Decision
When to Continue, Pivot, or Kill a Startup Idea
The evidence is in. The interviews are done. The landing page has been live for a week. The question founders keep asking is the one the publication has not yet answered: based on what I have learned, should I keep going, change direction, or stop?
Last updated · September 7, 2026
Quick answer
How do you decide whether to continue, pivot, or kill an idea?
The decision rests on three questions, in this order: (1) Is the critical assumption still un-validated, or has the evidence actually tested it? (2) Has the evidence replaced a belief, or only confirmed one the founder already had? (3) If the evidence were presented by a stranger, would it point to continue, pivot, or kill? If the critical assumption is still un-validated, the work is more validation, not a decision. If the evidence has tested the critical assumption and the answer is no, the work is to stop — not to add more validation. If the evidence is mixed, the work is a pivot to a sharper version of the question, not a re-run of the same test.
If the founder had not already built the prototype, would the evidence persuade them to start today? If the answer is no, the work is either more validation or a stop. If the answer is yes, the work is execution.
Key takeaways
The decision in 30 seconds
- Four kinds of evidence count: interview pattern, behavior on a smoke test, payment or pre-order, and observed workarounds. Verbal interest is not one of them.
- The decision is 'continue, pivot, or kill'. The 3-question test is: was the critical assumption tested, did the evidence replace a belief, and would a stranger act on the same evidence?
- Saturation — the same complaint or behavior showing up five times across unrelated interviews — is the stop-rule for the customer-research phase.
- A 7-day pre-mortem — 'it is 18 months from now and the idea failed. Why?' — surfaces the failure modes you cannot see from the inside.
- Traps founders fall into: sunk-cost, friends-as-data, 'one more interview', and the redefinition of pivot. Each one is a way to keep building when the evidence has already said no.
What counts as evidence
The four signal types that count
Most of what founders call 'evidence' is not evidence. Four kinds of signal actually test the assumption. If the data you have is not one of these four, it does not change the decision.
Interview pattern — the same complaint from unrelated people
The same problem, described in similar words, by people who do not know each other. Three to five interviews is usually enough to see the pattern; more is unnecessary if the pattern is consistent. Rob Fitzpatrick's Mom Test (2013) is the standard reference for the interviewing rules that make the pattern legible.
Behavior on a smoke test — clicks, signups, deposits
A real action on a real surface — a click, a signup, a deposit, a pre-order. The action costs the person something, even if only a few seconds. Behavior on a smoke test is the cheapest way to test willingness to act before any code is written.
Payment or pre-order — money on the table
A deposit, a pre-order, a paid pilot. Payment is the strongest signal because it is the only one that costs the buyer something real. Verbal commitment to pay is not payment; only an actual transfer is.
Observed workaround — time or money already being spent
Time, money, or labor the buyer is already spending to work around the problem. A spreadsheet, a hired contractor, a manual process. The workaround is signal that the problem is real and painful; the cost of the workaround sets the ceiling on the willingness to pay.
The 3-question test
Continue, pivot, or kill — the decision test
Run the evidence through the three questions below, in order. The first question determines whether the evidence actually tests the critical assumption. The second determines whether the evidence has moved the founder off a prior belief. The third asks whether a stranger, given the same evidence, would act on it.
- 1
Was the critical assumption actually tested?
If the evidence does not address the critical assumption the plan most depends on, the work is to test the critical assumption, not to make a continue-or-kill call. A no-go on a non-critical assumption is a no-op for the decision.
- 2
Did the evidence replace a belief, or only confirm one?
If the founder already believed the answer before the test, and the test produced the same answer, the test is not yet informative. An informative test is one where the result could have gone either way, and went one way.
- 3
Would a stranger act on the same evidence?
If a founder who had not built the prototype were handed the evidence today, would they start, pivot, or walk away? This is the question that exposes sunk-cost reasoning. The honest answer usually points to the right call.
Continue — the evidence supports the plan
The critical assumption has been tested, the test was informative, and a stranger would start. The work is now execution, not more validation. Build the smallest version that produces the next piece of evidence and keep going.
Pivot — the evidence points to a sharper version of the question
The critical assumption has been tested, but the evidence is mixed or partial. The pivot is to a sharper version of the same question — a narrower persona, a different price point, a different distribution channel — not a re-run of the same test with the same inputs.
Kill — the evidence points to no-go
The critical assumption has been tested, the test was informative, and a stranger would not start. The work is to stop and write down what was learned, not to add more validation. Adding more validation after the evidence has said no is the most common decision trap.
Saturation
When to stop interviewing
The customer-research phase needs a stop-rule, the same way the smoke-test phase needs a stop-rule. The standard rule is saturation: the same signal starts showing up, in the same words, from people who do not know each other.
Five interviews, if the pattern is consistent
If the same complaint shows up in three to five unrelated interviews, in similar words, the pattern is real. More interviews will not change the answer. The work is to act on the pattern, not to keep asking.
Ten interviews is the standard ceiling
If the first ten interviews are inconsistent, the work is to understand why — a different persona, a different framing of the question, a different channel — not to interview the eleventh person. Inconsistency is a signal about the inputs, not about the number of interviews.
The 'new' interview is not adding signal
If the eleventh interview is producing the same pattern as the first five, the interview is not adding signal. Stop interviewing. Switch to the next experiment in the validation ladder — usually a smoke test or a concierge — and test the next assumption.
Saturation is a stop, not a yes
Saturation tells you the qualitative phase is done; it does not tell you the idea is good. A consistent pattern of 'this problem is not real' is also saturation, and it is also a stop. The pattern, not the number, is what counts.
Pre-mortem
The 7-day pre-mortem
A pre-mortem is a structured exercise: assume the idea failed eighteen months from now, and write down why. The exercise is cheap (a single afternoon), and it surfaces the failure modes you cannot see from the inside the plan.
