How to Find Business Ideas: 9 Proven Methods, What the Research Says, and How to Test Yours

Published October 27, 2026

The most reliable way to find a business idea is to notice a problem that you, your industry or a group you know already pays to work around, or to spot something that just changed and ask what it now makes possible. Brainstorming from a blank page is the weakest route: when Y Combinator partner Jared Friedman studied the top 100 YC companies, "at least 70 percent of them" had found their idea organically. Then list several candidates, pick one, and get paying customers before you write much code.

This guide draws on 19 studies, the essays and talks founders learn from, 12 books and dozens of companies, each figure with its source and date.

Key takeaways

Notice, don't brainstorm. At least 70% of the top 100 YC companies found their idea organically (Jared Friedman, YC Startup School).
Insiders win. In US Census data on 2.7 million founders, prior work in the same industry raised success rates by up to 125%, and founders of the top 0.1% of startups averaged 45 years old (Azoulay et al., 2020).
Originality is overrated. Only 6% of the Inc. 500 founders interviewed by Amar Bhidé said they started with a unique product (Boston Fed review, 2001).
Generate several ideas, alone first. Thinking alone before meeting as a group produced about 3 times as many ideas per unit of time, and a better best idea (Girotra et al., 2010).
AI is a strong idea generator, a weak judge. In a Wharton study, 35 of the 40 best-rated product ideas came from GPT-4 and 5 from students, but the AI ideas were less novel (Terwiesch et al., 2026).
Validation means money. Noah Kagan's bar is 3 paying customers in 48 hours. His own untested sports-betting site cost him 6 months and $100K.
Testing changes results. In a randomized trial of 116 Italian startups, founders trained to test hypotheses pivoted more and earned more (Camuffo et al., 2020).
Bad fit kills. 43% of the failed startups studied by CB Insights in March 2026 had poor product-market fit, and running out of cash was the final cause in 70%.

Where do good business ideas actually come from?

Mostly from what founders already know. Twenty-five years of research find that people spot opportunities in the information they already hold, from their job, their industry or their own frustrations, far more often than from a deliberate hunt for something new.

In 2000, Scott Shane followed eight sets of entrepreneurs who each tried to commercialize the same MIT invention, a 3D printing process. One saw drug production, another rapid prototyping, another prosthetic bones. His conclusion, in Organization Science: entrepreneurs discover opportunities related to the information they already possess.

Robert Baron's review of the field (Academy of Management Perspectives, 2006) finds three factors that keep coming back: active or passive search, alertness, and prior knowledge. In a companion study, Baron and Ensley (Management Science, 2006) found that first-time founders judge an opportunity by how new it is, while repeat founders judge whether it can run as a business, down to positive cash flow.

The biggest dataset points the same way. Azoulay, Jones, Kim and Miranda (American Economic Review: Insights, 2020) used US Census records on 2.7 million people who founded a company with employees between 2007 and 2014. Founders of the fastest-growing 1 in 1,000 ventures averaged 45.0 years old, and a 50-year-old was 1.8 times more likely than a 30-year-old to reach that tier. Prior work in the same industry raised success rates by up to 125%. Their summary: "successful entrepreneurs are middle-aged, not young."

Originality matters less than people think. Amar Bhidé interviewed the founders of 100 companies on the 1989 Inc. 500 list. According to the Boston Fed's review of his book, most ventures came from ideas replicated or modified from a previous job, and only 6% of founders claimed to start with a unique product or service. The typical firm started with less than $30K, 41% had no business plan, and more than a third significantly changed their concept at least once.

Users are a large source too. A Kauffman Foundation study of 4,928 US firms founded in 2004 (Shah, Winston Smith and Reedy, 2012) found that 10.7% of startups surviving to age five were founded by people who first built the product for their own use, and 46.6% of those built around a new product. At 3M, Lilien, von Hippel and colleagues (2002) found that projects built with "lead users", customers already hacking their own solutions, were projected at $146M in annual sales after five years, more than 8 times a traditional project (forecasts, not booked sales).

Practitioners agree. Sam Altman, in a Startup Archive clip of one of his talks, says a "very high percentage" of transformational companies began with founders solving their own problem. Saras Sarasvathy, who studied 27 expert entrepreneurs with at least one IPO each, found they start from who they are, what they know and whom they know rather than from a goal (Darden interview).

