How to Find a Startup Idea: Search for Evidence, Not Inspiration

After learning to build with AI, many founders meet a harder question than how to make software: what deserves to be built?

The common sequence is backwards. A founder brainstorms, asks an AI for twenty ideas, chooses the most exciting one, and starts coding. The product becomes concrete before the problem has earned attention.

Do not search for inspiration. Search for evidence.

The six signals behind a useful startup opportunity

No single signal proves that a company should exist. A loud complaint may have no budget behind it. A growing category may be impossible for you to enter. A profitable incumbent may serve customers who have no reason to switch. The opportunity becomes more credible when several independent signals overlap.

1. Pain: a repeated task, delay, risk, cost, frustration, or unmet desire that matters to a specific person.
2. Spending: customers already pay for software, staff, contractors, workarounds, or the consequences of leaving the problem unsolved.
3. Direction: the need is durable, emerging, or becoming more urgent rather than disappearing.
4. Gap: current solutions are too expensive, complicated, slow, unreliable, generic, poorly supported, or designed for a different customer.
5. Distribution: you can identify where the customer searches, gathers, buys, asks for help, or trusts recommendations.
6. AI leverage: AI can materially improve the speed, cost, quality, access, personalisation, or operating model of the outcome without removing necessary human judgment.

Startup opportunity = pain × spending × direction × gap × distribution × AI leverage

Build a fixed evidence system instead of browsing randomly

The point of an information source is not to hand you a business. It is to reveal one kind of evidence. Use several sources, record what you find, and look for convergence rather than a single exciting screenshot.

Where is money already changing hands?

Use TrustMRR^[1] to study businesses with payment-provider-verified revenue, Acquire.com^[2] to inspect online businesses offered for sale and the operating metrics buyers evaluate, and Indie Hackers^[3] for founder case studies, experiments, and discussions. Treat self-reported figures as leads to investigate, not audited proof.

Ask what the buyer receives, who pays, why free alternatives have not removed the demand, how customers are acquired, and how much human operation sits behind the visible product. The goal is not to clone a company. It is to learn what outcomes already earn money.

What do customers repeatedly dislike?

Search relevant communities on Reddit^[4] for language such as “we do this manually”, “is there an alternative”, and “why is this so difficult”. Use software categories and reviews on G2^[5] to compare recurring complaints across products. Mid-range reviews can be especially useful because they may combine continued use with specific dissatisfaction.

A complaint is not automatically a market. Record the customer context, frequency, current workaround, consequence, and whether the person controls a budget. Repetition matters more than intensity.

What work are businesses already buying?

Freelance marketplaces are databases of paid business problems. Upwork^[6] exposes demand across AI, automation, development, marketing, data, and support work, while the Fiverr services directory^[7] shows packaged services that buyers can already order. Look for repeated, bounded work with clear inputs and outputs, then ask whether AI could improve the delivery model.

Do not assume that automating 80 percent of a task creates a product. The final 20 percent may contain the judgment, liability, relationship, or exception handling that customers are actually paying for. Investigate the whole outcome.

Which products and behaviours are appearing repeatedly?

Use the daily launch feed and archive on Product Hunt^[8] and the category, usage, traffic, and regional views on Toolify^[9] to notice clusters. One launch is an anecdote. Repeated attempts around the same customer job may indicate demand, easy imitation, temporary fashion, or all three. Your next job is to distinguish them.

For direction, compare relevant search interest and regions in Google Trends^[10]. For developer behaviour, watch GitHub Trending^[11] for technologies attracting current community attention. Neither source measures willingness to pay. Use them to ask what supporting needs—security, deployment, integration, testing, monitoring, migration, or cost control—may appear if adoption continues.

Build the daily three-problem habit

External sources help you see markets, but your own work and conversations contain higher-context evidence. Record three problems each day. Do not write solutions yet.

Too early: “AI invoice management startup.”

Useful observation: “I spent 30 minutes finding an invoice, checking its payment status, and writing a follow-up email.”

The second note preserves the person, situation, workflow, time cost, and current behaviour. Three observations a day produce 1,095 observations in a year. Most will not become companies. That is fine. The habit trains you to notice evidence before naming a product.

