All articles
idea validationstartup buildingcustomer interviews

How I Decide Whether a Business Idea Is Worth Building Before Writing Code

My build-or-walk system to validate a business idea before code: three kinds of evidence, a five-step sequence, and the scorecard I fill first.

On this page

Every idea looks brilliant at 11 p.m., and most of them should never leave your notes app. The way I decide whether a business idea is worth building comes down to three kinds of evidence: problem evidence, demand evidence, and willingness-to-pay evidence. I want at least two of the three before I write a line of code.

That rule has saved me from building things nobody wants more times than I can count. I’ve killed far more ideas than I’ve shipped, and the graveyard file is the most useful asset I own. To validate a business idea properly, you don’t need a methodology deck. You need a sequence of cheap tests and the discipline to walk away when they fail.

Here’s the exact sequence I run, the scorecard I fill, and the places where validation lies to you.

The three kinds of evidence

Every build-or-walk decision hangs on three questions: is the pain real, do people look for a fix, and will they pay you for yours. If you can’t answer all three with evidence, you’re guessing, no matter how good the idea feels. Here’s what each kind of evidence actually means.

Problem evidence is proof that people hit the problem often enough to do something ugly about it, like spreadsheets, duct-taped tools, or hired help. Demand evidence is proof that people actively search for and talk about solutions, not just nod politely at yours. Willingness-to-pay evidence is proof that someone gives up something scarce (money, time, reputation) to make the problem go away.

The failure data backs this ordering. CB Insights’ current report on why startups fail, updated in March 2026, analyzed 431 VC-backed companies that shut down and found poor product-market fit in 43% of them. Seventy percent “ran out of capital,” but the report itself is careful to call that the final cause of death, not the root problem. Two-thirds of the product-market-fit failures were early-stage companies that never found a market.

An earlier version of that same analysis put “no market need” at 42% back in 2018. The label changed over the years; the lesson didn’t. Building something nobody wants is still the most reliable way to waste a company.

The five checks I run, cheapest first

I test in order, cheapest test first, and stop the moment an idea flunks one badly. Nothing in this sequence requires a product. Most ideas die at step one or two, which is exactly why those steps cost nothing.

1. Look for problem evidence in real behavior.

Complaints are cheap, so I hunt for workarounds instead, workarounds cost people time, and nobody spends time on optional pain. Hand-built spreadsheets, two tools stitched together with Zapier, a virtual assistant paid to do it manually, a department quietly funding an intern to copy-paste between systems. These are people voting with their calendars.

I read niche subreddits, support forums for adjacent products, and job boards, and I pay special attention to anyone paying a human to do the thing. Paying a human is the most expensive workaround there is, which makes it the strongest problem evidence you can find for free. If nobody has rigged up even a bad workaround, the pain is probably optional.

2. Read search and community demand honestly.

Search data is my demand-evidence baseline because it shows people looking for solutions without me in the room. Google’s Keyword Planner lets you check estimated monthly search volume for free, and Google’s own help docs describe the numbers as estimates, forecasts refreshed daily off the last 7–10 days, adjusted for seasonality.

Read it honestly, because the tool won’t do it for you. The numbers often come as ranges rather than exact figures, very low-volume and sensitive keywords aren’t discoverable at all, and some of your buyers now research inside AI chatbots that keyword tools can’t see. Gartner predicted back in February 2024 that traditional search volume would drop 25% by 2026 as chatbots absorb queries, the exact figure is debatable, the direction isn’t.

Keyword difficulty needs the same honesty. Ahrefs’ Keyword Difficulty scores 0–100 based on how many referring domains the current top-10 pages have, and Ahrefs’ own guide admits it’s an absolute, backlink-only number that ignores your site’s authority and the search intent. So I treat it as a rough filter, then open the actual results, if forums and thin pages rank on page one, a new site can crack it too.

3. Run conversations that pass the Mom Test.

Rob Fitzpatrick’s The Mom Test (2013) is still the best handbook on this, and its subtitle tells you why: “how to talk to customers & learn if your business is a good idea when everyone is lying to you.” Its three rules are worth memorizing. Talk about their life, not your idea. Ask about specifics in the past, not opinions about the future. Talk less, listen more.