The popular idea that some people are simply "alert" to opportunities has, per Gaglio and Katz (2001), "equivocal" evidence behind it. Knowledge is the proven ingredient, and the nine methods below are nine ways to use yours, or to borrow someone else's.

What are the best ways to find a business idea? 9 methods that worked

The methods that produced the most successful companies all start from something observable: a problem, an industry, a change, a chore nobody wants, a dismissed niche, a model from elsewhere, an audience, an odd combination, or a business that already earns. Here is each one, with companies that used it.

1. Solve a problem you hit every week

Pick a friction you meet again and again and that you would pay to remove. Paul Graham opens his essay How to Get Startup Ideas (November 2012) with this line: "The way to get startup ideas is not to try to think of startup ideas."

Calendly. Tope Awotona, an enterprise software salesman, lost a full day to email back-and-forth to book one meeting and found no tool that worked. In January 2021 the company raised $350M at a $3Bn valuation, with 10 million monthly users and $70M in subscription revenue the year before, after raising only $550K in total (TechCrunch, Jan 26, 2021).
Slack. An internal chat tool built by Stewart Butterfield's studio while it worked on a game that failed. Its S-1 (April 26, 2019) reported $400.6M in fiscal 2019 revenue, 88,000 paying customers and more than 10 million daily active users. Salesforce closed its purchase for about $27.7Bn in July 2021 (The Register).
Spanx. Sara Blakely could not find underwear that worked under white pants, cut the feet off her pantyhose, and started the company with $5K of savings and no outside investors (Stanford GSB, 2018). In 2021 Blackstone took a majority stake at a $1.2Bn valuation (Retail Insight Network, Nov 19, 2021).

One warning from My First Million: the problems every smart programmer has (to-do lists, dating apps) are the ones everyone is already solving. Your problem is most valuable when few people like you share it.

2. Mine the industry you already know

Your current or past job is the most underused idea source you have. In his Startup School talk, Friedman ranks "what your team has an unfair advantage at" as the best of his idea recipes, and says about half of YC's most successful companies trace back to it (YC Library).

Veeva Systems. Peter Gassner helped build Salesforce's Force.com platform, then bet that cloud software could be built for a single industry: life sciences. Veeva reported $3.19Bn in revenue for the fiscal year ended January 31, 2026 (Veeva IR, Mar 4, 2026).
ServiceTitan. Ara Mahdessian and Vahe Kuzoyan watched their immigrant parents, who ran home-repair businesses, type invoices at the kitchen table at night (founders letter, Dec 2024). The software they built for the trades did $961M in fiscal 2026 revenue, up 24% (8-K, Mar 12, 2026).
Flexport. Ryan Petersen imported scooters from China and co-founded ImportGenius, a freight-data company, before building software for global trade. In early 2019 SoftBank invested $1Bn at a $3.2Bn post-money valuation; by April 2020 the company had raised $1.3Bn and served 10,000 customers (TechCrunch, Apr 26, 2020).

Breadth helps: a study of 496 technology ventures (Gruber, MacMillan and Thompson, 2013) found that teams with more diverse industry experience identified more market opportunities, and more distant ones.

3. Spot what just changed, then ask what it makes possible

A new technology, a new law or a new behavior opens a window before incumbents react. Altman calls it surfing "somebody else's wave" and points to the clusters it creates: Amazon and Google after the internet wave of the late 1990s, then Uber, Airbnb, Snapchat, Instagram and WhatsApp right after the iPhone in 2007 and 2008 (Startup Archive). Graham's version: "Live in the future, then build what's missing." Laws count too: YC's Kevin Hale notes the wave of healthcare startups that followed the Affordable Care Act (YC, 2019).

Cursor. Four MIT students saw GitHub Copilot as the first killer app of large language models, and early access to GPT-4 convinced them to build a code editor designed around AI (Lex Fridman #447, Oct 2024). By November 2025 the company reported more than $1Bn in annualized revenue and a $29.3Bn valuation (Cursor blog, Nov 13, 2025).
Photo AI. When Stable Diffusion came out, Pieter Levels launched AvatarAI, which he says made about $150K in its first week, "most money I ever made" (Lex Fridman #440, Aug 2024). He turned it into Photo AI, which he reports brings in $105K in revenue and $80K in profit a month (levels.io, Mar 6, 2026; self-reported, not audited).

