How to Validate a Startup Idea in Two Weeks

Spend an hour before the sprint writing down what has to be true for the idea to work. Most ideas rest on four assumptions:

Then set the pass and fail line for each test before you run it. The conversion optimization firm The Good calls these acceptance criteria and gives examples in the form "if we generate a given amount in pre-orders within a given number of weeks, it will be worth proceeding" (The Good). Writing the number down first matters because once data arrives, every founder is tempted to move the line. A product guide from Userpilot makes the same point about fake door tests: set the threshold before launch, commit to it, and do not move it once results come in (Userpilot).

A simple pre-registration note might look like this:

Idea: scheduling assistant for independent physical therapists
Segment: clinic owners with 1 to 3 therapists
Pass if, by day 14:
  - 8 of 12 interviewees describe double-booking or no-shows as a top-3 problem unprompted
  - landing page converts targeted visitors at or above our written threshold
  - 3 owners pay a refundable deposit or sign a paid pilot
Stop if: fewer than 4 of 12 mention the problem without prompting

Spend the first week talking to people, not building. The goal is to learn how the problem shows up in their lives today, what they already do about it, and what that costs them.

Find people fast. Start with your network, then go one step out. Todd Jackson's research for Lenny's Newsletter, based on interviews with the founders of companies including Vanta, Cocoon and Good Dog, describes founders who exhausted their own networks and then posted on Craigslist to reach a broader audience, or asked friends whether they knew anyone who had just been through the experience being studied (Lenny's Newsletter). Communities, industry forums and short cold emails also work. Aim for two or three conversations a day.

Ask about the past, not the future. Harvard Innovation Labs recommends the approach from the book The Mom Test: instead of "Would you buy this product?", ask "What challenges do you face in this area?" (Harvard Innovation Labs). Questions about what someone did last time are hard to fake. Questions about what they would do someday invite polite fiction.

Good interview questions for this stage:

• Tell me about the last time this happened. What did you do?
• What have you tried to fix it? What did that cost?
• How much time or money does this take each week or month?
• Who else is involved when it goes wrong?
• If you could wave a wand, what would be different?

Listen for unprompted pain. Cocoon's co-founder told Jackson that with the problem they chose, the pain was obvious, while in other areas "we were searching for the problem" (Lenny's Newsletter). One of Jackson's summary tips is that if it feels like pulling teeth to find the pain point, you have not found a problem worth solving.

How many is enough? For two weeks, ten to fifteen interviews in one segment is a realistic target. A guide from the research platform CleverX, read on 2026-10-17, suggests 15 to 20 participants per distinct customer segment before identifying patterns (CleverX). Treat interviews as direction, not proof. A startups.com guide warns that talking to 10 to 20 customers is anecdotal evidence and does not make you an expert on the market (Startups.com).

At the end of day five, write a one-page summary: the problems you heard, how often each came up unprompted, what people use today, and the exact words they used. Those words become your landing page copy.

A smoke test puts a description of the product in front of real prospects and measures whether they act. The Good describes it as testing the basic idea before implementing it fully, so you can abandon a failing project before spending much on it (The Good).

Build one page around one promise. KickoffLabs suggests a headline test: the first screen should say who the product is for, what problem it solves and what happens if you sign up (KickoffLabs). Use the language from your interviews. Collect an email and at most one qualifying question, such as "What are you using today?"

Send the right traffic. A page with no visitors proves nothing, and a page full of friends proves less. Post where your interviewees said they spend time, email people who asked to hear more, and consider a small paid campaign aimed at the segment. The Good quotes Eric Ries, author of The Lean Startup, arguing that anyone building an online service should test it with an ad campaign first (The Good).

Measure source by source. Track visits, signups and conversion rate for each traffic source, and how many signups reply to a follow-up email. KickoffLabs notes that 40 qualified leads from a community post can be worth more than 400 paid signups who never answer an email (KickoffLabs).

Use the thank-you page. After someone signs up, ask them to book a short call or answer one question. That converts a passive signup into a second round of interviews, and the share of people who say yes is itself a signal.

Fake doors need honesty. A fake door is a button or offer for something that does not exist yet. ConfigCat, a feature flag vendor, recommends telling people right after they click that the feature is under consideration and offering to notify them (ConfigCat). Chameleon, which makes product adoption software, warns that a poorly handled fake door can break trust in a brand (Chameleon). For a company with no customers yet, the risk is lower, but say plainly what exists and what does not.

Signups measure curiosity. A commitment measures intent. The closer the behavior is to paying, the stronger the signal. Rank your options by strength:

CleverX notes that for business products, paid pilots with design partners are especially strong evidence, because customers commit budget and integrate the solution into their operations (CleverX). The Good quotes a product consultant who argues that one of the most accurate ways to test a product page is to collect payments before the product is built, then refund buyers and tell them it is not ready yet (The Good).

Try a concierge offer. Harvard Innovation Labs recommends a concierge MVP, a manual version of the service you deliver yourself, before automating anything (Harvard Innovation Labs). Good Dog's founders did a version of this by finding dogs manually for hopeful owners before building their marketplace (Lenny's Newsletter). If people will pay you to do the job by hand, that is strong evidence they will pay for software that does it.

Show mockups, not code. Cocoon's founders signed their first customers from a design prototype shared with people they had already interviewed (Lenny's Newsletter). Vanta's co-founder advised testing with spreadsheets and mockups because "It's so easy to change a spreadsheet or mockup; it's much harder to change code" (same source).

Make the ask direct: "We are taking five pilot customers at this price. Do you want one of the spots?" A clear no is useful. A vague "maybe later" is a no.

Put every result next to the threshold you wrote before day one. Do not reinterpret. A simple scorecard:

Then make one of three decisions. Jeroen van den Boom, Chief Venture Strategy Officer at Bundl, quoted by The Good, puts it this way: move forward with a validated concept, make adjustments and re-test, or cut your losses and move on (The Good).

• Commit when people describe the pain without prompting and at least a few have paid or signed.
• Change and re-run when the pain is real but the segment, message or offer missed. Pick the one assumption that failed and run another two-week sprint on it.
• Stop when the pain is mild, rare or already solved well enough. Stopping after two weeks is a good outcome, not a failure.

A sign worth watching for in any decision is inbound pull. Jackson describes Vanta's turning point as the moment people the founder had not spoken to started reaching out to ask about the product (Lenny's Newsletter).

• Asking for opinions. "Would you use this?" produces kind lies. Ask about past behavior.
• Interviewing friends. They want you to succeed. Use them for introductions, not evidence.
• Counting the waitlist as traction. KickoffLabs warns that founders end up with "beautiful charts and no customers" by treating email signups as product-market fit (KickoffLabs).
• Confusing activity with progress. Startups.com lists naming the company, designing a logo and printing business cards as milestones that feel like progress but prove nothing (Startups.com).
• Looking for the answer you want. Cocoon's co-founder warned that if you try to make something a problem, you will find the evidence to support it (Lenny's Newsletter).
• Moving the threshold. If you change the pass line after seeing the results, you ran no test at all.

An AI assistant can speed up the clerical parts of this sprint, such as drafting interview guides, tagging interview notes and writing landing page variants. It cannot do the interviews or ask for the money. Those parts stay with the founder.