How to Find (and Vet) Startup Ideas

If you’re hunting for a startup idea, you don’t need a lightning bolt. You need a rigorous way to notice real problems, avoid the traps, and evaluate candidates with clear eyes. This blog distills a full talk into a practical playbook: the most common mistakes founders make, a 10-question framework to judge any idea, and concrete recipes for generating better ones. It’s written for teams who want to stack the deck toward ideas that customers truly care about and that can grow into durable businesses.

Part I: The Mistakes That Sink Good Founders

Great execution can’t rescue a fundamentally weak idea. Start by avoiding these four traps.

1) Solutions in search of a problem (SISP)

The classic move: “AI is cool; what can we bolt it onto?” That path almost guarantees you’ll invent a superficially plausible problem that users don’t really care about. Customers aren’t moved by your technology; they’re moved by pain. Flip the order. Fall in love with a concrete, specific problem before you pick your tool.

How to self-check

• Can you describe the user’s pain in their own words, without mentioning your tech?

• When you talk to target users, do they try to pull your solution from you, or do you have to push it on them?

2) Tar-pit ideas

Some ideas look obvious, universal, and easily solvable, yet thousands of smart people have already tried them and stalled. These are tar pits. The problem is real, but there’s a structural reason it keeps defeating founders (network effects you can’t ignite, chicken-and-egg dynamics, brutal retention math, etc.). You can pursue one, but do it with eyes open.

How to approach a tar pit

• Search thoroughly for prior attempts. Talk to past founders if you can.

• Identify the specific hard part others hit. What is your non-hand-wavy plan for that exact barrier?

• Set aggressive falsification milestones. If you can’t bend the hard part early, move on.

3) Over-indexing on the first idea vs. waiting for perfection

At one extreme, founders lock onto the first shiny concept without pressure-testing it. At the other, they stall for months waiting for the mythical “perfect idea.” Neither exists. Aim for a good starting point with enough merit to learn quickly and pivot intelligently if needed.

4) Choosing ideas too abstract or grandiose

“Global poverty” is a problem; it’s not a startup brief. Big issues need tractable wedges — concrete, narrow entry points where a small team can deliver value, measure impact, and expand.

Part II: A 10-Question Framework to Judge Any Idea

Use this checklist as a pre-mortem. If an idea survives these questions, it’s worth serious exploration.

1. Founder–market fit - Are you unusually qualified to build this? Think mix of domain knowledge, distribution access, credibility, and technical ability. The best ideas are not abstractly “good”; they’re good for your team.

2. Market size (now or soon) -You want a path to a billion-dollar outcome. That can be a large current market or a small but rapidly compounding one where a clear adoption curve (regulatory shift, platform change, new behavior) makes growth plausible.

3. Problem acuteness - How painful is the status quo? Best-in-class signals: users currently do nothing because options are unusable; they hack together ugly workarounds; or they’re actively begging for relief. Weak signal: “nice to have if it were free.”

4. Competitive reality - Competition is normal in good markets. The key is identifying a non-obvious wedge: underserved segments, a step-function UX change, new distribution, a different buyer, a workflow you can automate end-to-end, or a compliance/security posture others lack.

5. Direct personal demand - Do you (or people you personally know) want this badly enough to use and pay for it? If not, why? Beware distance from the user.

6. Recent change as catalyst - Great startups often arise because something just changed: new tech, regulation, distribution channel, cost curve, or a shock (e.g., remote work) that reshapes demand. Name the change. Tie your wedge to it.

7. Proxy validation - Is there a successful analog in a different geography, adjacent vertical, or neighboring persona? A credible proxy lowers idea risk, though it doesn’t guarantee execution.

8. Team stamina fit - Is this a business you could work on for years? Boring spaces can be goldmines, but you must be willing to grind through the unglamorous parts. Passion often follows traction; still, be honest about your appetite.

9. Scalability of the delivery model - Pure software scales. Service-heavy models can work but require clarity on margins, repeatability, and productization. If humans must be in the loop, design toward increasing automation over time.

10. Idea-space quality - Zoom out one level. Some “idea spaces” (clusters of related problems and buyers) are more fertile than others. A fertile space means multiple adjacent pivots if your first concept misses. Pick an ecosystem with many nearby shots on goal and users you can repeatedly interview.