The My First Million hosts call this a "one chart business": on episode #480, they argue that one chart alone can be your business plan, then ask "if that's true, then what?" Their example is cremation, which they say grew from about 3% in the 1960s to over 50% by 2020, and After.com, which pre-arranges cremation for $2.5K, as they describe it. They also retell how Jeff Bezos read that the internet was growing 2,300% a year, then picked books because they offer unlimited selection and ship easily. The trend says where to look; a filter says what to build.

Worked example: from signals to an idea in AI security. Here is how the method runs on live data, using signals collected by ideas.xyz in 2026.

The signals. Large tech companies now report about 2,400 high-severity and more than 600 critical flaws a month, against 250 and about 30 at the start of 2026 (the-nbs.fr, Sept 8, 2026). Bank of America hired 1,000 contractors this year to fix flaws found by Anthropic's Mythos model deployed at banks (Techpresso newsletter, October 16, 2026). HiddenLayer raised $100M for AI runtime security in September 2026 (SecurityWeek).
The trend. Searches worldwide on "AI and cybersecurity" keywords reach about 153,600 a month, up 37% over 12 months (ideas.xyz measurement, October 23, 2026).
The second-order question. If AI makes finding flaws nearly free, what becomes scarce? Fixing them. A bank hiring 1,000 people is behavior, not hype, which is exactly Altman's test for a real trend.
The idea. "Bug Bounty Hunters": a marketplace that sends vetted freelance fixers to companies drowning in AI-found flaws. Finding is cheap now; fixing is the bottleneck.

4. Take on the schlep everyone avoids

Tedious, regulated, unglamorous problems are less crowded because most founders' minds filter them out. Graham named this in Schlep Blindness (January 2012): "Your unconscious won't even let you see ideas that involve painful schleps."

Stripe. Graham's prime example: thousands of programmers knew online payments were painful, and almost none wanted to deal with banks, fraud and regulation. Stripe processed about $1.9 trillion in total payment volume in 2025, up 34%, and ran an employee tender offer at a $159Bn valuation (Stripe Newsroom, Feb 24, 2026). YC's Dalton Caldwell and Michael Seibel add that Stripe was not first and priced itself as the most expensive option, and people still wanted it (YC, 2022).
Gusto. The three founders had personally felt the pain of payroll and health insurance in their previous companies (Josh Reeves in Fortune, 2015). Gusto crossed $1Bn in trailing 12-month revenue in February 2026 and is valued at $9.3Bn (Fortune, May 7, 2026).

Friedman lists three traits that make a good idea look bad: it seems hard to start (Stripe), it sits in a boring space (Gusto), or it already has competitors (Dropbox was roughly the 20th cloud storage player). Arvid Kahl's version is the most useful filter of all: "Find the critical problem at the intersection of something mandatory and something wasteful" (The Bootstrapped Founder, Dec 5, 2019).

5. Go where others look down

The least crowded markets are the ones respectable people dismiss. Arnaud Labossière makes the case in Autistic Antisocial Money: The Secret Art of Suspicious Wealth (May 5, 2025): big money hides in ambiguous, neglected or looked-down-upon niches, and the way to find it is to reject mimetic desire, René Girard's term for wanting what the crowd wants. It echoes Peter Thiel's argument in Zero to One that great businesses are built on secrets, and his reminder that "There are a lot of secrets left" (Chicago Booth Review, Oct 15, 2014). Altman gives the mimetic version in numbers: the year he joined the board of the fusion company Helion, someone counted about 10,000 new photo startups and one nuclear fusion startup.

