How to validate a startup idea before you spend money building it
Opinions are free. Evidence costs the person giving it something, and that is how you spot it.
Why validate a startup idea before building it?
Because most startups do not die of technical problems; they die building something the market does not want enough. Validation costs weeks. Finding out after the build costs months and much of your runway.
43%
The headline reason in the CB Insights analysis was running out of capital, at 70%, but the authors stress that cash is usually the final cause, not the root one. The sample is VC-backed and mostly American; the lesson applies just as much if you are bootstrapping from savings.
There are no facts inside your buildings, so get the heck outside.
Blank, who formalised customer development, has made the same point for years (Stanford GSB): until you test a business plan against real customers, it is an act of faith dressed up as a spreadsheet.
How do you validate a startup idea, step by step?
Turn the idea into testable assumptions, then gather evidence from cheapest to most expensive.
Five steps to validate an idea
Break the idea into assumptions
Every idea is a bundle of assumptions: who has the problem, how often, what they use today, what they would pay, how you reach them. Write one per line and mark the riskiest. Start there.
Talk to people who have the problem
Conversations about the past, with people in your segment, not friends. Keep going until the answers start repeating.
Ask for something that costs them
A second call with a colleague, access to a process you can watch, a letter of intent, a pre-order.
Run a small market test
A hand-delivered service or a page with real prices measures behaviour rather than opinion; Buffer, for one, put a plans-and-pricing page before its sign-up form to see who would click on a paid plan. See types of MVP for formats and landing pages to validate a startup idea for the page itself.
Decide against a bar you set earlier
Carry on, change direction or stop: write down in advance which result leads to which decision, so the data cannot bend to your hopes.
How do you run customer interviews that do not lie to you?
Ask about the past, never the future. Asked whether they would use your app, most people say yes to be polite; asked what they did the last time the problem came up, they give you a fact you can build on.
In his Startup School talk, Y Combinator partner Eric Migicovsky warns against asking whether someone would use or pay for a feature you have not built, and suggests a short script instead. Our paraphrase:
- What is the hardest part of [the job you want to improve]?
- Walk me through the last time it happened.
- Why was that hard?
- What, if anything, have you done to fix it?
- What do you dislike about the fixes you have tried?
What not to ask
Skip hypotheticals (would you pay £20 a month?) and requests for approval (do you like the idea?). Do not pitch your solution in the first few minutes: the moment you do, people start being kind to you instead of honest about their problem. Rob Fitzpatrick wrote a whole book on avoiding that trap, The Mom Test: good questions are so concrete that even your mum, who loves you, could not answer them with a kind lie.
What counts as real evidence?
Anything that costs the giver something. A compliment costs nothing; a deposit does. This is the ladder we use to weigh signals during validation.
| Signal | What it costs the giver | Strength |
|---|---|---|
| Great idea, likes, comments | Nothing | None |
| Waitlist sign-up | An email address | Weak |
| A second call, or an intro to a colleague | Time and a little reputation | Medium |
| Access to real data or processes for a trial | Trust and effort | Good |
| Letter of intent signed by a decision-maker | A formal commitment | Good |
| Pre-order, deposit or payment | Money | Strong |
| Repeat use without reminders | A changed habit | Strongest |
How do you validate an idea with no audience or distribution?
Go where your customer already asks for help and request conversations one at a time instead of chasing traffic. At this stage ten good conversations beat a thousand visits.
- Communities where your ideal customer asks questions, such as subreddits, Slack and Discord groups or LinkedIn threads: read before you post, then ask the person with the right question for twenty minutes.
- Trade bodies, professional associations and industry events: full of people with the problem and little time, so keep the ask short and specific.
- Former colleagues and clients from your previous job, if that is where the problem comes from.
- Short personal emails with one clear request: twenty minutes to understand how they handle something today, no sales pitch.
This is the spirit of Paul Graham's Do Things that Don't Scale: early users are recruited by hand, and the manual work teaches you things no campaign will.
Can you validate an idea while you still have a job?
Yes, and it is often the sensible way to do it. Interviews fit into lunch breaks and evenings, small tests into weekends, and a pre-agreed bar tells you whether it is time to leave.
Two cautions. Check your employment contract for non-compete, exclusivity or intellectual-property clauses, and take advice if anything is unclear. And do not use your employer's time, data or customer list for the side project.
When should you stop validating and build the MVP?
When the evidence has cleared the bar you set and the remaining question, usually repeat usage, can only be answered by a product. From there, what is an MVP walks through the build; if you are unsure whether you need a prototype or a proof of concept first, start with MVP vs prototype vs proof of concept.
Validation does not stop at launch; it changes tools. Interviews give way to usage data, and the question becomes who comes back and who drifts away. When customers return on their own and complain if the product disappears, you are approaching product-market fit.
If you are tempted to speed things up by building an AI prototype straight away, that can work, as long as it stays a tool for gathering evidence: see how to build an MVP with AI.
Frequently asked questions
How many people should I interview to validate an idea?
There is no magic number. Keep going until answers start repeating within the same segment; if they never do, your segment is probably too broad and needs narrowing.
How do I validate a startup idea without spending money?
Interviews, a hand-delivered concierge service and a simple pricing page cost little beyond your time. What you cannot skip is asking for commitments, because free signals such as likes and sign-ups prove very little.
Is an online survey enough to validate an idea?
No. Surveys collect opinions and intentions, which are weak evidence. They can help later to size a problem you have already observed in interviews, not to decide whether to build.
Should I ask people to sign an NDA before I describe my idea?
For validation interviews, usually not: early conversations should be about their problem, not your solution. If you have genuinely sensitive technology, ask a lawyer how best to protect it.
What is the difference between validation and product-market fit?
Validation shows that a problem exists and that people will commit to solving it. Product-market fit comes later, when a product is used steadily by a market large enough to build a company on.
Sources
- CB Insights, The top reasons startups fail (2026) (19 October 2026)
- Stanford Graduate School of Business, Lean Messaging: How Simple Messages Really Stick (Steve Blank, 2022) (19 October 2026)
- Y Combinator, Startup School Week 1 Recap: Eric Migicovsky on talking to users (2019) (19 October 2026)
- Rob Fitzpatrick, The Mom Test (sito ufficiale del libro) (19 October 2026)
- Paul Graham, Do Things that Don't Scale (2013) (19 October 2026)
- Buffer, Idea to Paying Customers in 7 Weeks: How We Did It (19 October 2026)
Figures and definitions checked on the dates shown.
