How to validate a business idea before you build it

The checks are simple. Get twenty real names of people who’d buy it. Work out whether you came up with the idea or just agreed with it. Ask what would have to be true for the answer to be no. Then be honest about whether you can run the thing once it exists.

None of this is new, and most of it isn’t mine. What’s changed is how badly it’s needed, because the gap between an idea that sounds good and an idea that is good has never been easier to miss. I’ve written about why that is in the article Centaur or reverse centaur: where AI belongs in your business. The short version is that if you ask an AI to build a case for your idea, it builds one, including for a bad idea, because a case is what you asked for.

Check one: twenty real names

Natalie Bullen of Unapologetic Wealth, whose website system I built to support her offers, put it plainly in a Facebook post a while back: “Make a list of 20 people who would actually buy it. Real names.” If you’re not sure whether someone belongs on that list, her advice is to go and ask them and get the intel. Don’t launch to zero leads.

Twenty is a well-chosen number. It’s small enough that you can’t hide behind an addressable market, and big enough that you can’t fill it with friends being kind to you. If you can only get to six, you don’t necessarily have a bad idea. You have an idea for six people, which is a real business with completely different economics, and you’d want to know that before you build a funnel for it.

Getting the names is the easy part. The harder part is the conversation, because asking people about your idea is the fastest way to get a polite lie. Rob Fitzpatrick’s The Mom Test is the best short book I’ve read on that. In the book, he suggests asking people what they’ve already done rather than what they’d do, because anything about the future is an optimistic lie that costs them nothing to say. Then treat commitment as currency: a compliment is free, so it tells you nothing, while time, reputation, and money are the three things a person has to actually give up.

I’ve developed a simple tracker I use to help quantify the feedback from your twenty chosen people. Four columns against each name: how you know them, whether you’ve actually asked, what they said, and which of those three currencies they’ve put up. The last column does most of the work and is what really tells you if you’ve got something worth pursuing.

The uncomfortable truth is that this check takes an afternoon and most people still don’t do it. The answer arrives too fast, so it’s easy not to trust it. A chunk of time going down a rabbit hole developing a minimum viable product (MVP) and building a landing page to sell it feels more like progress than one afternoon of asking people whether they want the thing, even though only one of those tasks tells you anything.

This is where I part ways with the lean startup idea, or at least with what it’s turned into. Eric Ries argued for shipping an MVP and learning from what happens, which is reasonable enough. What it became, in practice, is permission to build first and ask later. It slotted neatly into a culture that already read visible effort as evidence of progress, the same one that gave us “move fast and break things”.

Twenty names costs you an afternoon. An MVP costs you three weeks and answers a question you could have asked directly. Building something is one of the more comfortable ways to avoid finding out whether anyone wants it. I’ve done it myself.

Check two: did you come up with it, or just agree with it?

Sarah Moon, a marketing and business strategist whose advice is always refreshingly direct, makes the point that your expertise got you this far and you already know what you’re doing. At first glance, this pulls against check one. One says go and get external evidence. The other says trust your gut.

They’re doing different jobs. Your instinct comes from years of watching your market up close, and it’s the best tool you have for deciding what’s worth testing at all. What it can’t do is tell you whether people will pay. That comes from the twenty names. Instinct picks the direction, evidence confirms the appetite, and problems start when you use one to do the other’s job.

There’s an AI-specific version of this worth watching for. Reading something and thinking “yes, that’s exactly right” is a much weaker signal than arriving at the same conclusion yourself. Recognition feels like confirmation and isn’t. If you can’t remember whether the idea was yours before your AI chat started, that’s information.

Check three: what would have to be true for this to be a no?

This one came out of a conversation with the founder of an Australian service-based business I do strategic business consulting for. Rather than asking whether the idea is good, ask what would have to be true for it to fail, then go and check those specific things.

It’s a small reframe that changes what you go looking for. “Is this a good idea?” sends you hunting for support, and you’ll find it, because support is easy to find for almost anything. “What would sink this?” sends you at the load-bearing assumptions. Usually there are two or three, they’re specific, and they’re checkable in a week.

This is also where paying someone can be the validation. An outside expert’s real value is that they have no stake in the idea being good. You have a stake. Your friends have a stake in you being happy. An AI has no stake at all, which sounds neutral but means it also carries no consequences for being wrong. Someone who’ll be embarrassed in twelve months if they told you the wrong thing is a different proposition entirely.

The downside is that it costs money and you have to actually take the answer. Buying an opinion and then arguing with it is an expensive way to feel validated.

Check four: can you run it once it exists?

Everyone validates demand. Almost nobody validates delivery, and this is the check I’ve got wrong more than once.

A new idea/offer/product isn’t one job. It’s five. You have to build it, market it, sell it, deliver it, and then maintain it for as long as it exists. Most people scope the first one honestly and hand-wave the rest. Maintenance is the one that gets you, because it never ends and it doesn’t feel like work until you’re doing it for three different offers at once.

So before you commit, write down what launching the thing actually involves. For me, that’s usually a landing page, an email sequence, two or three supporting articles that auto-generate some socials. Ads only if there’s already a funnel that converts without them. Your launch plan will be different, but if you can’t name the steps and roughly how long each one takes, you don’t have a launch plan, you have a vague intention that is impossible to quantify.

The test isn’t whether you could do all five of those jobs that sit around your new idea. You probably could. It’s whether you can do all five for this offer while still doing them for everything you already sell.

Where AI can actually help

Search for how to validate a business idea and you’ll turn up a hundred versions of the same prompt, all promising to do it for you.