Christian apps and products. Worldwide searches on this trend's keywords reach about 203,000 a month, up 103% over 12 months (ideas.xyz measurement, October 25, 2026). The leader, Hallow, a Catholic prayer and meditation app, had 10 million App Store downloads when it raised $50M in May 2023 (Built In Chicago). The recent signals point to hardware: Elah launched the Elah Ring S1, a $249 smart ring for Christians with prayer reminders, health sensors and an assistant (Elah, Oct 2026); Glorify, one of the most popular Christian apps, unveiled its own Christian smart ring; and CBN's Superbook launched a free Bible app for kids.
Covet & Mane. Hair extensions sold direct to salons. As one of the hosts told it on My First Million #480, the founder started with $100K to $150K, has six employees, did $4M in her first year and was doing tens of millions in revenue by year four or five (his account).
Daydream Dental. Insurance back-office for dentists, started by a founder whose mother was a dentist and who worked summers at her practice. The host estimated it was "close to 10 million a year" in revenue in year two (My First Million; his recollection).

6. Copy what works somewhere else

Taking a proven model to a new country, vertical or customer is the most underrated method. Shaan Puri's essay Shamelessly Cloning argues that "humans are surprisingly bad at copying successful ideas", so the copier who adds a twist often faces little competition.

Mercado Libre. Founded by Marcos Galperin right after his Stanford MBA and nicknamed the eBay of Latin America; eBay itself bought 20% in 2001 (Stanford GSB, 2011). Net revenue reached $28.9Bn in 2025, up 39% (Mexico Business News, Feb 25, 2026).
Flipkart. Two IIT graduates who had worked in tech, including at Amazon, launched an online bookstore from a small apartment in Bangalore: the Amazon model, transposed to India. Walmart paid about $16Bn for roughly 77% of it (Walmart 8-K, May 9, 2018).
Zalando. An online shoe store "modeled after Zappos in the U.S." that reached €1.8Bn in revenue in 2013, up more than 50% (VentureBeat, Mar 31, 2014).

Friedman calls it "a new variant of a recent success": Rappi copied DoorDash for Latin America, Nuvocargo built Flexport for Mexico. Models also jump verticals: the My First Million hosts trace online booking from yoga studios (Mindbody) to stylists (StyleSeat) to barbers (Squire).

7. Pick an audience first, then find its problem

Choose a group of people you can reach and like serving, study where they complain, and let the product come last. Arvid Kahl's audience-first method (October 2020) starts by listing 100 or more audiences and scoring your affinity for each from 0 to 10, then checks problems, willingness to pay and size. Amy Hoy's Sales Safari asks for 10 to 30 hours of research where the audience gathers before you brainstorm a product (Stacking the Bricks).

FeedbackPanda. Kahl's partner taught English online and spent more than two unpaid hours a day writing mandatory parent feedback, due within 12 hours to get paid. The tool they built reached more than 5,000 customers with no employees and no funding, and was sold about two years after launch (Kahl, 2019).
Morning Brew. Two Michigan students who coached more than 75 classmates for interviews found they all considered the Wall Street Journal too dense. Their newsletter had nearly 200,000 subscribers and about a 50% open rate in 2018 (TechCrunch, Mar 26, 2018). Insider bought a majority stake at about a $75M valuation, per Axios (Talking Biz News, Oct 29, 2020).
Lenny's Newsletter. After leaving Airbnb in 2019, Lenny Rachitsky published 40 free issues in 10 months, most answering readers' questions. Six weeks after turning on paid subscriptions he had about 450 paying readers and roughly $56K in annual recurring revenue; today he has about 1.2 million subscribers (First Round Review, Apr 15, 2026).

Greg Isenberg's version for Reddit: skip the giant subreddits and go to fast-growing communities of 10,000 to 100,000 members, then read the anger, the requests for advice and the requests for a solution (YouTube, 2024). He says the agency Design Scientist became a seven-figure business that way. An audience also lowers launch risk: Hale's Wufoo had 100,000 developers subscribed to its blog before launch, raised only $118K in its whole life and returned over 30,000% to investors (YC, 2019). Rob Walling estimates that only about 4.5% of TinySeed companies had an audience before building (Nov 22, 2023), so this method is still uncrowded.

8. Combine two fields nobody connects

Many big ideas are a familiar model with one unusual ingredient. In science, a study of 17.9 million papers (Uzzi et al., Science, 2013) found that papers combining a highly conventional core with a tail of novel combinations hit the top 5% of citations 9.11 times per 100, almost double the 5-per-100 baseline. Transposing that to business is an analogy, not a finding, but the pattern shows up in companies too.

