Updated October 2026. Validating a startup idea means testing, as cheaply as possible, whether real people have the problem you want to solve, whether they want your solution and whether they will pay for it, before you spend money on a prototype or a minimum viable product (MVP). It is the step that comes before building anything.
It matters because poor product-market fit is one of the most common reasons startups fail. In a March 2026 analysis of 431 VC-backed companies that shut down since 2023, CB Insights found identifiable failure reasons for 385. "Ran out of capital" topped the list at 70%, but CB Insights calls it "almost always the final cause of death, not the root problem"; poor product-market fit was cited for 43%. Validation is how you find out about product-market fit while it still costs a few weeks, not a year of building.
The six steps at a glance
- List your riskiest assumptions, and decide in advance what would prove each one wrong.
- Talk to potential users about the problem, not your idea.
- Check willingness to pay with money or a commitment, not compliments.
- Run a landing-page or pre-order test to see whether strangers act.
- Test a clickable prototype of the core flow.
- Decide: build, pivot or drop.
Step 1: List your riskiest assumptions
Every idea rests on a handful of beliefs. Write them down as statements that could turn out to be false, then rank them: the one that would kill the idea if it were wrong, and that you have the least evidence for, gets tested first. Most ideas depend on the same five kinds of assumption:
- Problem: a specific group of people has this problem, often, and it hurts.
- Customer: you can name who they are and where to find them.
- Solution: your approach solves it better than what they do today.
- Willingness to pay: they, or their employer, will pay enough for the business to work.
- Channel: you can reach them at a cost the price supports.
For each assumption, set the test and the pass mark before you run it. Deciding afterwards what counts as success is how founders talk themselves into ideas. A simple way to keep it honest:
| Assumption | Test | Pass mark, set in advance | Result |
|---|---|---|---|
| Clinic billing staff spend hours every week on denied claims | Problem interviews with billing staff | Most interviewees describe a recent denial without prompting, and what it cost them | To fill in |
| They would trust an AI draft if a person approves it | Show a mocked-up draft in the interview | They point to what they would change, not whether they would use it at all | To fill in |
| A clinic would pay a monthly fee for it | Name a price at the end of the interview; ask for a paid pilot | At least a few clinics agree to a paid pilot or a letter of intent | To fill in |
| We can reach clinic managers online | A landing page with a small ad budget | Sign-up rate you decided on before the ads ran | To fill in |
The example is based on an idea from our collection, claimjumper, a hackathon project for denial triage in medical billing. The rows are illustrations, not research results. Choose pass marks that fit your market.
Step 2: Talk to potential users about the problem
Interviews are the cheapest test you can run, and they usually tell you more than a survey. The rule that makes them work: ask about what people have actually done, not what they think they would do. "Would you use an app that does X?" gets a polite yes. "Tell me about the last time this happened" gets the truth. Rob Fitzpatrick's book "The Mom Test" is a widely used guide to asking questions that even your mother couldn't answer with a polite lie.
Questions that work:
- Tell me about the last time you had to deal with this. What happened?
- What did you do about it? What did that cost you, in time or money?
- What have you tried to fix it? Why did you stop using it?
- Who else is involved when this happens? Who decides what to buy?
- If you could change one thing about how you handle this, what would it be?
How many is enough? There is no fixed number. LaunchRock (May 2026) says 30-50 interviews "typically reveal the patterns you need" and that below 10-15 you "can't separate signal from noise". A practical stopping point is when you can predict what the next person will say. Interview the same type of person, so the patterns aren't muddied by five different markets.
Find them where they already talk about the problem: industry forums and subreddits, LinkedIn groups, trade associations, your own network's introductions. Say upfront that you are researching, not selling; people are more open when there is no pitch coming.
Step 3: Check willingness to pay
Saying "I'd use that" costs nothing. LaunchRock's guide puts the difference simply: "I would use that" doesn't predict purchase, while "I've spent $5K on workarounds" does. Look for evidence that money already moves:
- Current spend. What do they pay today for tools, staff hours, agencies or workarounds? That is the budget you are competing for.
- A named price. End the interview with a real number and watch the reaction. Hesitation tells you more than enthusiasm.
- A commitment. Ask for something that costs them: a deposit, a paid pilot, a letter of intent, an introduction to their manager, an hour to test a prototype. A yes to any of these is worth more than a dozen compliments.
For business software, find out who signs off on purchases. An enthusiastic user who can't buy is a useful interview, not a customer.
Step 4: Run a landing-page or pre-order test
Interviews tell you what people say in a conversation. A landing page tells you what strangers do when nobody is watching. Describe the product as if it exists, state the price, and ask for an action: join a waitlist, book a demo or pre-order.
