How to Validate a Business Idea
The five steps
Write the idea as a set of assumptions
An idea is not one claim, it is a stack of them: this problem exists, this segment feels it acutely, they will change behaviour to fix it, they will pay this much, and we can reach them at a cost the price supports. Write each one as a separate sentence. Most ideas fail on one specific link in that chain, and you cannot test a chain.
Example: Instead of "a scheduling tool for clinics", write: (1) independent clinics lose bookings to phone tag, (2) practice managers own this problem, (3) they will switch systems for it, (4) they will pay $150/month, (5) we can reach them through practice-manager networks.
Rank by risk, not by ease
The assumption to test first is the one that would kill the business fastest if it were false — not the one that is most convenient to check. Founders naturally test the assumptions they are confident about, because those tests feel good. Sort by consequence instead.
Example: In the clinic example, whether practice managers can switch systems at all is riskier than whether the price is $150 or $200. Test switching first.
Set the kill criteria before you look
Decide in advance what result means yes and what result means no. Write it down with a number and a date. Without a threshold set beforehand, every ambiguous outcome reads as partial validation, and no idea ever gets stopped.
Example: "If fewer than 8 of 25 practice managers say phone tag costs them bookings weekly, we stop and reconsider the segment."
Run the cheapest test that could prove you wrong
Match the test to the assumption. Problem assumptions are tested by conversation. Willingness-to-pay is tested by asking for money, a deposit, or a signed letter of intent — never by asking whether someone would pay. Reach assumptions are tested by spending a small amount to acquire ten people and measuring what it cost. Build only what the test requires.
Example: A concierge test — doing the work manually for three clinics for a month — proves demand and reveals the real workflow without writing software.
Read the result against the criteria, not against your hopes
Compare the outcome to the threshold you wrote in step three. Three outcomes are legitimate: proceed to the next assumption, adjust one variable and retest once, or stop. Repeating the same test hoping for a different answer is not one of them.
Example: If 6 of 25 hit the bar you set at 8, you did not "almost validate". You learned the segment is narrower than you thought — which is useful, and cheap.
Signal vs. noise
The difference between evidence and encouragement is whether the person gave up something — money, time, or reputation. Everything else is conversation.
Counts as evidence
- Money changed hands, or a deposit was placed
Someone rearranged a budget. This is the strongest signal available before launch.
- They described the problem before you did
You did not lead the witness. The problem exists independently of your pitch.
- They already built a workaround
A spreadsheet, an intern, a manual process — evidence the problem costs enough to be worth solving badly.
- They introduced you to someone else with the problem
Unprompted referral is behaviour, not opinion.
Does not count
- "That's a great idea"
Costs nothing to say. Measures your likeability, not the market.
- "I'd definitely use that"
A prediction about a hypothetical future self. Weakly correlated with behaviour.
- A large TAM figure
Market size says a market exists. It says nothing about whether you can reach it.
- Signups with no activation
Curiosity is cheap. Only the second visit means something.
Where founders go wrong
A demo produces feedback on your design. It cannot tell you whether the underlying problem is worth money.
Validating with friendly contacts proves nothing about a channel. Test inside the segment you would actually have to sell to.
A guessed conversion rate copied into a forecast becomes a false certainty within two weeks. Keep ranges as ranges and unknowns as unknowns.
Without kill criteria written in advance, validation becomes an indefinite research project that always concludes 'promising'.
Do this next
- Write your idea as five separate assumptions on one page.
- Circle the one whose failure ends the business fastest.
- Write the number and date that would mean it is false.
- Pick the cheapest test that could produce that number within four weeks.
- Once demand looks real, pressure-test the economics with the break-even calculator — a validated problem at an unviable price is still not a business.
Common questions
What does it mean to validate a business idea?
Validation means gathering evidence that the riskiest assumption behind your idea is true, using the cheapest test that could prove it false. It is not gathering encouragement — it is trying to break the idea before the market does.
How long should validating an idea take?
Most single assumptions can be tested in two to four weeks. If a test would take three months, it is usually testing several assumptions at once and should be split into smaller ones.
How many customer conversations are enough?
Signal usually stabilises somewhere between 15 and 30 conversations within one narrow segment. If you are still hearing genuinely new problems after 30, the segment is too broad to test.
Keep going
Where validation sits in the six stages, and its exit condition.
Turning validated assumptions into a plan someone else can check.
Test whether the price your evidence supports can carry your cost base.
How much validation each funding path expects before it opens.