What it means to validate a startup idea
To validate a startup idea is to gather evidence, before you commit months of work and money, that a specific group of people has a specific problem, cares enough to change their behaviour, and will give up something real to solve it. The last clause matters most. Compliments, likes and 'I would totally use that' are not validation; they cost the person saying them nothing, so they carry almost no information.
Validation is also not the same as product-market fit. Fit is what you measure after a product exists and customers keep using it. Validation comes first and answers a smaller question: is this worth building at all? If you want the later stage, our guides to reading product-market-fit signals and to the MVP cover it. This article is about the weeks before the first line of code.
Step 1–2: Write the idea as assumptions and find the riskiest one
An idea is a stack of assumptions wearing a trench coat. 'A scheduling tool for physiotherapy clinics' quietly assumes that clinics lose money on no-shows, that the owner rather than a receptionist decides on tools, that existing software is inadequate, and that they will pay a monthly fee. Write every such assumption down as a plain sentence. Most founders find between eight and fifteen, and the exercise alone exposes how little is known.
Then rank them by two questions: how badly does the idea collapse if this is false, and how little evidence do you currently have? The assumption that scores high on both is your riskiest one, and it is the only one worth testing first. Founders tend to test the assumption that is easiest to test, which is usually technical feasibility, while the fatal one is almost always about the customer: whether the problem is painful and frequent enough that someone pays to remove it.
Step 3: Run customer discovery interviews that are not sales pitches
Customer discovery interviews are short conversations, usually 20 to 30 minutes, with people who plausibly have the problem. The rule that separates useful ones from wasted ones is simple: talk about their life, not your idea. Ask about the last time the problem occurred, what they did about it, what it cost them, and what they already pay for. Past behaviour is reliable evidence. Hypothetical answers about what they would do are mostly fiction.
Aim for around fifteen interviews inside one narrow segment before you draw conclusions, because patterns only show up when the same story repeats. Keep a simple log with the quote, the workaround they currently use and the money or time at stake. If ten people describe the same painful workaround without being prompted, you have a problem worth solving. If you have to explain the problem to them, that is a finding too, and usually a negative one.
Step 4: Ask for commitment, not opinions
Once the problem looks real, test the solution with signals that cost the customer something. A smoke-test landing page measures whether strangers will leave an email address, or better, pay a deposit, for a product that does not yet exist. A concierge version has you deliver the result by hand to three or four customers, which proves the outcome is valued before you automate it. A letter of intent, a paid pilot or a pre-order is stronger still.
The pattern behind all of them is escalation. Each step asks for a little more time, reputation or money, and each step filters out people who were only being polite. The table below sorts common tests by how much they cost the customer, which is a decent proxy for how much you should trust the result.
| Opinion signals | Commitment signals | |
|---|---|---|
| Typical examples | Survey answers, likes, 'sounds great' | Pre-orders, paid pilots, signed letters of intent |
| Cost to the customer | Almost nothing | Money, reputation or a real chunk of time |
| Reliability | Low: politeness inflates it | High: people rarely pay for what they do not need |
| What it proves | The topic is interesting | The problem is painful enough to act on |
| Speed to collect | Hours | Days to weeks |
Fake validation: the signals that mislead founders
Some evidence feels like validation and is not. Friends and family are biased toward you. A survey sent to your own followers measures your audience, not the market. Waitlist counts inflate when sign-up is free, and traffic from a lucky post says nothing about intent. Even a real customer can mislead if they are an outlier: the one enterprise buyer who loves the idea may represent a market of one.
A quieter trap is confirmation bias in how you ask. Leading questions such as 'wouldn't it be useful if...' collect agreement by design. A cleaner habit is to define in advance what a failed test would look like, then record the results as they come in rather than as you remember them. If every conversation seems to confirm the idea, either you are in a rare situation or you are asking questions that cannot disagree with you.