Part III: Three Traits That Make Ideas Look “Bad” But Are Actually Good

Savvy founders don’t just chase shiny. They pick ideas others avoid for the wrong reasons.

1. Hard to start - If getting started requires tedious integrations, regulated partnerships, gnarly domain knowledge, or long slogs with gatekeepers, most founders look away. That friction is a moat. If you can stomach the schlep, you thin the field dramatically.

2. Boring domain - Unsexy categories (compliance, payroll-like processes, procurement, taxes, logistics ops) repel many builders, yet the pain is undeniable and buyers pay. Day-to-day, your work will still be code, calls, and iterations — not a never-ending party. Boring often equals bankable.

3. Existing competitors - An empty market is usually empty for a reason. A crowded market can be a green flag if adoption remains low despite many products — that suggests unsolved fundamentals. The opportunity is a step change (e.g., embed in the OS/workflow instead of a web upload screen; automate the actual job, not just provide a dashboard).

Part IV: Seven Recipes to Generate Better Ideas

You can wait for ideas to occur organically (the highest hit rate), or you can systematically generate them. If you’re ideating now, start with these — ordered from most to least likely to produce quality.

Recipe 1: Start with your team’s superpowers

List your unique assets: domains you’ve worked in, permissioned data you can access, buyer relationships, technical depth, and hard-earned insights. Brainstorm only within those circles. You’re hacking founder–market fit at the ideation step.

Exercise For each founder: list every job, internship, research area, side project, and community you’ve been embedded in. For each:

• What chronic problems did people grumble about?

• What workarounds did they create?

• What truths do you know that outsiders don’t?

Synthesize overlaps across the team.

Recipe 2: Problems you’ve personally felt (especially non-obvious ones)

The best vantage points are weird intersections you occupy — roles, industries, or geographies where engineers rarely sit. If practitioners hate a task but rarely start companies, that opportunity can sit untouched for years.

Tell-tale signs

• Antiquated processes (fax, phone, spreadsheets) in mission-critical workflows

• Fragmented vendor landscape with unhappy customers

• Buyers who say, “If someone built x, we’d switch tomorrow”

Recipe 3: “I wish this existed”

Make a list of tools you want but can’t find. Then interrogate why they don’t exist. Sometimes the gap is a tar-pit; other times it’s a timing or distribution failure you can overcome.

Guardrails

• Identify the structural blocker (supply, regulation, unit economics, cold start)

• Draft a concrete plan to neutralize it (bundling, manual bootstrap, wedge segment)

Recipe 4: Map recent changes

Catalog shifts in technology, regulation, platforms, or behavior. Each change creates asymmetries. Ask: “Given this new reality, what becomes possible or necessary for [specific persona] that wasn’t before?”

Examples of change vectors

• New foundation models or cheap inference

• Data portability mandates or new reporting rules

• Hardware/OS capabilities (on-device AI, secure enclaves)

• Behavior shocks (remote/hybrid, split-shift work, BYO-AI)

Recipe 5: Variant of a proven pattern

Find a successful model elsewhere and adapt it to a new geography, segment, or adjacent workflow. The key is non-trivial localization: different buyer incentives, compliance, languages, integrations, or offline steps that incumbents won’t prioritize.

Recipe 6: Talk to people inside a fertile space

Pick a promising idea space, then over-interview. If you’re young or domain-light, this is the equalizer.

How to run it

• Define a narrow user: “fleet managers with 20–100 vehicles,” “multi-site dental office ops,” “clinical trial coordinators at mid-size CROs.”

• Schedule 30–50 calls. Ask about their last 3 painful weeks, not hypotheticals.

• Shadow workflows. Take screenshots (with permission). Time tasks.

• Speak with founders who tried and failed in the space. Learn the “hard parts.”

• Iterate fast: propose small automations and ask what would break.

Recipe 7: Hunt for broken big industries

Scan large, regulated, or operationally heavy categories that still run on paper, email, and phone trees. There’s often low-hanging automation with clear ROI. Be ready for longer sales cycles and heavier trust requirements; that’s part of the moat.

Bonus hack: If you’re missing a cofounder and an idea, join a cofounder network and filter for people with domain expertise plus early traction signals. Sometimes the best wedge is joining the right nucleus.

