A page of startup ideas creates a peculiar kind of confidence. Every row can carry a market estimate, an AI feature, and a plausible buyer. The sheet looks analytical, yet the founder is still stuck. One idea has obvious pain but difficult access. Another is easy to demo but hard to sell. A third has a large market and no credible first transaction. More scoring has not created a choice.
Choose the idea with the strongest combination of painful demand, reachable users, founder access, testability, and a believable first transaction. Those dimensions bring the comparison close to what can actually be learned and sold. The goal is not to identify the largest possible company from a spreadsheet. It is to select the opportunity that deserves the next week of evidence.
Market-size scoring is insufficient
Market size matters after a product has a credible path into the market. Before that, a large category can conceal almost every practical weakness. It does not show whether a specific workflow hurts, whether the founder can reach the people living with it, whether a small test can challenge the assumption, or whether anyone can authorize a purchase.
Early estimates are also unusually sensitive to definitions. A founder can make a market appear enormous by counting everyone adjacent to a problem, then make it look focused by describing a narrow initial segment. Neither move explains why the first ten conversations will happen. Market size is a boundary condition, not a substitute for an entry strategy.
It is tempting to choose the grandest idea because ambition feels like commitment. The opposite mistake is choosing the easiest demo because visible progress feels like evidence. Both avoid the uncomfortable middle: a painful problem that can be reached, tested, and transacted around before the full product exists.
The best idea to pursue now is not the one with the biggest story. It is the one that can earn stronger evidence without pretending the hard parts away.
Compare five dimensions that affect the next decision
Painful demand asks what people already lose when the problem remains unsolved. Look for delayed revenue, repeated expert labor, avoidable risk, broken service, or a workaround the team protects. Interest in an outcome is not enough. The idea becomes stronger when the cost appears often, has a clear consequence, and belongs to someone motivated to change it.
Reachable users asks whether the relevant people can be found and engaged without first building a mass audience or signing a distribution partnership. A modest market with concentrated communities, obvious job titles, or warm paths may be more useful than a huge population reachable only through expensive broad marketing. Reachability is part of the product hypothesis because learning depends on contact.
Founder access goes deeper than a list of prospects. It includes earned context, trust, vocabulary, and permission to observe the workflow. A founder who understands the constraints can ask better questions and spot false positives sooner. Access does not require lifelong industry experience, but there must be a believable path to acquiring context without performing expertise that is not there.
Testability asks whether the riskiest assumption can be challenged before building the complete system. Can the workflow be delivered manually? Can a user bring a real case? Can a prototype operate on representative data with responsible boundaries? Ideas that require a marketplace, proprietary dataset, regulatory approval, and deep integration before any signal appears deserve a steep discount at the choosing stage.
A believable first transaction asks who pays, what they receive, and why the exchange can happen early. The transaction may be a paid pilot, a scoped service, or a small software purchase. It should deliver an outcome narrow enough to promise honestly. “Enterprises will subscribe after the platform is complete” is not a first transaction. It is a postponed assumption.
Score each dimension from one to five and write one sentence of evidence beside every score. A five requires observed behavior or direct access, not confidence. A one should name the missing condition. Multiply scores by the weights, but keep the evidence visible. The calculation disciplines comparison; it does not turn uncertainty into fact.
The suggested weights favor demand and transaction because a young company cannot survive on technical plausibility alone. They are not universal constants. A founder with several credible opportunities might increase the weight on founder access. A team facing a short runway might increase testability. Change weights to reflect a real constraint, never to rescue a favorite idea after seeing the result.
Eliminate before ranking
Some ideas should not receive a weighted score yet. Eliminate an idea if no specific person owns the cost, if the team cannot reach users without solving distribution first, or if the first transaction depends on a complete platform. Park an idea if it requires reliability the current approach cannot responsibly demonstrate. Reject any test that would require inappropriate access to sensitive data or would place unreviewed output into a high-consequence action.
An elimination rule protects the comparison from compensating errors. Severe pain should not cancel out an impossible path to users. Easy prototyping should not cancel out the absence of a buyer. Founder enthusiasm should not cancel out a safety boundary. If a missing condition can be changed, write the prerequisite and revisit the idea when it becomes true.
