Concept Validation
/ Playbook.
How to validate a product concept before you build it. For founders and PMs at the zero-to-one stage.
Owner: _____
Stage start: _____
Most failed startups didn't fail at building — they failed at validating. This playbook walks four stages, from cheapest to most expensive. Move forward only when the prior stage hits its success bar. Skip a stage and you trade speed for false confidence.
The Hypothesis Statement
Write this before the smoke test. If you can't fill all four blanks, you don't have a concept — you have a vibe.
Stage 1 — Smoke Test
A landing page + paid traffic. The cheapest signal you can buy. You're testing demand for the promise, not the product.
- Goal
- Measure intent — will the audience click "sign up" for the promised value?
- Method
- One-page landing site with the headline, the 3 promises, a single email-capture CTA. Drive 500–1,000 targeted visitors.
- Time-box
- 7 days end-to-end. 1 day to build, 5 days to run, 1 day to read.
- Budget
- $200–$500 in LinkedIn or Google search ads. Cap it.
- Success
- >5% visitor-to-email conversion in your ICP. >15 emails from named companies you'd actually serve.
- Kill
- <1.5% conversion after 1,000 visits = no demand at this price point or value prop. Reword once. Don't reword three times.
Stage 2 — Concierge MVP
You deliver the value by hand to 3–5 customers. No product yet. The goal is to learn whether the outcome you promised is actually wanted — and what it really takes to deliver.
- Goal
- Confirm the JTBD, learn the actual workflow, watch real users react to real value.
- Method
- Deliver the outcome manually. Spreadsheet, Loom video, weekly Zoom call, hand-typed Slack summary — whatever it takes. You are the algorithm.
- Time-box
- 4 weeks. Three customers minimum, five preferred.
- Success
- 3 of 5 say "I'd pay for this" in cash, not future-tense. At least one unprompted referral. Repeatable workflow you could automate.
- Kill
- You can't replicate the value across customers without bespoke work. Or: nobody opens the deliverable twice.
- Graduate
- You've done it manually 3× in a row, and the steps look mechanical. Now build the smallest automation.
Stage 3 — Wizard of Oz
Front-end looks real, back-end is a person clicking buttons. Customers don't know. This stage tests whether the productized version works without paying full build costs.
- Goal
- Test the user-facing experience while the backend is still humans + duct tape.
- Method
- Real signup flow, real product UI, fake intelligence. Examples below.
- Time-box
- 4–6 weeks. 25–100 active users.
- Success
- Activation rate >30%. WAU/signup >15%. Users stop emailing you support questions you predicted.
- Kill
- Manual fulfillment can't keep up with active users (good problem — build automation). Or: users sign up and never return.
Three product-type examples:
- AI feature ("the model writes for you"): first 100 outputs are written by you in 4 hours, served via the same UI. Watch which outputs get used vs. ignored.
- Marketplace ("auto-matched suppliers"): manually email-introduce buyers and suppliers behind a "match" button. The button is real; the matching is you.
- Workflow integration ("syncs Slack to Jira"): read Slack, type Jira tickets yourself for the first 20 customers. Find out which messages they want synced before you build the rules.
Stage 4 — Pre-Sales
Money on the table before code is shipped. The strongest signal in concept validation.
- Goal
- Get cash or signed commitment for a thing that doesn't exist yet.
- Method
- Three formats, ranked by strength of signal:
1. Pre-orders — wire transfer or Stripe charge today, deliver in 60 days.
2. Paid pilots — reduced rate for 8–12 weeks, with a real PO or invoice.
3. LOIs — signed letter of intent. Weakest signal. Useful for enterprise where procurement takes months. - Time-box
- 30 days to land 3–5 paying or signed accounts.
- Success
- 3+ paid pilots OR 5+ LOIs from named, in-ICP buyers. Free design partners do not count.
- Kill
- Lots of "this is interesting, send me when it's ready." That's a polite no. Believe people only when they pay.
The Decision — Pivot, Persevere, or Kill
After each stage, three outcomes. Be honest about the band you're in.
| Stage Result | Confidence | Move | Verdict |
|---|---|---|---|
| Hits success bar with margin (≥1.5×) | High | Advance to next stage. Document what worked. | Persevere |
| Lands in band (0.7–1.5×) | Medium | Run the stage again with one variable changed (audience, price, message). | Pivot |
| Misses by >30% across two attempts | Low | Kill the concept. Or pivot to a different audience that surfaced unexpectedly. | Kill |
Common Validation Mistakes
- Vanity metrics. "100 newsletter signups" is not validation. Email is the cheapest commitment a human can make. Demand cash, calendar, or referral.
- Leading questions. "Would you use a tool that…?" almost always returns yes. Past-tense behavior is the only signal: "the last time you did this, what did you do?"
- Building too soon. If you find yourself opening Figma in week one, you skipped Stage 1. Engineers especially fall into this. The cheaper the test, the more you'll learn before money is spent.
- Ignoring negative signal. One customer saying "I love it but I won't pay" is more informative than ten saying "looks great." Treat dissent as data.
- Sample-size hubris. Two friends and an old colleague is not validation. ICP-fit, paying, repeat-using strangers is. Don't run validation on people who'd never let you fail.
The frame: validation isn't permission to build. It's evidence that builds conviction — conviction you'll need when you hit the inevitable Trough of Sorrow at month 8. Skipping it doesn't save time. It moves the failure later, when failing costs more.