The structure is always roughly the same. You tell the AI model you’re using to take on a role, usually something like “venture architect” or “solopreneur strategy advisor”. You give it your concept, your target audience, your skills, how many hours a week you have, and your budget. It then walks a fixed protocol: is the problem a painkiller or a vitamin (i.e., a thing people have to fix versus a thing that would be nice to have), who else already solves it, could one person deliver this repeatedly, here are three ways to make money from it, here’s your positioning. At the end, it produces a formatted report with green flags, red flags, and a viability score out of 100. Good versions of this prompt also tell the model to be brutally honest and not over-validate weak ideas.

The report reads well. And that’s where the problems start.

Every input to that score is your own description of your idea. The model has no information about your market, your buyers, or what anyone in your local area currently pays for the nearest equivalent. It’s pattern-matching your paragraph against everything it has read, then expressing the result as a number. Seventy-eight out of 100 looks like a measurement but has none of the substance, and a number is far stickier than a purely text response. You’ll still remember the 78 long after you’ve forgotten every caveat that was ignored to arrive at that number.

The individual stages the prompt looks at have the same issue. Painkiller versus vitamin is a real and useful distinction, but the AI model is guessing at it from how you phrased things. Describe the same idea in confident language and then in cautious language, and you’ll often get different results. The competitor matrix comes out of large-language model training data, so it’s probably months or years out of date, and it won’t know about the two people in your niche who are actually selling to your buyers.

The brutal honesty instruction sounds good, but makes the AI worse rather than better. Asking for brutality produces brutal-sounding text. The model will generate red flags because you’ve told it a good answer contains red flags, and you finish reading with the distinct feeling of having been put through the wringer by something that checked nothing. That feeling is the danger. It’s much harder to go and do the uncomfortable work of asking twenty real people after you’ve already been told your idea scores 78 and here are its three weaknesses.

So I’d tweak the questions these prompts ask and throw away the score they produce. What the same tool is genuinely good at is check three. Describe the idea and ask what would have to be true for it to fail, then get those ranked by how cheaply each one could be tested.

The prompt I’d use instead

Here’s my idea: [describe it in a paragraph, including who it’s for and what you’d charge]. List the assumptions this idea depends on to work. For each one, tell me how I could test it in under a week and what that would cost. Rank them by how badly the idea breaks if the assumption turns out to be wrong. Don’t tell me whether the idea is good.

That last line matters. You want a list of things to go and check, not a verdict. Producing the list is the AI’s job. Working out what the answers mean is yours (you’re the centaur), and so is the afternoon of asking twenty people, which no prompt is going to do for you.

What to do when a check fails

A failed check rarely kills an idea outright. What it does is tell you which part of the idea is wrong, and the four checks fail in different ways, so they point at different fixes.

Here are some examples of how that could play out:

1. If you couldn’t get to twenty names, look hard at who you did get. Six people who all run the same kind of business says something about audience rather than demand. The offer might be right and the market wrong, or the reverse. Work out what those six have in common, then go looking for twenty more of exactly them before you conclude anything about the idea itself.
2. If you couldn’t tell whether the idea was yours, put it away for a week and write it out again from memory with the chat closed. Whatever survives is yours. Whatever you can’t reconstruct was AI scaffolding.
3. If an assumption turned out to be expensive or impossible to test, restructure so the risky part comes last. Sell the manual version to five people before you build anything that automates it. You’ll have tested the assumption by doing the work rather than by researching it, and you’ll get paid while you find out.
4. If the capacity check is the one that failed, change the format, rather than the idea. The same expertise might be a done-for-you service, a small group program, a one-off workshop, a template someone buys and runs themselves, or something you license to a person who delivers it for you. Those are wildly different in delivery load and maintenance, and the thing you know (your expertise) can be identical inside all of them. This is the most common fix available, and usually the last one people reach for, because switching the format of your idea feels like a retreat from the original plan. It isn’t. It’s the plan meeting what you found out. And that’s actually a win.

The point of running all four of these checks early is to learn this in an afternoon rather than in month four, when you’ve got a landing page, an email sequence, and no buyers.

Pick whichever idea you’re closest to building and do check one this week. Twenty names, real ones. It’s the cheapest information you’ll ever buy about your own business, and if the list fills up easily, you’ve lost an afternoon and gained a launch list.

I’ve put the sheet I use for this together as a download, twenty rows with the four columns above already set up. It’s free, and you’ll get my newsletter with it, which goes out every month with one useful thing you can do in about ten minutes. Grab the sheet below.

Validate your ideas with the Twenty Names Check

Get the spreadsheet I use to get unbiased quantification of my ideas in an afternoon.

Get the Twenty Names Check spreadsheet →

Sources

• Rob Fitzpatrick, The Mom Test: How to Talk to Customers and Learn if Your Business Is a Good Idea When Everyone Is Lying to You (2013). About 130 pages, which is the best argument for reading the book rather than a summary of the book. Both ideas I’ve used are his: ask people what they’ve already done rather than what they say they’d do, and treat commitment as a currency in which only time, reputation, and money matter.
• Eric Ries, The Lean Startup: How Today’s Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses (Crown Business, 2011). Read it before you argue with it. My disagreement is with what the minimum viable product turned into in practice, not with what Ries actually wrote.
• Natalie Bullen, Unapologetic Wealth. The twenty names advice comes from a Facebook post of hers.
• Sarah Moon, Sarah Moon Consulting. A marketing and business strategist in Portland, Oregon, who specializes in search for service providers and other experts. Sarah is excellent. If you need search advice, or help crafting your framework, drop everything and go talk to her. Sarah helped me develop my Amplify Your Clever framework, and it’s now the filter around everything I do.