Four questions every startup idea must answer

1. What specific problem changes?

“Help companies use AI” is not a useful problem definition. Name the customer, the painful moment, the current workflow, the consequence, and the better after-state. Specific pain is easier to verify and easier to sell.

2. Who is the first reachable customer?

“Everyone” hides the work. Start with a group narrow enough to understand: for example, Australian plumbing businesses with 3–15 employees or independent Shopify stores processing more than 500 orders per month. The first market does not need to be enormous. It needs to be identifiable and reachable.

3. How will the first 100 customers hear about it?

Find where the customer already spends attention and whom they already trust: search, communities, industry associations, software marketplaces, local directories, partners, creators, events, or direct outreach. Distribution is not a task to postpone until launch. It is part of the idea.

4. What proves they will pay?

Existing competitor subscriptions, staff time, freelance contracts, internal tools, deposits, pilots, and purchase commitments are stronger than polite enthusiasm. Ask for behaviour that costs the customer something—money, time, data, reputation, or workflow change.

Competition is often evidence, not a reason to stop

Competitors can show that customers recognise the problem, budgets exist, pricing has been attempted, and distribution channels are available. A market with no competition may contain an overlooked opportunity—or no meaningful demand.

Look for a specific advantage: a simpler workflow, a neglected niche, local requirements, better service, stronger integration, a different pricing model, lower operating cost, greater trust, or a complete outcome instead of another tool.

The strongest position is usually defined by the deliverable a customer values, not access to a fashionable model or framework.

Ask how AI changes the whole business, not one feature

A powerful question is not “what new AI app can I invent?” It is “what existing valuable business would look different if it were started today?” Study where customers already pay substantial amounts, then reconsider the research, service, onboarding, support, reporting, marketing, and product operations around that outcome.

AI leverage matters when it creates a better customer result or a defensible operating advantage. It does not rescue weak demand, remove professional responsibility, or make every workflow safe to automate. The human still defines the need, sets boundaries, judges the work, and owns the outcome.

That is why productive vibe coding begins with customer value before product. AI is the execution leverage. The customer outcome is the reason to build.

A repeatable startup idea discovery pipeline

1. Discover a market. Find categories, customer groups, and business models with visible activity.
2. Collect pain. Save exact complaints, workflows, consequences, and workarounds without jumping to a solution.
3. Verify spending. Find software subscriptions, salaries, contractor jobs, internal tools, or costly failure that put economic weight behind the pain.
4. Check direction. Compare several indicators and ask whether the need is durable, emerging, seasonal, regional, or fading.
5. Find a narrow gap. Identify the customer and moment current solutions serve poorly, not a generic list of missing features.
6. Verify distribution. Name a credible path to the first 10, 100, and 1,000 customers and the cost or effort each path may require.
7. Define AI leverage. Explain which parts AI can execute, which parts need human judgment, and why the resulting system is materially better.
8. Choose the riskiest assumption. Decide what must be true for the business to work and what evidence would prove you wrong.
9. Build the smallest credible test. Use a conversation, manual service, landing page, paid pilot, concierge workflow, or focused prototype to learn before committing to a full product.

In the AI era, the first product can be disposable. What matters is whether the minimum viable product produces useful evidence about demand, trust, repeated use, and payment.

Prefer costly behaviour over easy opinions

• Weak evidence: likes, upvotes, trend lists, compliments, survey intent, and generic “I would use this” responses.
• Useful evidence: repeated complaints, active workarounds, competitor use, customer interviews with specific recent behaviour, and willingness to share data or time.
• Strong evidence: existing spend, a deposit, signed pilot, procurement step, repeated use, referral, renewal, or measurable improvement in the promised outcome.

Let the evidence point to what deserves to be built

A startup idea does not need to arrive as a flash of inspiration. It can emerge from a repeatable practice: observe problems, trace money, inspect current solutions, check direction, find a reachable customer, and test the most dangerous assumption.

AI makes it easier to turn that evidence into an experiment. That increases the value of judgment; it does not replace it. Build only after you can say what you are trying to learn, from whom, and what result would change your mind.

Discover → verify pain → verify spending → check direction → find the gap → verify distribution → define AI leverage → build the smallest test