Duolingo. Luis von Ahn took the principle behind his reCAPTCHA, millions of people doing small useful tasks, and applied it to language learning (MIT Technology Review, 2011). In 2025 Duolingo passed $1Bn in bookings for the first time, with more than 50 million daily active users (8-K, Feb 26, 2026).
Peloton. A home bike plus a studio class plus streaming. Fiscal 2019 revenue was $915M, with 511,202 Connected Fitness subscribers (S-1, Aug 27, 2019).
Dollar Shave Club. Razors made by South Korea's Dorco, sold by subscription, direct to consumers. It did $152M in sales in 2015, and Unilever bought it for a reported $1Bn (Retail Dive, Jul 21, 2016).

9. Buy, or study, a boring business that already makes money

Codie Sanchez's thesis in Main Street Millionaire (2024) is to buy rather than build: baby boomers own 60% of small businesses and many have no successor (Next Big Idea Club, Jan 13, 2025). As she puts it in her playbook video, "boring is often where the billions are."

Her filter. In her 80-minute playbook on YouTube, she targets fragmented blue-collar services under $10M in revenue, profitable today, with at least a 15% margin, priced at 2 to 3 times profit. An HVAC company she reviews on screen, listed at $4M on $10M of revenue, fails her test at a 12% margin.
The Venmo challenge. Export your Venmo and PayPal history, list the small businesses you already pay, and ask what each one is missing: a website, subscriptions, a price increase. Her example is a property manager she pays $6.5K a month, which she estimates at $420K a year with no price increase in three years (her numbers).

Even if you never buy anything, the exercise is an idea machine: every gap you list (no site, no reviews, prices never raised) is a service you could sell to a whole category of small businesses.

Which method fits you? The 9 methods at a glance

Pick the method that matches what you already have: a recurring frustration, an industry, a front-row seat on a new technology, patience for paperwork, a taste for odd niches, travel, an audience, a cross-disciplinary background, or capital.

Method	Where to look	Real example with figure
Solve your own problem	Tasks you repeat weekly and would pay to remove	Calendly: $70M subscription revenue, $3Bn valuation (Jan 2021)
Mine your industry	Every job, client and supplier you have known	Veeva: $3.19Bn revenue, fiscal 2026
Spot what just changed	New tech, new laws, new behaviors, funding news	Cursor: over $1Bn annualized revenue (Nov 2025)
Take on the schlep	Regulated, tedious, mandatory work	Gusto: $1Bn trailing revenue (Feb 2026)
Go where others look down	Niches your peers dismiss or mock	Hallow: 10M downloads, $50M raised (May 2023)
Copy what works elsewhere	Proven models in other countries or verticals	Mercado Libre: $28.9Bn net revenue (2025)
Audience first	Communities of 10K to 100K members, review sites	FeedbackPanda: 5,000+ customers, sold in about 2 years
Combine two fields	A familiar model plus one unusual ingredient	Duolingo: over $1Bn bookings (2025)
Buy a boring business	Small businesses you already pay, retiring owners	Boomers own 60% of small businesses (Sanchez, 2025)

How many business ideas should you generate before choosing one?

Enough to have a real choice, and probably more than you think. Founders who compare several opportunities do better, but each extra idea adds less than the last: make a wide list, then cut hard.

A choice set beats a single bet. Gruber, MacMillan and Thompson (Management Science, 2008) found that serial entrepreneurs learn to build a "choice set" of alternative markets, that identifying several opportunities before entry improves performance, and that the gain is nonlinear with decreasing marginal returns.
Experience widens the list, up to a point. Among 630 British entrepreneurs, experienced founders identified more opportunities and more innovative ones, but those who had owned more than 4.5 businesses identified fewer (Ucbasaran, Westhead and Wright, 2009).
Alone first, then together. In a Wharton experiment, 44 students produced 443 ideas. Working alone for 10 minutes before meeting as a group yielded about three times as many ideas per unit of time as a group session, and better ones, roughly 30 percentile points higher. Both formats were poor at judging their own ideas, and building on others' ideas did not help: those ideas were "not statistically better than any random idea" (Girotra, Terwiesch and Ulrich, 2010).
Quantity yields quality. In a test of 246 students, the number of responses correlated with judged creativity at r = 0.73 (Jung et al., 2015), consistent with Dean Simonton's equal-odds rule: the more works a creator produces, the more hits.