A worked three-idea comparison
Consider a deliberately illustrative portfolio, not a claim about real market performance. Idea A prepares evidence for marketplace payment disputes. Idea B produces general meeting summaries. Idea C reviews technical documentation for regulated product teams. Assume the founder can directly reach online sellers, has ordinary access to team-software users, and has no trusted network or domain background in regulated product development.
Idea A receives scores of five for painful demand, four for reachable users, four for founder access, five for testability, and five for a first transaction. The weighted result is 4.65. The reasoning is more important than the number: disputed revenue creates consequence, real evidence packets can be tested manually, and a seller can purchase help on a bounded case.
Idea B receives two for painful demand, five for reachability, four for founder access, five for testability, and two for a first transaction. The weighted result is 3.35. It is easy to find users and demonstrate the feature, but the current alternatives are abundant and the cost of leaving the problem unsolved is often modest. Convenience does not automatically create a durable purchase.
Idea C receives five for painful demand, two for reachability, one for founder access, two for testability, and four for a possible transaction. The weighted result is 3.15. The workflow may be valuable, but this founder cannot yet inspect it responsibly or recruit the necessary expertise. The idea is not declared bad. It is parked until access and a safe test become credible.
Idea A wins this round because demand, access, testability, and transaction reinforce one another. The conclusion is not that payment disputes form the best AI market. The example only shows how present founder conditions change the choice. A different founder with trusted regulatory relationships could rank Idea C very differently without either analysis being dishonest.
Run a sensitivity check before committing
A ranking is fragile when a small, reasonable weight change produces a different winner. Recalculate after increasing the weight of the constraint most likely to determine survival: perhaps distribution, time to evidence, or transaction. Then lower any score based mainly on assumption rather than observed behavior. If the leading idea changes repeatedly, the next move is not a confident selection. It is a test that separates the tied ideas.
Also inspect dependency risk. An idea can score well while depending on model accuracy, data access, a platform policy, or a long procurement path that has not been tested. Place that dependency beside the score and ask whether one week of work can reduce it. Weighted totals should concentrate attention on uncertainty, not hide it behind decimals.
Give two finalists seven days to separate themselves
When the sensitivity check leaves two plausible finalists, day one is for naming the exact assumption that changes their order. Perhaps one idea leads only because reachability is scored four instead of three. Perhaps the other wins when the first-transaction weight rises. Keep both finalists and turn that ranking flip into one comparison question. On day two, prepare matched evidence requests with the same effort ceiling and a predeclared result that would justify changing the disputed score.
Use days three through five to run the two evidence tracks side by side. If painful demand is disputed, ask each audience about a recent consequence before describing a solution. If reachability is disputed, record what it takes to reach the actual user and budget owner through the channel that idea would depend on. If founder access is disputed, test whether the conversations reveal constraints the team can interpret without borrowing authority it does not have. If testability is disputed, require a falsifiable test design for each idea. If the first transaction is disputed, put a narrow outcome, scope, and price in front of the relevant buyer and ask for a decision.
Keep the comparison fair without making it artificial. Give each finalist the same attention budget, but use the real distribution path each would need; a warm network for one and cold outreach for another is founder access evidence, not noise to erase. Do not build two prototypes when the uncertain dimension is reachability. Do not count an easy meeting as demand when the uncertain dimension is transaction. The week should isolate the ranking disagreement instead of generating a pile of incomparable activity.
On day six, replace only the disputed scores with what the week revealed and calculate the ranking again. On day seven, write the strongest alternative explanation for the result, then rerun the original weights and the sensitivity weights. Choose the leader if it remains ahead under both reasonable views. If the order still flips, name the next cheapest separating test or admit that neither idea has earned priority. The output is a defensible selection, not simply more evidence for the original favorite.
Multiple ideas do not need to be resolved by instinct or by a false promise of mathematical certainty. Eliminate the ones that cannot yet be reached or tested. Weight the conditions that matter now. Keep assumptions visible. Then make the finalists compete on the disputed fact instead of defending them with broader stories. The page of possibilities has done its job when it produces one honest next action.