How do you validate a startup idea?

What does it mean to validate a startup idea?

Validating a startup idea means finding proof that a real group of people has a painful problem, has it often, and would pay to make it go away. You are testing the problem first, not your product.

Most ideas fail because nobody needed the thing. Paul Graham calls the most common mistake "solving problems no one has." Validation finds that out in weeks instead of years.

Think of your idea as a guess with three parts. Who has the problem. How bad it is. What they do about it today. Your job is to check each part with real people and real behavior, not with your own excitement.

A simple way to write the guess: "I believe [this group] struggles with [this problem] because [this reason]." Theanna's founder teaches founders to start every round of interviews with that sentence.

• Real: people can describe a specific time it happened to them.
• Painful: it wastes their time, money, or sleep, and they can say how.
• Frequent: it happens every week or every day, not once a year.
• Paid for: they already spend time, money, or effort working around it.
• Specific: you can name who has it, down to job, situation, or moment.

Who should I talk to first?

Talk to strangers who have the problem right now, not your friends or family. People who love you will be nice about your idea, and nice is useless here. Nomiki Petrolla, Theanna's founder, gives the founders she coaches a simple rule: no friends or family interviews for the first 30 days.

Start narrow. Paul Graham's test is to ask "who wants this right now?" A small group that needs it badly beats a big group that thinks it sounds neat. If you cannot name that group, your idea is not ready to test. Rule people out on purpose. The smaller and clearer your group, the easier your first customers are to find.

Kevin Hale of Y Combinator suggests looking for problems that are widespread, growing, urgent, expensive, mandatory, or frequent. You do not need all six. You need at least one to be strong.

• Your own network: ask for intros to people with the problem, not opinions on the idea.
• Where they already gather: online groups, forums, Slack and Discord communities, subreddits.
• In person: events, meetups, and places where your users work. One YC company got meetings by walking into fire stations.
• People who complain in public: reviews of tools in your space and posts asking for help.
• Every person you talk to: end by asking who else you should talk to.

How do I run a customer interview that tells the truth?

Ask about their life and what they did in the past, never about your idea or what they might do in the future. Those are the core rules of Rob Fitzpatrick's book The Mom Test, and they work because people are bad at predicting their own behavior but good at telling you what happened last Tuesday.

1. 01

   Write down your riskiest guess

   Before any call, fill in: "I believe [this group] struggles with [this problem] because [this reason]." Pick the part that would kill the idea if it were wrong. That is what this round of interviews is for.

2. 02

   Book short calls, and say why

   Ask for a short call to learn how they handle the problem today. Do not lead with your idea, or people start being polite instead of honest. Nomiki Petrolla opens with something like: "I'm not selling anything. I'm doing research, and I'll be asking about your experience."

3. 03

   Start with the last time it happened

   Ask: "Can you walk me through the last time this happened?" Get the details: when, what they were doing, what they tried, how long it took. Specific stories are hard to fake.

4. 04

   Dig into what they already do about it

   Ask what they have tried, what tools they use, and what they dislike about those tools. If they have never looked for a fix, the pain is probably not big enough for them to buy yours.

5. 05

   Talk less, listen more

   Let them do most of the talking. When they say something vague like "I would," ask for a real example. Take notes in their words.

6. 06

   Do not pitch until the end, if at all

   As The Mom Test puts it, once you start talking about your idea, they stop talking about their problems. Save any pitch for the last few minutes, and only if the pain is clear.

7. 07

   Ask for a next step

   End with a small ask: another call, an intro to a coworker or their boss, a look at an early sketch, or a deposit. Book the next meeting on the call, not by email later. A clear no is still useful.

8. 08

   Score it the same day

   Right after the call, fill in the scorecard below. Memory makes every call sound better than it was.

What questions should I ask in a customer interview?

Ask questions that make people tell you stories about the past. Avoid questions that ask for opinions, predictions, or prices for something that does not exist yet.

Eric Migicovsky's Y Combinator talk on talking to users gives five questions that work for almost any idea. They are in the left column below. The right column shows the questions founders ask instead, and why they mislead you.