The past is evidence; the future is hope. What counts as real signal is specific, too: compliments are noise, while commitment means the person gives up something they value (time, reputation, money) for a next step.

One rule I’ve adopted from First Round Review’s September 2025 roundup of validation tactics: don’t pitch friends, because social pressure ruins the data. Sprig’s founder Ryan Glasgow cold-emailed strangers instead, on the logic that if people who don’t know you will spend an hour on your problem, you’re onto something. I stop interviewing when answers become predictable, usually somewhere around 10–15 conversations.

4. Run the cheapest possible smoke test.

A smoke test is a landing page that presents your product as if it already exists and measures how many visitors try to buy it. The folks at early.tools define it exactly that way, a realistic page, a buy flow that redirects to a “coming soon” reveal, and a conversion number you commit to before spending on traffic.

That last part matters most. Decide the bar first (“at least 3% of visitors attempt to purchase at this price”) then send targeted visitors and judge the result without negotiating with yourself. For reference, the median landing page converts around 6.6% across industries, per Unbounce’s Q4 2024 data from 41,000 pages, but low-commitment asks like email signups need much higher bars than purchase attempts.

So I ask for money: a pre-order, a refundable deposit, a paid pilot slot. Email captures curiosity, and deposits capture intent. Smoke tests work best for simple self-serve products. For complex B2B with long sales cycles, the equivalent is a signed paid pilot or letter of intent, which is slower but unambiguous.

5. Hold it against the build bar.

My bar is blunt: two of the three evidence types at green, and willingness to pay must be one of them. Problem evidence without payment is a charity project. Payment signals without problem evidence are a fad.

Two extra questions sit at the bottom of my scorecard. Can I reach these buyers this month without a budget I don’t have? Distribution is decided at validation time, not after launch. And which single assumption, if wrong, kills this idea, did I actually test that one, or the comfortable one?

The scorecard I fill before saying yes

I score every idea on the same eight signals before any code gets written. It’s qualitative on purpose, because false precision is just lying to yourself with decimals. Mostly green means build; a red in willingness to pay means walk.

SignalGreen flagRed flag
Problem visibilityPeople complain unprompted and already use workaroundsYou have to convince them it’s a problem
Search demandReal monthly volume for the problem, plus community chatterVolume too low for tools to even estimate
CompetitionSeveral players making money, demand is provenZero competitors, or one giant for whom it’s a feature
Interview signalStories of past pain, failed fixes, money spentCompliments, hypotheticals, “would you use this?”
CommitmentDeposits, pre-orders, signed pilots, real time givenEmail signups, “keep me posted,” likes
ReachThree places you can find buyers this month”It’ll go viral” or “ads, eventually”
DifferentiationA reason to switch that incumbents can’t copy quicklyOnly cheaper, or only prettier
EconomicsPlausible price times reachable buyers beats your costsNeeds 100,000 customers at $5 to matter

Two red flags anywhere is a pause, and a red in commitment or economics is a kill. Ideas rarely fail on one dramatic signal. They fail as a pattern of yellows nobody wanted to read.

When validation lies

Validation is data about people, and people are polite, optimistic, and terrible at predicting themselves. I’ve been fooled by every pattern below at least once. The data was never fake, I just read it wrong.

  • Buyers who say yes. Ask “would you use this?” and people say yes because it’s kind, not because it’s true. The fix is structural: talk to strangers, ask about the last time it happened, and end every conversation by asking for something that costs them something.
  • Crowdsourced noise. Upvotes, waitlist counts, and launch-day traffic measure curiosity, not purchasing. Communities reward novelty and entertainment, and a loud niche of two hundred redditors can drown out a quiet market of twenty thousand buyers. I use communities to find language and problems, never to decide go/no-go.
  • Small samples. Five friendly interviews are not a market, and five rejections from the wrong audience aren’t a verdict either. The real skill is telling “no signal” apart from “negative signal”, one means dig deeper, the other means stop.
  • The derivative-idea trap. Competition proves demand, but it doesn’t prove you’ll win any of it. Cloning a working product is easy; the question that matters is who switches to you and why. Meanwhile, “no one is doing this” usually means no one wants this, a market with zero competitors is a warning, not a moat.