Part V: How to Pressure-Test an Idea Fast

Once you have a candidate, move from theory to evidence in days, not months.

1. Formulate your falsifiable thesis - “In 2 weeks, we will find 10 target users who each commit to a paid pilot for [outcome] because [pain] is costing them [quantified cost].” If you can’t write that sentence, the idea is still vapor.

2. Run 15–20 discovery calls - Ask about the last time they did the job: what triggered it, how they solved it, who was involved, what broke, what it cost. Collect artifacts: spreadsheets, emails, PDFs, screenshots.

3. Prototype the step-function - Don’t build an app; build the moment that feels 10x better (e.g., “drop a contract; get a clean, verified summary with flagged exceptions and suggested clauses”). Use duct tape, scripts, even manual work behind the scenes.

4. Charge early - A small, clear price filters compliments from commitment. Pilots with no money tend to mislead.

5. Instrument with ruthless clarity - Measure time saved, error reduction, throughput, and user touchpoints avoided. Show “before/after” with numbers. Buyers use this to justify renewal and expansion.

6. Observe usage gravity - Where does the work naturally happen (email, file systems, spreadsheets, chat, tickets)? Meet users there first. Add your own UI later if it earns daily attention.

7. Document risk, controls, and logs - Trust is product. Even in a pilot, show permissions, scopes, redaction, audit trails, and rollback paths. You’ll stand apart from toy demos.

Part VI: Pricing and Go-to-Market for Agent-Shaped Products

Agents change both what you sell and how you charge.

• Sell throughput/outcomes, not seats. Tie price to unit economics your buyer already tracks (contracts reviewed, claims processed, SKUs enriched, tickets resolved).

• Hybrid model. Platform fee + committed usage tiers. This de-risks seasonality while aligning price to value.

• Put ROI in the proposal. Convert your metrics into dollars: time saved × hourly fully loaded rate; error reduction × downstream cost; revenue lift × historical close rates.

• Land narrow, expand adjacent. Start with a painful slice you can automate end-to-end, then radiate into neighboring workflows once you’re embedded.

• Design the human-in-the-loop. Autonomy is earned. Start with draft/review/apply. Introduce confidence thresholds where the agent acts automatically and logs the action.

Part VII: Signs You’re On the Right (or Wrong) Track

Green lights

• Users volunteer to introduce you to peers or their boss.

• They give you real data, not sample brochures.

• They chase you for the next build.

• They ask for procurement paperwork unprompted.

Red flags

• “This is neat” with no next steps.

• Pilots stuck in “evaluation” after 60–90 days without an owner.

• Usage only by the champion, not the broader team.

• You’re spending all your time educating the market rather than solving a hair-on-fire problem.

Part VIII: Boring Beats Flashy (Most of the Time)

Fun, consumer-adjacent ideas get piled onto quickly; boring enterprise workflows languish for years. When you’re 6–12 months into any startup, the day-to-day looks similar: code, debugging, customer calls, ops triage. Initial “fun” has almost no correlation with actual founder happiness. Progress does. Boring + progress is more satisfying than flashy + stagnation.

Part IX: The Mindset to Keep You Moving

• Be hypothesis-driven, not dogmatic. Ideas morph. Keep the problem constant and let the solution evolve.

• Bias to shipping. Debating quality from the bleachers is a stall tactic. Launch a thin slice, face reality.

• Use idea spaces as safety nets. If your first swing misses, pivot adjacent without losing momentum.

• Respect the hard parts. Write them down explicitly. If your plan for them is “we’ll figure it out later,” that’s a warning.

• Treat trust as a feature. Especially with AI-infused products, governance is not a compliance afterthought — it’s table stakes for adoption.

A Final Word: When in Doubt, Launch and Learn

Even with a solid framework, it’s often impossible to know whether an idea is truly good without putting it in front of users. If you’re on the fence, choose a falsifiable milestone, ship a scrappy version that demonstrates the 10x moment, charge something, and see who leans forward.

Most of the ideas that become great companies didn’t start that way. They started as good beginnings pursued by teams who listened carefully, learned quickly, and moved where the signal was strongest. Your job isn’t to predict perfectly. It’s to start smart, test hard, and keep going.