How to Validate a Startup Idea Before You Build (2026)

Most startups do not fail because the product was bad. They fail because nobody wanted it. CB Insights has put "no market need" at the top of the failure list for years, around 42 percent of post-mortems. The good news: you can find that out in an afternoon instead of a year.

Here is how to validate a startup idea with evidence, not optimism.

Start with the problem, not the product

Write your idea as one sentence: "X for [who] solving [Y]." If you cannot fill in the who and the why, you do not have an idea yet, you have a feature. The "who" is the most important word in that sentence. A tool for "everyone" is a tool for no one.

Look for a starving crowd, not a clever idea

The order that matters is market first, then offer, then product. A average product sold to a desperate audience beats a brilliant product sold to people who do not care. So before you build, go find the people who already have the problem and listen to how they describe it.

The cheapest place to do this is where people complain in public: Reddit, niche forums, review sites, and communities. Search for the phrases your customer would actually type when they are frustrated: "how do I", "is there a tool that", "I hate that", "I wish". Real demand looks like dated posts you can click, with other people agreeing in the comments.

Size the market honestly

You do not need a precise number. You need to know if the market is big enough to matter and reachable by you. Ask three questions: how many people have this problem, how many can you actually reach, and how many have money to pay. If the answer to any of those is "very few," you have your answer.

Define a pass or fail test

Validation is not a feeling, it is an experiment with a kill criterion you set in advance. For a web product, get a landing page in front of 1,000 of the right people and measure if they sign up or pay. For a service or local business, talk to 20 potential buyers and count how many say "yes, I would pay for that today."

Write the number down before you run the test. "Pass = 5 of 20 say yes." If you do not, you will move the goalposts to protect your idea.

Do not let AI flatter you

If you ask a chatbot "is my idea good," it will almost always say yes, in a confident and detailed way. That confidence is the trap. It is generating an encouraging answer, not checking reality. Validate against real people and real data, not a generated opinion.

The fast version

You can do all of this manually. Or you can run a free DemandSonar scan: it mines the real demand for your idea, defines your ICP, and gives you the exact test to run, in about ninety seconds.

The point is the same either way. Find the evidence before you write the code.