The one to remember: never ask how much they will pay. Ask how much they have paid.

• What is the hardest part about [the task]?

  Not this:

  Do you think this is a good idea?

  Why:

  Opinions about your idea are free to give and mean nothing.

• Tell me about the last time you ran into this.

  Not this:

  Would you use a product that did X?

  Why:

  Answers about the future are too hopeful. Past stories are real.

• Why was that hard?

  Not this:

  Is this a big problem for you?

  Why:

  Yes or no questions get polite yeses.

• What, if anything, have you done to solve it?

  Not this:

  What features would you want?

  Why:

  People are good at problems, not at designing your product.

• What have you paid to fix it so far?

  Not this:

  How much would you pay for this?

  Why:

  People guess high for things that are not real yet.

• Who else should I talk to?

  Not this:

  Can I add you to my list?

  Why:

  Intros show they care. A list signup shows very little.

How many customer interviews do I need?

Keep going until you stop hearing new things. Nomiki Petrolla tells founders to expect 20 to 30 conversations if you sell to people (B2C) and 5 to 10 if you sell to businesses (B2B). A simple rule: stop a round when three interviews in a row tell you nothing new.

If ten people all give you totally different answers, the problem is usually not your questions. It is that your group is too broad. Narrow it to one type of person and start again.

Do not count calls with friends, people without the problem, or calls where you mostly pitched.

What signals mean my idea is good, and which ones are false?

A good signal asks the other person to give up something real: time, reputation, or money. A false signal is anything they can give you for free, like a compliment or a promise about the future.

Compliments feel great and tell you almost nothing. Watch for fluff words too: "usually," "always," "would," "might." When you hear them, ask for a specific time it happened.

Use this scorecard after every interview. Give each line 0, 1, or 2 points. Look at the totals across all your calls, not at your favorite one.

• Signals that count: a detailed past story, a homemade workaround like a messy spreadsheet, money already spent, an intro to a coworker or boss, a deposit or pre-order.
• False positives: "great idea," "I would totally use that," likes and follows, people who want to be told when it launches but will not book a second call.
• Mixed signals: lots of pain but no attempt to fix it. Often the problem is real but not urgent enough to pay for.
• Do not ask people to design the fix. They know the pain and what they tried. Coming up with the answer is your job.

• Did they tell a specific recent story?

  0 points:

  Only general opinions

  1 point:

  A vague memory

  2 points:

  A clear story from the past few weeks

• How often does it happen?

  0 points:

  Rarely or never

  1 point:

  Every month or so

  2 points:

  Every week or every day

• How much does it hurt?

  0 points:

  Mildly annoying

  1 point:

  Wastes real time

  2 points:

  Loses money, customers, or sleep

• Have they tried to fix it?

  0 points:

  Never looked

  1 point:

  Searched but gave up

  2 points:

  Built a workaround or pay for a tool

• Do they spend on it now?

  0 points:

  Nothing

  1 point:

  Time or effort

  2 points:

  Money, or a hire, or a paid tool

• Did they give you a next step?

  0 points:

  No

  1 point:

  Another call or a sketch review

  2 points:

  An intro, a pilot, a letter of intent, or a deposit

When do I stop interviewing and start testing demand?

Move on once most of your interviews score high on pain, frequency, and current spend, and at least a few people gave you a real next step. At that point you know the problem is real. The next question is whether people will act on your offer.

That is a different test. Instead of asking about their past, you put something in front of strangers and count what they do: sign up, click buy, or pay a deposit. Our guide on how to validate demand covers waitlists, fake door tests, and pre-sales in detail.

Keep the first test small. Theanna's first version was a plain list in a Notion doc. When users kept asking for more, the team shipped a simple fake AI agent. People used it so much they broke it, and that told them to build the real one.

If your scores stay low after a full round, do not keep pushing. Change one thing, like the type of person or the specific problem, and run a new round. Dropping a weak idea early is a win, not a failure.