Practitioners land on the same shape. Arnaud Labossière's filter, inspired by the French film Le Dîner de Cons, is 100 → 10 → 1: list 100 niches other people dismiss, research 10 of them deeply, then commit fully to one, and treat the first dollar as the first real milestone (Arnaud Labossière, May 5, 2025). Pieter Levels announced 12 startups in 12 months on March 1, 2014; Nomad List was his fourth launch (levels.io). Josh Pigford went through more than 50 ideas before Baremetrics (Justin Jackson).

A practical routine based on that research:

Keep a "problem book", as Michael Seibel recommends, and add to it every day for a few weeks (YC, 2019).
Generate alone first, then compare lists with a cofounder. Aim for 100 entries.
Cut to 10 with the filters below, then research each where its customers gather; Amy Hoy budgets 10 to 30 hours per audience.
Commit to one and chase the first dollar. Levels gives each launch about a month to show demand.

Can AI come up with good business ideas?

Yes for volume and average quality, no for novelty or for deciding which idea to bet on. The best evidence comes from Wharton.

In Study 1 of a July 2026 preprint (Terwiesch, Meincke, Girotra, Mollick, Nave and Ulrich), the challenge was a physical product for students priced at $50 or less. Researchers compared 200 student ideas, collected before AI tools were common, with 100 GPT-4 ideas generated cold and 100 generated after seeing six high-scoring examples. College-age respondents rated all 400 on purchase intent. The averages were 40.4% for humans, 46.4% for GPT-4 and 49.3% for GPT-4 with examples. Of the top 40 ideas, 5 came from humans, 15 from plain GPT-4 and 20 from the prompted version. Rewriting the human ideas in the AI's style did not change their scores.

The catch: novelty went the other way (40.6% for humans against about 36% for AI), and the AI ideas were less diverse. Use AI to widen your list, not to replace customers: Rob Fitzpatrick's rule is that "customers behavior never lies their words do" (Digital Product School), and a simulated customer has no behavior at all.

How do you tell a good business idea from a bad one?

A good idea solves a frequent, urgent, expensive problem for a market that is growing, and you have some unfair advantage in reaching it. YC partners and investors have turned that into checklists.

Kevin Hale, Wufoo's cofounder, frames an idea as a hypothesis with three parts: the problem, the solution and the insight that gives you an edge (YC Startup School, 2019). His six traits of a good problem make a useful scorecard:

Trait	Question to ask
Popular	Do a lot of people have this problem?
Growing	Is the market growing fast? YC likes 20% a year.
Urgent	Does it need to be solved very quickly?
Expensive	Would solving it let you charge a lot?
Mandatory	Do people have no choice, as with healthcare rules after the Affordable Care Act?
Frequent	Does it come back over and over, like Slack several times a day?

Hale adds five unfair advantages: the founder, a growing market, a product 10 times better, free acquisition through word of mouth, and a monopoly through network effects. A growing market alone, he warns, is the weakest of them.

Friedman's longer list of questions, from his 2022 Startup School talk, adds examples to each:

Founder-market fit: PlanGrid paired a construction industry veteran with a strong developer.
Market size: Coinbase's market was tiny in 2012 but growing fast.
Acuteness: no bank would give Brex's customers, startups, a credit card. Good problems are ones where "the alternative to your solution is literally nothing".
Competition: usually a good sign, not a reason to quit.
Do you want it yourself?
Did something recently change? Checkr rose with gig platforms.
Is there a proxy? DoorDash proved demand for Rappi in Latin America.
Would you work on it for years, and can it scale?
Is it a good idea space? Fivetran pivoted twice inside data tooling and kept the same customers.

Sam Altman adds the trend test (How to Succeed with a Startup, August 2018). The key skill, he says, is "learning how to differentiate between real trends and fake trends." In a real trend, early adopters use the product obsessively and tell their friends: the iPhone in 2007. In a fake trend, people buy but don't use: his example, as of August 2018, was VR headsets that owners rarely put on. His Startup Playbook adds that the best answer is often a large share of a small market, and that "The best ideas sound bad but are in fact good."