Step 5: Set pass and kill thresholds before you test
The most useful validation habit is deciding the success bar before you see the data. 'If at least 8 of 15 interviewees describe the problem unprompted and 3 agree to a paid pilot, we build; if fewer than 3 do, we change segment or drop it.' A number written down in advance prevents the common failure of moving the goalposts until a weak result looks acceptable. The figures below are illustrative, not universal benchmarks, and the right ones depend on your market.
Thresholds also protect your runway. Each test costs weeks, and weeks are cash for a team that is paid or a founder who is not. Treat every experiment as a small purchase of information with a budget and a deadline. When the deadline arrives, the decision is one of three: proceed, change the assumption and retest, or stop. Stopping early on a bad idea is a good result, not a failed one, because you keep the months you would otherwise have spent.
15
interviews in one narrow segment
8 of 15
describe the problem unprompted
3
agree to a paid pilot or deposit
4–6 weeks
deadline before the go / change / stop call
Step 6–7: Decide, then build the smallest thing that tests the next risk
A first validation round should usually take four to eight weeks, not six months. If the evidence is strong, you do not build the full product; you build the smallest thing that tests the next riskiest assumption, which is often whether customers will keep using the solution after the novelty fades. That is exactly what a minimum viable product is for, and why validation and the MVP are two stages of one process rather than rivals.
If the evidence is mixed, resist both extremes. Do not push ahead because you have already invested, and do not abandon the idea because one segment said no. Look for the segment that said yes most clearly and narrow the idea toward them. Many companies that later found fit began with a pivot toward the customers who showed up with the most pain, and the pivot guide covers when that change is wise.
Practising the same discipline in a simulation
Founder Runway puts you in the position of making these calls under time pressure. A run lasts 20 turns, each decision has four options, and you choose a B2B, B2C or B2G model and a starting stage of Pre-Seed, Seed, Pre-Series A or Series A. Choices that chase growth before the customer signal exists burn cash without moving product-market fit, which is the same trap as building before validating, only compressed into minutes.
A run ends as a failure, a promising company, EBITDA positive or high-value exit potential, so you can compare how a patient, evidence-first approach plays out against a rush to build. The game is a rehearsal, not a substitute for talking to real customers. Use it to practise ranking risks and deciding with incomplete information, then take the habit into your actual interviews.
Frequently asked questions
How do you validate a startup idea?
List your assumptions, find the riskiest one, and test it with customer discovery interviews followed by tests that ask for commitment, such as pre-orders, paid pilots or deposits. Set pass and kill thresholds before you collect data. A first round usually takes four to eight weeks.
How do I validate a business idea without building a product?
Talk to potential customers about past behaviour, run a smoke-test landing page, or deliver the result manually to a few customers as a concierge service. These tests cost days, not months. They show whether people will pay before you invest in software or inventory.
How many customer interviews do I need to validate an idea?
Around fifteen interviews inside one narrow segment is a practical starting point, because patterns only appear when the same story repeats. If the first eight or ten sound identical, you may learn enough sooner. If answers are scattered, your segment is probably too broad.
What is the difference between idea validation and product-market fit?
Validation asks whether an idea is worth building and is answered with interviews and commitment tests before a product exists. Product-market fit is measured after launch through retention, usage and revenue. Validation reduces the risk of building the wrong thing, while fit confirms the right thing is working.
Is a survey enough to validate a startup idea?
No. Surveys capture opinions, which cost the respondent nothing and tend to overstate demand. Use them to find people to interview, then rely on behaviour and commitment such as time spent, deposits paid or letters of intent signed.
What are signs a startup idea is not worth pursuing?
Warning signs include people who must be convinced the problem exists, no existing workaround or spend, polite interest that never turns into a pilot, and a customer who is an outlier rather than a pattern. If your pre-set threshold is missed after a fair test, change the segment or stop.
How long should idea validation take?
A focused first round typically takes four to eight weeks. Set a deadline in advance and make a go, change or stop decision when it arrives. Open-ended validation tends to become a way of avoiding the decision.
Test this decision in the game.
Apply the same assumption across one run and see which metric weakened three turns later.