- 1
Write the failure post-mortem first
Open a blank page. Title it: 'It is 18 months from now, and the idea failed. Here is why.' Write the post-mortem as if you were reading the analysis after the fact. The format is what matters: a written post-mortem is harder to fool yourself with than a verbal one.
- 2
List every plausible cause of failure
For each plausible cause, write one or two sentences. The cause can be a distribution failure (the channel did not work), a willingness-to-pay failure (the price was too high), a product failure (the MVP did not deliver), or a market failure (the problem turned out to be rare). Be specific.
- 3
Sort the causes by likelihood and impact
For each cause, write down a rough likelihood (low, medium, high) and the impact it would have on the business. The top three causes — by likelihood times impact — are the assumptions to test next, not the ones you already feel good about.
- 4
Identify the cause you cannot test
Some causes cannot be tested cheaply — a regulation that changes, a competitor who raises a large round, a technology shift. These are real risks; they are not stop signs. The pre-mortem is a way to see them in writing, not a way to eliminate them.
- 5
Decide which causes change the decision
Some causes, if they happen, would invalidate the plan. Others, if they happen, would be a setback but not an invalidation. The first set changes the decision; the second set is a risk to manage. Mark which is which.
Decision traps
Five traps that keep founders building after the evidence has said no
Each of these is a way to keep building when the evidence has already pointed to a different call. None of them are reasons to override the evidence; they are reasons to be more careful about how the evidence is read.
Sunk-cost reasoning
'I have already built the MVP, so I should keep going.' The decision is about the next six months, not the last six months. A prototype that has not produced signal is a sunk cost, not a reason to continue. The honest test is: would you start this today, knowing what you know now?
Friends and family as data
'My friends say the idea is great.' Friends and family will say the idea is great because they want to support you, not because they would use the product. The signal is from strangers who have no reason to be polite.
'One more interview'
The pattern is inconsistent, so the response is to interview more people in the hope that the eleventh person will change the answer. Inconsistency is a signal about the inputs, not about the number of interviews. The right response is to understand the inconsistency, not to add more data points.
Redefining pivot as 'another feature'
When the evidence says no, the response is to redefine the pivot as a small feature change — 'if I just add this, the conversion will be fine'. A feature change that does not address the critical assumption is not a pivot; it is the same plan with a different surface.
Conflating effort with progress
A busy founder is not necessarily a founder making progress. Long hours on the wrong plan are not a substitute for the right plan. The honest question is: based on what the evidence says, what should I be doing this week? If the answer is not what you are doing, the work to do is to change the work.
How Yibud helps
How Yibud connects to the decision
Yibud's tools do not make the continue-or-kill call for you — there is no algorithm that can. What they do is name the critical assumption, surface the per-dimension evidence, and keep the same plan connected to the same report so the call is grounded in what was learned, not in what was hoped for.
Use the Decision Guide
The Decision Guide walks through how to read the Startup MRI report, find the critical assumption, and pick the cheapest experiment that tests it. The Guide is the in-product version of the 3-question test above, with the steps adapted to the report's structure.
Open the Decision Guide →Compare two ideas side by side
When the decision is between two candidate ideas rather than continue-or-kill on one, the compare surface puts them side by side on the same dimensions. The side-by-side makes the call auditable and prevents the trap of picking whichever idea the founder has been thinking about longest.
Open comparison →FAQ
Frequently asked questions
How many customer interviews is enough?
If the pattern is consistent, three to five interviews from unrelated people is usually enough. If the pattern is inconsistent, ten is the standard ceiling — and the work after ten is to understand the inconsistency, not to interview more. The Mom Test (2013) describes the rules that make the pattern legible.
What if the evidence is mixed?
Mixed evidence usually means the assumption is partially tested, not untested. The right response is to test the part that is unclear, not to keep collecting more data on the parts that are already clear. The pivot is to a sharper version of the question, not a re-run of the same test.
What if my partner disagrees with the decision?
Write the evidence down. The disagreement is usually about which piece of evidence to weight most heavily, not about the evidence itself. A written list of the evidence — what was tested, what the result was, what it means — usually produces a faster agreement than a verbal argument.
What if I have already built the MVP?
Sunk cost is not a reason to continue. The honest test is: would you start this today, knowing what you know now? If yes, the work is execution. If no, the work is to stop and write down what was learned. A prototype that has not produced signal is a sunk cost, not a plan.
How do I know if a 'no' is really a no?
A 'no' is informative if (a) the critical assumption was actually tested, (b) the test could have gone the other way, and (c) the result is consistent with the other evidence you have. If all three are true, the 'no' is a real 'no'. If any one is missing, the work is to test again, not to override the answer.
Should I tell investors or cofounders about the no-go?
Yes. The honest framing is the most useful one. 'We tested X, it said no, here is what we learned' is the kind of update that builds trust. A no-go that becomes a new idea is the most valuable kind of pivot; the people who saw the decision get to see the next decision.
What if I have no evidence yet?
The work is to collect evidence, not to make a decision. The 3-question test is only useful after the critical assumption has been tested. Before that, the right output is the experiment that produces the next piece of evidence — usually a smoke test, an interview, or a concierge.
How does Yibud help with the decision?
Yibud does not make the continue-or-kill call — no tool can. What the tools do is name the critical assumption in the report, surface the per-dimension evidence, and connect the experiment workspace to the same plan so the call is grounded in what was learned. The Decision Guide walks through the same 3-question test, adapted to the report's structure.
See the critical assumption in your own idea
Generate a Startup MRI report. The report names the single critical assumption, the per-dimension evidence to collect, and the cheapest experiment to run first. Bring the evidence back to the same report and the Decision Guide walks through the call.
Analyze my idea →