Marc Andreessen puts market above everything (Pmarca Guide, 2007): "When a great team meets a lousy market, market wins." His 2007 examples of bad markets: videoconferencing, workflow software and micropayments. For a number, Josh Kaufman's Personal MBA scores a market on 10 factors from 0 to 10: 50 or less means move on, 75 or more means very promising (personalmba.com).

How do you test a business idea before building it?

Ask people about what they already do, then ask them for money. A payment, a deposit or a pre-order is evidence. A compliment, a survey answer or "let me know when it launches" is not.

Talk about their life, not your idea. Rob Fitzpatrick's The Mom Test (2013) rests on asking about past behavior: the last time you had this problem, how did you solve it? Once you pitch, people only give you opinions. "The future is always a lie," he told the SaaS Podcast. He once killed his own idea, a deal-flow tool for investors, over one coffee: thousands of pitches a week were already filtered down to 10 or 20, in a couple of hours. No real problem.
Measure what they spend now. People predict their own spending badly. In his Digital Product School talk, Fitzpatrick describes a customer with six full-time people on a problem he had planned to price at about $2K a year, roughly a $1M-a-year problem by his estimate. His pricing rule: guess a price, then keep doubling it until people stop paying.
Watch the job, not the demographics. Clayton Christensen's team found that about 40% of a chain's milkshakes were bought in the morning by commuters, who "hired" them to make a long drive less boring. Demographics, he said, "may be correlated with the decision, but it doesn't cause it" (HBS Working Knowledge, 2011).
Get three paying customers in 48 hours. Noah Kagan's bar in Million Dollar Weekend (2024): "If you can get three customers, to me that's validation." His example is Jake, a golf-trip organizer: about $2.5K profit per trip means roughly 1,000 sales to reach $1M, and Jake took 5 deposits totaling $500 in 48 hours (Tim Ferriss Show #717). For B2B, Altman suggests trying to sell before building, for example with a letter of intent.

Four cheap tests that worked:

Buffer: a pricing page before the product. Joel Gascoigne put up a landing page, then a pricing page to see who would click a paid plan. The product took seven weeks to build and the first paying customer came four days after launch (Buffer, Feb 16, 2011). As of October 26, 2026, Buffer's public dashboard shows $27.7M in annual recurring revenue (buffer.com/transparency).
Dropbox: a demo video. Drew Houston posted a video full of in-jokes for the Digg community before the product was finished; the beta waiting list went from 5,000 to 75,000 people overnight (Eric Ries in TechCrunch, Oct 19, 2011). Dropbox reported $2.52Bn in revenue for 2025 and 18.08 million paying users (Feb 20, 2026).
Nomad List: a public spreadsheet. In June 2014 Levels tweeted a public Google Sheet of cities for digital nomads and let users fill it in. The site that followed hit number one on Product Hunt and Hacker News on July 29, 2014, drew about 50,000 visitors from Hacker News and 2,500 email signups in two weeks, and was profitable from day one (levels.io, Aug 16, 2014).
AvatarAI: a Stripe link and a Typeform. Levels launched with an HTML page, a Stripe payment link and a Typeform to collect photos, and processed the first few hundred orders by hand (Lex Fridman #440). The counter-example is his Hoodmaps: viral, never made a dollar, and left him with a $20K Mapbox bill, by his account. Attention is not demand.

Does testing change results? A randomized trial says yes. Camuffo, Cordova, Gambardella and Spina (Management Science, 2020) followed 116 Italian startups for about a year; only the treated group was taught to state hypotheses and test them rigorously. Treated founders were more likely to pivot (19 of the 30 firms that pivoted), and among observations with non-zero revenue, average revenue was about €7.8K against €900 for the control group. Steve Blank's line sums it up: "there's no facts inside your building" (Innovation Engine podcast).

What mistakes kill business ideas most often?

Starting from a solution, chasing an idea that has failed many times, and building for a market that isn't there. Failure data and YC partners point to the same short list.