Then there’s the opposite error: validating when you should just build. If you can ship a rough version in a weekend with AI-assisted tools, you’re your own target user, and being wrong costs days, build it and let usage argue for you. The loop matters more than the ceremony: cheap ideas get built and tested, expensive or unfamiliar ones get validated first. What I refuse to do is spend three months researching what a weekend would have answered.

How I decide in a week, not a quarter

Validation should end in a decision, so I timebox it to two weeks and track a short list of numbers. When the deadline arrives, the numbers decide, not my mood, not the sunk cost of research. If they aren’t there, the idea joins the graveyard file, and that counts as a win too.

The checklist I fill before the clock runs out:

  • 10+ conversations with people who actually have the problem, most of them strangers
  • 3+ unprompted complaints or documented workarounds
  • A keyword set with real estimated volume and a results page a new site could crack
  • At least 100 targeted visitors to a smoke test page
  • 3+ purchase attempts, deposits, or signed pilot conversations

Missing one number doesn’t kill an idea automatically. Missing most of them does, and pretending the deadline doesn’t exist is how “validating” turns into a full-time job. Over-validation is its own trap: endless research always feels more productive than shipping or killing, which is exactly why it’s dangerous.

FAQ

How do I validate a business idea without building it?

Combine all four cheap signals: observe workarounds, read search and community demand, run past-focused interviews, and smoke test the offer. You never need a product for the first three, and the fourth only needs a page. The strongest no-build proof is money, a deposit, pre-order, or paid pilot.

What signals say an idea is worth building?

Two of the three evidence types at green, with willingness to pay as one of them. In practice that looks like unprompted complaints, buyers already spending time or money on bad fixes, strangers giving you their time, and at least a few people committing money before the product exists.

How do I find out if people will actually pay?

Ask for money before you build, pre-orders, refundable deposits, paid pilots, letters of intent. Stated willingness to pay almost always overstates the real thing, so treat every “I’d definitely buy that” as zero until it arrives with a payment attached.

When should I skip validation and just build?

When the build costs days rather than months, you’re your own user, and being wrong is cheap. Ship the smallest useful version and let real usage be the test. If the build is expensive or the market is unfamiliar to you, validate first.

How long should validation take?

One to two weeks for a self-serve or consumer idea, closer to four for B2B with real sales cycles. If you’re still “gathering data” after a month, you’re usually avoiding a decision.

Does keyword research really prove demand?

It proves people search for solutions, which is problem-aware demand, the most honest signal you can get without talking to anyone. It can’t see buyers who describe the problem differently or ask an AI assistant instead. Treat volume as one input in the scorecard, never the verdict.

Sources

Questions & answers

What is "How I Decide Whether a Business Idea Is Worth Building Before Writing Code" about?

My build-or-walk system to validate a business idea before code: three kinds of evidence, a five-step sequence, and the scorecard I fill first.

Who wrote this article?

Aditya Mallah is a growth marketer for SaaS, AI tools, and fintech. Full bio: https://adityamallah.com/about

About this article

The content on this blog is for educational and informational purposes. These are my own views and do not constitute professional, financial, legal, or business advice. I can be wrong; do your own research before making decisions. Third-party links are provided for convenience and do not imply endorsement.

Was this article helpful?

Get my growth notes

Notes on growth, AI, and finding customers. Delivered to your inbox.

All articles

keyword research 11 min read

AI Keyword Research Tools: Can They Replace Traditional SEO Platforms?

AI keyword research tools handle ideation and clustering well, but they can't produce real search volume. Here's the split that actually works in 2026.

startup building 12 min read

What Building a SaaS Product Taught Me About SEO and Distribution

Lessons from building a SaaS product: keep the scope focused, plan SEO early, and start reaching potential customers before launch.