Guide · Starting out
How to validate a startup idea before you build anything (and before you pay someone like me)
How to validate a startup idea before building: interviews, landing pages, pre-sales, concierge and Wizard of Oz tests, with AI or no money, and when to stop.
Short answer
Validate a startup idea by getting people to give up something real before you build: a call about their problem, an email next to a price, a deposit or a signed pilot. Start with interviews about past behaviour, then a landing page or pre-sale, then a test you run by hand. Most of it costs almost nothing.
This is the post where I tell you not to hire me yet. I build MVPs for a living, and the most expensive mistake I see founders make is paying for a build before anyone has shown they want the thing. Code is a slow, costly way to find out an idea doesn't have buyers. There are faster ways, and most of them cost almost nothing.
How do you validate a startup idea?
You validate a startup idea by getting the people you want as customers to give up something real before you build: time talking about the problem, an email address next to a price, a deposit, a signed pilot, or repeated use of a version you run by hand. Opinions and compliments don't count. Commitments do.
It's worth the effort. CB Insights looked at 431 venture-backed companies that shut down since 2023 and found poor product-market fit among the reasons for 43% of them. Running out of money showed up in 70%, but CB Insights notes that's usually the final cause of death rather than the root problem (CB Insights, 2026). Most of those companies built something. Not enough people wanted it.
What are the ways to test a startup idea?
Six tests cover almost every idea. Run them roughly in this order, because each one costs a little more and tells you a little more than the one before.
| Test | What it proves | Rough cost | Best for |
|---|---|---|---|
| Problem interviews | The problem is real, frequent and painful, and how people solve it today | Time only | Every idea, first |
| Landing page with a price | Strangers care enough to leave an email at your price | A domain plus a free-tier page builder | Consumer and self-serve products |
| Pre-sale or paid pilot | They'll pay before it exists | Time and a payment link | B2B, courses, anything with a clear buyer |
| Concierge test | The outcome is worth paying for when you deliver it by hand, openly | Your time | Services, marketplaces, B2B workflows |
| Wizard of Oz test | People will use the product, with you secretly doing the work behind it | Your time plus simple tools | Anything where the "automation" is the expensive part |
| Demo video or clickable prototype | People want it once they see it working | A few days | Products that are hard to explain in words |
The cost column is my rough estimate for a founder doing it themselves with free or low-cost tools. None of these need a developer.
How do you run customer interviews that don't lie to you?
Ask about their life and what they did last time, not about your idea and what they'd do in future. This is the core of Rob Fitzpatrick's The Mom Test: if you pitch, people are polite, and polite answers are useless. If you ask about specific past behaviour, you get facts.
Questions that work:
- "Tell me about the last time this happened. What did you do?"
- "What have you tried to fix it? What did that cost?"
- "Who else is involved when this goes wrong?"
- "If you could wave a wand, what would be different?" (Then ask why it matters.)
Questions that mislead you: "Would you use an app that…?", "How much would you pay for…?" and "Do you think this is a good idea?" Everyone says yes to those. End each conversation by asking for something small: an introduction to someone else with the problem, a second call, or permission to send them the first version. People who won't give you that won't give you money either.
How do you validate with a landing page?
Put the real price on it, send traffic you can trace, and treat sign-ups as the start of a conversation rather than the result. A page without a price measures curiosity. A page with a price measures intent.
Three rules I'd follow:
- One promise, one action. Say who it's for, what it does for them and what it costs. One button.
- Know where every visitor came from. Tag the links you post in each community or email, so you know which audience responds. Fifty sign-ups from one forum is a lead. Fifty sign-ups from nowhere is noise.
- Follow up within a day. Email each sign-up and ask for a 15-minute call. The share who say yes tells you more than the sign-up count.
A demo video can do the same job for a product that's hard to explain. The famous case is Dropbox: according to TechCrunch, the team expected around 15,000 people on its beta waitlist, and 75,000 signed up within 24 hours of a demo video posted during the private beta (TechCrunch). The video showed the product working; the waitlist showed the demand.
What is a concierge test, and what is a Wizard of Oz test?
Both are ways to deliver the product by hand before you build it. In a concierge test the customer knows a person is doing the work. In a Wizard of Oz test, the customer sees what looks like a product, and a person does the work behind the curtain.
Zappos is the classic Wizard of Oz. Nick Swinmurn's early model, in his own words, was "going into stores, taking pictures, and selling the shoes online". He offered a store: "if people buy them, I'll buy them from you at full price" (Fortune). No warehouse, no inventory system, just proof that people would buy shoes online.
The modern version is usually a form, a spreadsheet, a payment link and you. For a booking marketplace, you take requests through a form and match them by phone. For an AI tool, you do the "AI" part yourself and email back the result. You learn exactly what customers need from the output, which is the spec you'll eventually build to.
When a concierge test starts to hurt, because you can't keep up by hand, that's one of the best signals there is that it's time to build.
How can a quiz or assessment capture demand before you build?
