How to Evaluate a Startup Idea With Evidence

How to Evaluate a Startup Idea With Evidence

No checklist proves that a startup idea is good. The strongest early evidence usually concerns real behavior: a specific user repeatedly experiences a consequential problem, already spends time or money on an alternative, can be reached, and makes a commitment stronger than praise.

Evaluate the idea by separating facts from assumptions, ranking the assumptions that could kill it, and running the smallest ethical test that could disprove the weakest critical claim.

Write the idea as a change in behavior

Begin with one sentence:

A specific user in a specific situation currently does this because of this problem. We believe this change will create this observable value for the user or buyer.

"An AI platform for insurance" does not identify a user, workflow, consequence, buyer, or value. "A commercial property underwriter spends two hours gathering missing exposure information before a renewal decision; we believe a verified intake workflow can reduce that work without removing underwriting judgment" creates testable claims.

The sentence can still be wrong. Its value is that evidence can reach it.

Build a claim ledger

Write the current evidence, source, date, quality, contradiction, owner, and next test beside every critical claim.

Distinguish opinion from behavior

A user can sincerely praise an idea and never adopt it.

Stronger signals require proportionate effort or risk. A user may give access to a workflow, share authorized data, introduce the actual buyer, schedule a procurement review, accept a manual service, sign a test agreement, pay, use the product repeatedly, retain it, or refer another user.

Each behavior supports a different claim. A paid pilot can support willingness to pay while revealing that the service costs more to deliver than the customer will sustain.

Negative evidence belongs in the ledger. A declined introduction, repeated no-show, unsupported integration, or unacceptable data right may be more decision-relevant than ten compliments.

Use market research as context

The U.S. Small Business Administration describes market research and competitive analysis through demand, market size, location, saturation, pricing, barriers, competitors, and direct customer research.

The Census Business Builder supplies selected U.S. demographic and economic data for market and location research.

These sources can bound a market. They do not prove that the intended buyer has the problem, that the proposed product will be adopted, or that the founder can reach the buyer.

Check every dataset's geography, period, definition, population, suppression, sampling, revisions, and relationship to the actual segment.

flowchart TD
    A["Idea narrative"] --> B["Claim ledger"]
    B --> C["Rank fatality and uncertainty"]
    C --> D["Choose weakest critical assumption"]
    D --> E["Run smallest ethical disconfirming test"]
    E --> F["Record behavior and negative evidence"]
    F --> G{"Decision"}
    G -->|"Support grows"| H["Continue or narrow"]
    G -->|"Evidence conflicts"| I["Revise, change, pause, or stop"]

Treat familiar heuristics as questions

YC's official Startup School video index includes Jared Friedman's talk on getting and evaluating startup ideas.

Prompts about boring markets, competition, timing, hard starts, personal interest, scalability, and product-founder fit can widen investigation. They should not become score points.

A boring market may have real spending and neglected users, or it may have low willingness to change. Competition may confirm demand, or it may expose a distribution barrier. A recent enabling change may improve delivery, or it may be irrelevant to the buying process.

Convert each attractive heuristic into a claim that could fail.

Rank assumptions by fatality and uncertainty

An assumption is fatal when the idea cannot work if it is false. It is uncertain when current evidence is weak.

A color choice is rarely fatal. Legal access to the required data may be. A broad market category may be large, but the reachable buyer may be too narrow. Users may feel pain while the budget owner benefits from the current arrangement.

Test the assumptions that are both fatal and uncertain before polishing the lower-risk parts of the product.

Choose a proportionate test

Use source research for rules, market definitions, and existing evidence. Use interviews and observation for workflow, consequence, alternatives, and language. Use a prototype for comprehension or feasibility. Use a concierge service for delivery learning. Use a commitment request for demand. Use a paid pilot for authorized real use and economics.

Do not use one method for every question. A landing page cannot establish whether a regulated workflow is lawful. A prototype cannot establish that a buyer controls budget. An interview cannot establish retention.

Money, private data, regulated services, vulnerable users, employment, health, safety, or binding commitments require appropriate consent, authority, security, privacy, legal, and domain review.

Record the decision

At the end of a test, record what happened before explaining why.

Then choose continue, narrow, revise, change the buyer, change delivery, pause, or stop. State which evidence changed and which critical claim remains weakest.

The goal is not to protect the idea from a no. It is to earn the next investment of time and money or avoid it.

This page was developed with AI assistance from the E056 transcript and linked primary sources, then structured for human market, product, domain, data, legal, ethics, and editorial review. It does not validate a venture, market, customer, investment, or regulated activity.