Continue, pivot or kill: how to decide on a startup idea

Continue when every critical claim has evidence behind it that you would trust from a stranger, and a deliberate search for counter-evidence came back without a serious contradiction. Continue does not mean the idea is proven. It means nothing you have found yet is a reason to stop.

• The problem shows up in people's own words, and they already spend time or money on it.
• There are enough of those people, and a way to reach them.
• Competitors exist and leave a gap you can name, or do not exist for a reason you can explain.
• Nothing in law, regulation or a platform's policy blocks it.

A Continue that still has untested critical claims is a Continue with homework. Say which claims those are, and test them before you build much.

Pivot when the evidence supports the problem but contradicts one part of how you planned to solve it. The test of a real pivot is that the evidence itself points at the new direction. A pivot chosen because the old idea was uncomfortable to keep, with nothing suggesting the new one, is a fresh guess.

The part that changes is usually one of five:

• The customer. The problem is real, but for a different group than the one you named.
• The problem. The people are right, but what hurts them is next to what you assumed.
• The solution. They have the problem and will not adopt your way of solving it.
• The channel. They would buy, and you have no affordable way to reach them.
• The price or the model. They want it, at a price that does not cover serving them.

Change one part at a time. If you change the customer and the solution together, a better result next time will not tell you which change caused it.

Kill when a claim the idea cannot live without is contradicted by evidence you trust, and you cannot describe a version of the idea that works without that claim. The strongest reasons are structural:

• The people you named do not have the problem, or have it so rarely that they will not act on it.
• Winning and serving a customer costs more than they will ever pay, with no path to changing that.
• A law, a regulation or a platform's policy forbids the thing the product has to do.
• The idea depends on something a single other company can take away.
• Several well-funded attempts have failed for a reason that still holds.

A Kill reached in a week is the cheapest result validation can give you. The expensive version of the same answer arrives after the product is built.

Weak evidence means nobody has published what you need to know. Bad news means somebody has, and it goes against you. They call for opposite responses, and mixing them up is the most common way to kill a good idea or keep a bad one.

• Untested: nothing was found either way. Run an experiment.
• Mixed: credible findings on both sides. Find out which side describes your customer.
• Contradicted: credible findings against, none for. This is the one that can justify a Kill.

The same applies to scores. A low confidence figure on a lens says the sources are thin, not that the idea is poor. New ideas often have little written about them, and that is a reason to test, not to stop.

1. Set the bar before the evidence arrives. Write down what result would make you stop. A threshold chosen afterwards moves to wherever the results landed.
2. Ignore what you have already spent. Months of work are a reason you want the idea to be good, not evidence that it is.
3. Weigh the source, not the sentence. A confident claim on a vendor's blog is worth less than a dull table from a statistics office.
4. Read the case against first. Start with the red flags and the contradicting findings, then read what supports you.
5. Say the decision out loud to someone who gains nothing from your answer, and listen for the part you hurry through.

Buteo, the AI idea validator, recommends one of the three and shows the evidence behind the recommendation. It never makes the decision. A validation pauses at a dedicated step and waits for a person to choose, and an idea is never killed automatically. If the reply at that step is unclear, it is read as Continue, never as Kill.

• Continue sends the idea back through research, aimed at the claims that are still untested or mixed.
• Pivot restarts from your new direction and keeps the evidence already gathered, because it still describes the same problem space.
• Kill closes the validation with the reasons written down.

Continue and Pivot each start another research pass, at 8 credits. Kill costs nothing.

You can decide against the recommendation. Buteo sees the published evidence; you know your runway, your skills and what your customers told you in person. How the recommendation is produced is on the methodology page.

An illustrative case, not a real validation: a tool that sends payment reminders for freelance designers.

• Problem: supported. Late payment comes up repeatedly in freelancers' own discussions, and they already pay for tools around it.
• Competition: mixed. The space is crowded for agencies and thin for solo designers.
• Financial: untested. Nobody has published what a solo designer will pay for this.

The call is Continue, with one condition: run a pricing test before building. The untested claim is critical, so it is the next thing to settle, and a fake door or a concierge test would settle it. If the same research had found that solo designers are rarely paid late, the call would be Pivot, towards the agencies who are.