A short free assessment lets people tell you about their problem in a structured way, and gives you a reason to follow up. It's a lead-capture tool that doubles as research: every answer is a data point about who's interested and what they're struggling with.
For Breathe n' Being, a yoga and meditation studio, I built a free wellness-assessment quiz as the front of the funnel, alongside the public site and booking. The pattern travels well; the physio clinic booking story uses it to route new patients, with stricter privacy rules because the answers are health information. If your idea is "help people with X", a five-question assessment that tells them where they stand on X can show you who cares, how they describe the problem, and which answers come up most, before you've built the thing that solves it. As a validation test, it needs nothing beyond a form tool and a results page; you don't need it built into a product to learn from it.
How do you validate a business idea without spending money?
Spend time instead. Interviews are free. A concierge or Wizard of Oz test costs your evenings. A pre-sale costs a payment link. The only spend most founders need is a domain name, and even that can wait until the landing-page step.
What you're trading is speed for certainty. A week of interviews can save you months of building the wrong thing. That trade is worth it almost every time.
How do you validate a business idea with AI?
Use AI to do the work around real customers faster, not to replace them. It's good at:
- Drafting interview questions, then checking them for leading phrasing.
- Summarising interview notes and pulling out repeated problems.
- Sorting complaints from forums, reviews and support threads into themes.
- Writing several landing-page versions to test against each other.
- Building a clickable prototype or demo in a day.
What it can't do is be your customer. Simulated users and AI personas tell you what sounds plausible. They can't pay you, and they can't surprise you, and surprise is most of what validation is for.
What counts as a strong signal and what's a weak one?
A strong signal costs the person something. A weak one costs them nothing. Here's how I'd rank them.
| Weak signal | Strong signal |
|---|---|
| "Great idea, I'd use that" | "Here's how I solved it last month, and what it cost me" |
| Likes and comments on a post | Replies asking when they can start |
| Waitlist sign-ups with no price shown | Sign-ups with the price shown, who then take a call |
| Friends and family enthusiasm | Strangers introducing you to colleagues |
| "Let me know when it's ready" | A deposit, a pre-order or a signed paid pilot |
| One person who loves it | Several people with the same problem, described the same way |
When have you validated enough to build?
When a few of the right people have given you something real, and you can describe exactly what the first version must do for them. In practice that looks like:
- You can name the type of customer who has the problem worst, and you've talked to several of them.
- At least some have paid, pre-ordered, signed a pilot, or used a manual version more than once.
- You know how they solve the problem today, and what that costs them.
- You can write the one job the MVP does on a single line.
- Running it by hand has started to hurt.
You won't get certainty. Validation lowers the risk; it doesn't remove it. Past this point, building is the cheapest way to learn the next thing, which is whether people keep using it. My step-by-step guide picks up from here: how to build an MVP step by step.
When should you skip straight to building?
Rarely, but there are cases. If you are the customer and have the problem every day, you can often build a narrow version for yourself and validate with people like you. The customer-onboarding SaaS story is close to that case: a founder building the tool she had wanted in her old job, for companies with the same problem. If the product is a small internal tool where the cost of being wrong is a few weeks, testing may take longer than building. And if the real risk is technical, such as whether an AI model can do the job accurately enough, a quick technical prototype is the validation.
Even then, show the first version to someone who isn't you before you spend real money on it.
What to do this week
- Write your riskiest assumption as a sentence that could be proved false.
- List twenty people who have the problem, and message ten of them for a 20-minute conversation about how they handle it today.
- Put up a one-page site with the real price and one button.
- Offer three people the outcome by hand, for money.
When that's done and you've got real commitments, the build is the next step. If you want to know what it would cost first, here's how to estimate what your app will cost, and here's how I work when you're ready.
Questions founders ask
How do you validate a business idea without spending money?
Talk to people who have the problem about how they handled it last time, not about your idea. Then ask for a small commitment: a second call, an introduction, a pre-order. Run the service by hand for a few customers using email, a spreadsheet and a payment link. All of that costs time, not money.
How do you validate a business idea with AI?
Use AI to speed up the work around real customers: draft interview questions, summarise notes, cluster complaints from forums and reviews, and write landing-page variants to test. Don't use AI-generated personas or simulated users as the evidence. They tell you what's plausible, not what real people will pay for.
How many customer interviews do you need to validate an idea?
There's no magic number. Keep going until you stop hearing new answers about how people solve the problem today, and until you can name who has it worst. For a narrow B2B idea that often happens within a few dozen conversations. More important than the count is that they're with the right people.
Is a waitlist enough to validate a startup idea?
On its own, rarely. An email address is a weak commitment, especially with no price on the page. A waitlist becomes useful when the page shows the real price, when you follow up and some people agree to a call or a pre-order, and when you know where the sign-ups came from.
When is an idea validated enough to build an MVP?
When a few of the right people have given up something real, such as money, a signed pilot or repeated use of a manual version, and when you can describe exactly what the first version must do for them. At that point, building is the cheapest way to learn the next thing.