your team are unusually good at. Expertise creates founder-market fit for free.
2. Mine problems you have personally hit. Go back through previous jobs and ventures and ask what was needlessly difficult.
3. Ask what you wish existed. Powerful, but check carefully that it is not a tarpit.
4. Look for something that recently changed. Unusually fertile right now: AI capability, regulation, new APIs, new hardware, behavioural shifts.
5. Take a recently successful company and build a variant. X for Europe, X for healthcare, X for SMEs, X for construction.
6. Pick a promising industry and talk to people inside it. Do not ask them for startup ideas; they do not have any. Ask what is painful, manual, expensive, slow or ridiculous about their week.
7. Find huge industries that appear broken. Logistics, construction, healthcare administration, government processes, insurance, procurement. The mess is the opportunity.

The Whole Framework in One Sentence

If I had to compress forty minutes into a single line:

Find a painful, specific problem in a large or rapidly growing market that your team is unusually equipped to understand, where something has changed recently that lets you solve it substantially better than the current alternatives, and then start, rather than trying to prove the idea is perfect first.

And the mental model I have actually adopted, which is the one that reframes everything else:

Do not ask "is this a billion-dollar idea?" Ask "is this a strong enough starting point, in a fertile enough space, for us to start learning?"

The first question is unanswerable and invites months of theorising. The second is answerable this week.

What Actually Changed My Mind

Three things, and they were not the ones I expected.

Competition as a positive signal. I had absorbed the standard instinct that a crowded market is a warning. Framed properly it is the opposite: somebody else has already paid to prove customers exist, and the remaining question is narrower and much more tractable: what do they get wrong?

Difficulty and dullness as moats. This one genuinely reorganised how I look at opportunities. I have spent a lot of time on the technically interesting end of problems, and the honest reading of schlep blindness is that the technically interesting part is usually the part everyone else also wants to do. The unglamorous adjacent work is where the defensibility hides.

Idea space over idea. Evaluating a single idea is a low-information exercise, because the idea will change. Evaluating whether the surrounding space would still be valuable after three pivots is a far better predictor, and it is a question I had simply never asked in that form.

What it confirmed, meanwhile, was the direction-of-travel rule. Watching a lot of teams start from "we have AI, now what?" was the thing that pushed me to write From Model to Product in the first place, and it is the same failure mode I described when arguing that producing artefacts got cheap while verifying them stayed expensive. Cheap production makes solutions-in-search-of-problems easier to build than ever, which means the discipline of starting from a real problem is now worth more, not less.

My own route into building things was much closer to the organic 70% than to any brainstorm. EchonLabs started in a boarding room next to a university, out of a problem two students kept running into, not out of a list. Having a name for that pattern is more useful than it sounds, because it tells you what to do next time: get closer to the work, not further into a document. That is also the argument I ended up building From Model to Product around: the constraint has moved, and the habits have not caught up yet.

Frequently Asked Questions

What is a solution in search of a problem? An idea generated by starting from an exciting technology and looking for something to apply it to, rather than starting from a painful problem and asking what now solves it. It is hard to detect from the inside because plausible use cases are easy to invent for any capable technology.

What is a tarpit idea? An idea that repeatedly attracts founders because the problem seems obvious and solvable, but which has a structural difficulty underneath that has defeated many previous attempts. Apps for organising plans with friends are the standard example. The test is simple: if the problem is decades old and obvious, find out what killed the last twenty attempts before assuming you are the first to notice it.

Is competition a bad sign for a startup idea? Usually not. Competition demonstrates that customers exist and will pay, which is the expensive thing to validate. No competition can mean nobody wants the product. The useful question is what you understand that the incumbents do not.

What is founder-market fit? The degree to which your specific team's experience, knowledge, relationships or technical ability gives you an advantage on this specific problem. The reframe is to stop asking whether something is a good idea and start asking whether it is a good idea for you.

How do most successful startups find their ideas? According to Friedman's analysis of YC's top 100 companies by valuation, around 70% found their idea organically, by encountering a problem in the course of their work, rather than through deliberate brainstorming sessions.

Listen to It Yourself

Summaries lose things, and this one is worth forty minutes of your own attention: How to Get and Evaluate Startup Ideas on Spotify. YC also publishes a large amount of this material free in the YC Library.

The framework will not tell you whether your idea works. Nothing will. What it does is stop you spending two years finding out something you could have established in two weeks, and given how much of a founder's life is measured in exactly that currency, that is a considerable thing for a free podcast to hand you.

Before you close this tab: pick the idea you are actually thinking about and give it a score. Download the free Startup Idea Scorecard. One A4 sheet, ten weighted criteria, no email required. It takes about fifteen minutes and is considerably cheaper than finding out in year two.