A friend calls you with an idea. He's animated, sketching diagrams on a napkin, already talking about the domain name. Six months later he's burned through his savings building a dashboard nobody logs into. I've been on both sides of that conversation, and I still wince thinking about the feature I built for four months straight, only to discover that the three people who'd been asking for it were asking for something slightly different — and had already found a workaround they preferred.
So let's talk about how to validate a startup idea before building. Not in the "believe in yourself" way. In the "here are the thresholds and the false signals and the tests that cost you almost nothing" way. Because the uncomfortable truth is that most validation advice online is too vague to be actionable, and vague validation is worse than no validation — it gives you a false green light.
Key Takeaways
- Validation means evidence of behavior, not evidence of politeness.
- Fifteen to twenty properly recruited interviews usually reveal whether a pain is real; two hundred bad ones can lie to you.
- A landing page test needs a clear conversion benchmark before you run it — decide your threshold first, or you'll rationalize whatever number comes back.
- The cheapest honest test is a manual "concierge" delivery: do the job by hand for a handful of users and see if they come back.
- A signed letter of intent with a price and a date beats a hundred "I'd definitely buy that"s.
What idea validation actually means (and what it doesn't)
Idea validation is the process of gathering evidence that a specific group of people will take a costly action to solve a problem you intend to solve, before you commit engineering time to the solution.
Notice the phrase "costly action." That's the whole thing. Costly action means money, a signed commitment, a deposit, a calendar invite for a pilot, switching away from an existing tool. It does not mean enthusiasm, agreement, or a thumbs-up emoji on your prototype video. Those are free for the person giving them, which is exactly why they're worthless to you.
Look, I get why people conflate the two. When you've spent weeks on a concept, a compliment feels like oxygen. But compliments are the single largest source of wasted runway in early companies, and I say that having been the guy who collected them like trading cards.
The three false signals people mistake for validation
- "That's a great idea" — this is a social reflex. It costs the speaker nothing and ends the conversation politely.
- "I would definitely use that" — hypothetical future behavior. People are notoriously bad at predicting their own purchases, myself included.
- An unsigned letter of intent, or a "maybe we could pilot this next quarter" — non-binding enthusiasm that evaporates the moment a budget approval cycle enters the room.
The test I apply now: if this person's enthusiasm were money, would I be able to pay a single invoice with it? If the answer is no, it's a signal about their manners, not your market.
A startup idea validation framework you can actually run
Here's the sequence I use. It's not novel — it borrows heavily from the customer development tradition — but the details matter more than the shape.
Step one: write the claim you're trying to disprove
Before any test, write one sentence: "At least [number] of [specific audience] will [costly action] within [timeframe] because [problem] costs them [specific consequence]."
Mine once read: "At least 5 of the 30 independent bike shops I contact will prepay $40/month for automated inventory alerts within three weeks." That sentence forced me to name the audience, the price, the deadline, and the consequence. Vague ideas die in vague sentences. Mine nearly did.
Step two: run interviews that don't lead the witness
The core failure of founder interviews is asking about the future instead of the past. "Would you use X?" is a fiction prompt. "Walk me through the last time this problem bit you — what did you do that day?" is a fact prompt.
Ask about the last occurrence. Ask what they tried. Ask what it cost them in hours or dollars. Ask what they did instead when the workaround failed. Then shut up and let the silence pull more out of them. The most useful sentence in any interview tends to arrive right after you stop talking.
Step three: test with behavior, not opinion
This is where a small budget buys a lot of truth. Three approaches, ranked by how much I trust them:
- Pre-sale or deposit. Highest signal. Real money leaving a real account. Even $20 refundable is worth more than an hour of praise.
- Concierge delivery. You do the job manually for five to ten users, no product required. If nobody comes back for a second round, your problem isn't a software problem, it's a demand problem.
- Landing page plus targeted ads. Weakest of the three, but useful for scale and for testing your positioning language. Set the pass threshold before you launch it.
How many interviews, and what conversion rate, before you commit?
Numbers are where most guides go soft, so here are mine. These are heuristics from my own projects, not laws of physics — treat them as starting thresholds, then tighten them for your category.
| Test | Reasonable sample | Pass threshold I use | Typical cost |
|---|---|---|---|
| Problem interviews | 15–25 recruited prospects | At least 60% describe the problem unprompted, in their own words, with a recent example | Your time; small thank-you incentives |
| Landing page + ads | 300+ targeted visitors | Email or waitlist signup rate of 8% or better on qualified traffic | Modest ad spend |
| Pre-sale | 30–50 qualified outreach contacts | 5–10% convert to a paid deposit or signed order | Almost nothing but effort |
| Concierge pilot | 5–10 users | At least half request a second round without you chasing them | Your labor |
Two caveats that I learned the hard way. First, the interview threshold is about saturation, not statistics — once the last five conversations stop surprising you, you've probably heard the pattern. Second, the 8% landing page figure only means something if the traffic is properly targeted; broad traffic will flatter you with curiosity and starve you of buyers.
How do you recruit prospects who won't just be nice to you?
Never start with friends, family, or your own network's warmest contacts. They're structurally incapable of giving you a clean signal.
Go where the audience already complains in public: niche forums, trade associations, local meetups, communities around the tool they currently use. Introduce yourself as someone researching the problem, not selling a solution. Say plainly that you're not pitching anything and you want twenty minutes of their experience. Roughly a third of the people you approach this way will say yes, and their answers will be worth ten times what your college roommate told you.
Can AI help you validate a startup idea?
Partly. AI is genuinely good at three jobs in this process, and genuinely dangerous at a fourth.
Use it to synthesize interview notes — dump twenty transcripts in and ask it to cluster the recurring problems and the language people used to describe them. Use it to draft interview scripts and pressure-test whether your questions are leading. Use it to generate the objections you haven't thought of, and to draft your landing page copy for the ad test.
The dangerous part: using AI to simulate your market. Ask a model whether people would pay for your idea and it will produce a confident, well-structured, entirely fictional answer. It has no access to purchase behavior. I've seen founders walk away from a chatbot conversation feeling validated, which is a bit like asking a mirror whether you're a good dancer.
AI accelerates the thinking. It cannot supply the evidence.
The one thing I would do differently
On my second attempt at a side project, I skipped the interviews entirely. I was sure. I built a scheduling tool for freelance tutors, launched it, and got forty signups in the first week — which felt amazing until I checked retention. Two users came back in month two. I'd validated a curiosity, not a need.
The lesson wasn't "validate more." It was "validate the thing you're actually afraid to test." I hadn't wanted to ask tutors whether they'd pay, because I was scared of the answer. That fear was the most useful data in the whole project, and I ignored it for four months.
So when you sit down with your own idea this week, pick the single question whose answer would hurt most to hear. Then go find twenty people who can answer it honestly. If they can't, you haven't validated anything — you've just been keeping yourself company.