A solution in search of a problem. YC calls it a SISP: starting from "AI is cool, what could I apply AI to?" rather than from a person with a problem (Friedman). Seibel's antidote: "Be in love with your problem."
Tarpit ideas. Ideas that many founders pitch and that keep failing, such as restaurant discovery or the app for planning outings with friends, which Friedman says has been pitched to YC for 20 years. Caldwell and Seibel note that tarpits get lots of praise from friends, and that "it's been done a lot, a lot of people have failed, and nothing has changed" (Tarpit Ideas: The Sequel, 2024). A new technology such as LLMs can reopen one, but you have to say what changed.
Sitcom ideas. Graham's name for ideas that sound plausible but that almost nobody actually needs.
Waiting for the perfect idea. Google was roughly the 20th search engine. Justin.tv started as an online reality show, became Twitch and sold to Amazon for almost $1Bn (Seibel). YouTube began as a video dating site (Steve Chen at SXSW, via Tubefilter, 2016), and Twitter was a side project of the podcasting startup Odeo (History.com).
A lousy market. Low-demand categories such as to-do apps look easy and pay badly. Justin Jackson's rule: "No demand? No business." Fitzpatrick's cautionary tale is Pets.com, which by his telling spent about $300M, including Super Bowl spots, to learn that pet food does not sell online like books.

The failure data agree. CB Insights' March 5, 2026 edition studied 431 venture-backed companies that shut down since 2023. 70% ran out of capital, which the report calls "almost always the final cause of death, not the root problem." The roots: 43% had poor product-market fit (two thirds of them at seed stage), 29% cited bad timing and 19% unsustainable unit economics. The often-quoted "42% no market need" comes from an older edition based on 101 post-mortems.

Changing course is normal, not a failure. Bhidé found that more than a third of Inc. 500 founders significantly altered their concept at least once, and a 2011 study of more than 650 web startups found that those that pivoted once or twice raised 2.5 times more money and had 3.6 times better user growth than those that pivoted more often or never (Startup Genome, via TechCrunch; a correlation, not proof). Research on seven energy and cleantech firms shows pivots usually happen as a series of small changes rather than one dramatic decision (Kirtley and O’Mahony, 2023).

Frequently asked questions

How do I come up with a business idea if I have no ideas at all?

Start with people, not ideas. Rob Fitzpatrick suggests choosing someone you want to serve and studying their life until problems appear. Jared Friedman's exercise: list every job and project you have had and the problems you saw there.

What is the best way to find a profitable business idea?

Follow money people already spend. Look for problems that are frequent, urgent and expensive, where buyers already pay for a workaround or staff to handle it. Noah Kagan's one-minute business model checks whether $1M is reachable: profit per sale times the number of sales you can realistically make.

How many business ideas should I look at before choosing one?

Several. Gruber and colleagues (2008) found that founders who identify multiple opportunities before entry perform better, with diminishing returns. Arnaud Labossière's 100 → 10 → 1 filter is a practical version: list 100 niches, research 10 deeply, commit to 1.

Does my business idea need to be original?

No. Only 6% of the Inc. 500 founders Amar Bhidé interviewed said they started with a unique product, and Google was roughly the 20th search engine. What matters is being clearly better for a specific group: Sam Altman says "new" can mean 10 times better.

How long does it take to validate a business idea?

Days to weeks, not months. Noah Kagan's target is three paying customers in 48 hours. Pieter Levels gave each of his 2014 launches about a month to show demand. Rob Fitzpatrick says the learning phase is measured in calendar weeks, so you can run it alongside a full-time job.

Can ChatGPT give me a good business idea?

It can give you many decent ones. In a 2026 Wharton study, GPT-4 produced 35 of the 40 best-rated product ideas, against 5 from students, but its ideas were less novel and less diverse. Use AI to widen your list, then validate with real customers.

Should I start a business or buy one?

It depends on your capital and goals. Buying gets you customers and cash flow on day one: Codie Sanchez points out that baby boomers own 60% of small businesses and many have no successor. Starting costs less: Chris Guillebeau's survey of more than 1,500 entrepreneurs earning at least $50K a year found an average startup cost of about $450 (Entrepreneur.com, 2012).

Where can I find business ideas online?

Fast-growing Reddit communities of 10,000 to 100,000 members, recurring complaints on review sites, feature requests competitors turn down, job listings, search trends and funding announcements.