- Write the headline in the words your interviewees used for the problem.
- Send traffic you didn't hand-pick: a small ad budget, a post in the community where your users are, a cold email to a list that matches your customer.
- Measure the action that costs the visitor something. An email address is weaker than a booked call, and a booked call is weaker than a deposit.
- Write down the conversion rate that would count as a pass before the traffic starts.
- Try a fake door inside an existing product or newsletter: a button for the feature that leads to "coming soon, want early access?"
A landing-page builder and a form tool are enough; see MVP software tools for options. If you would rather have it built, MVP development cost lists what agencies charge for landing-page and concierge tests.
Step 5: Test a clickable prototype
Once you know the problem is real, test whether your solution makes sense to the people who have it. A clickable prototype shows the screens of the core flow and lets people tap through them, with no real code behind it. Why you need a clickable prototype first explains what it gives a startup; gathering user feedback on a prototype covers how to run the sessions.
Before you draw screens, decide what belongs in the first version. User story mapping lays out the user's journey and slices off the smallest set of steps that delivers the value; that slice is what the prototype, and later the MVP, should cover. AI prototyping tools can turn a description into clickable screens in minutes.
You don't need many sessions per round. Jakob Nielsen's often-cited analysis found that tests with five users uncover about 85% of usability problems, and recommends several small rounds over one large one (Nielsen Norman Group, 2000). Give people a task ("book a viewing for Saturday"), stay quiet, and watch where they hesitate.
Step 6: Decide whether to build, pivot or drop
Go back to your assumption log and compare each result with the pass mark you set. Then make one of three calls:
| What you found | Decision |
|---|---|
| The problem is real, people describe it unprompted, and some have committed money or time | Build: scope an MVP around the one flow they cared about |
| The problem is real, but they don't want your solution, or a different group feels it more | Pivot: change the solution or the customer, and test again |
| Interviewees shrug, nobody commits, and the landing page misses the pass mark | Drop it, or park it, and move to the next idea |
Dropping an idea after a few weeks of tests is a good outcome. LaunchRock frames the trade as a few weeks of interviews against six to twelve months of building the wrong thing. When the evidence says build, keep the first version as narrow as the evidence: our guides to MVP development cost and building an MVP as a non-technical founder cover what comes next.
Extra checks for an AI idea
AI ideas have two more assumptions to test before you build:
- The model can do the job. Take twenty or thirty real inputs from your interviewees (with permission) and try them in a chat interface. If a strong model can't get them right with a good prompt, the product won't either. Our guide to how to build an AI app explains how that test grows into an eval set.
- The unit economics work. Estimate the tokens per user and compare the cost with the price your interviewees accepted. AI app development cost has a worked example at current model prices.
Where to find ideas worth validating
If you don't have an idea yet, start from what other builders are already working on. Our collection of 7,800+ AI business ideas indexes the projects submitted to the OpenAI hackathon, each with its tagline and team. The industry hubs narrow it down:
- AI ideas for healthcare
- AI ideas for finance
- AI ideas for e-commerce
- AI ideas for marketing and sales
- AI ideas for HR and recruiting
- AI ideas for real estate and construction
- AI ideas for restaurants and food
- AI ideas for travel and hospitality
For ideas by function rather than industry, see AI agent examples and AI SaaS ideas. Keep in mind what a hackathon project proves: that something can be built. It says nothing about whether anyone will pay for it. That is exactly the question the six steps above answer.
Frequently asked questions
How long should validating a startup idea take?
Weeks, not months. LaunchRock (May 2026) describes a round of customer interviews as costing "a few weeks", set against six to twelve months spent building the wrong thing. A landing-page test and a prototype round can run alongside the interviews.
How many customer interviews are enough?
Enough that you can predict the next answer. LaunchRock's 2026 guide says 30-50 interviews typically reveal the patterns and that fewer than 10-15 can't separate signal from noise. Talk to one type of customer at a time so the patterns are clear.
Can I validate a startup idea without building anything?
Yes. Problem interviews, a request for a deposit or a paid pilot, a landing page with a price, a fake-door button and a concierge test (doing the service by hand for a few customers) all work before any software exists. A clickable prototype is the first thing you build, and it has no real code behind it.
When is an idea validated enough to build an MVP?
When the assumptions that could kill it have passed the marks you set in advance, and some people have committed money, time or a signature, not just praise. At that point the MVP tests the next question: whether they keep using it.
Validated an idea and ready to build the first version? See how we approach custom MVP development, or compare our three plans (bi-weekly sprints, weekly sprints and 24/7 development) on the pricing page.