TL;DR An MVP is the point at which a digital product meets reality. It should come after research has identified a real problem and prototyping has tested a possible direction. When the live product still cannot retain users, attract paying customers, or survive without continuous founder effort, more iteration may only deepen the loss.
This article is about closing the product itself, not removing an unpopular feature. In many startups, that also means closing the company. Research on escalation of commitment explains why founders delay this decision.
Controlled experiments with more than 750 startups suggest that teams trained to treat ideas as hypotheses stop weak projects earlier and perform better. The practical defence is to define evidence and a deadline before launch, then review the result with someone who is not emotionally invested.
An MVP Comes After Research and Prototyping
In product work, we keep seeing the same mistake. A team starts with a solution, speaks to people who are easy to reach, hears encouraging comments and begins building. The founders say they know their customers already.
When the product struggles, they call the next round of development an experiment. By then, however, they are often using an expensive live product to answer questions that better research should have addressed much earlier.
The discipline behind good digital product development did not begin with startup culture. Human factors studies how people interact with tools and systems so those systems can be designed around human capabilities and limitations. Human-computer interaction brought that tradition into computing.
In 1985, John Gould and Clayton Lewis described three principles for usable systems: an early focus on users and their tasks, empirical measurement and iterative design. Those principles later became part of the broader human-centred design process reflected in ISO 9241-210. The Human Factors and Ergonomics Society and NIST both describe this work as applying knowledge about people to the design of systems.
The practical sequence is research, prototype and MVP, although real projects loop between them. Generative research investigates people, their context, the problem, the alternatives they already use and the differences between relevant customer segments.
A prototype is a low-cost representation used to test a proposed direction before the team commits to a production system. An MVP is the smallest live product that allows real customers to reveal, through behaviour, whether the product deserves to exist.
Startup methods later popularised the MVP and compressed product work into a build-measure-learn loop. The shorthand is useful, but it becomes dangerous when teams treat it as permission to skip generative research. An MVP should test the remaining business and product assumptions under real conditions. It should not carry every unanswered question about the customer, the problem and the solution into production.
Weak Research Produces Expensive MVPs
Poor research can make a product look validated before it has met the people whose behaviour matters. Teams recruit the wrong segment, ask whether an idea sounds useful, or mistake politeness for demand.
They study stated preferences but ignore current behaviour, switching costs and the alternatives people already tolerate. A polished prototype can then confirm that people understand the interface without proving that the underlying problem is important enough to change what they do.
This distinction matters because usability and demand are different questions. A product can be easy to use and still solve the wrong problem. It can also solve a real problem for a segment that cannot or will not support the business.
When research, prototype testing and an honest MVP all point in the same negative direction, the team must be willing to accept the result.
Closing the product means ending the product as a whole. It is not the routine removal of a feature or the retirement of one line inside a healthy platform. In an early-stage company whose identity and economics depend on a single product, killing the MVP may mean shutting down the company. That is precisely why founders find the decision so difficult.
Why Founders Keep Products Alive
Many startups formally die when the money runs out, but cash exhaustion is often the last event rather than the first problem. In an analysis of 431 venture-backed companies that shut down after 2023, CB Insights reported poor product-market fit, bad timing and unsustainable unit economics among the recurring underlying causes.
The median company closed 22 months after its last funding round, and nearly a quarter had remained in a zombie state for more than three years. CB Insights documents the delay between losing a credible path and admitting that the company is over.
Psychology has a name for the tendency to keep investing in a losing course of action: escalation of commitment. A review in the Academy of Management Annals found that positive expectations appeared consistently across very different contexts. Identity also matters because founders persist longer when the business has become part of who they are.
A study of 66 entrepreneurial teams across 569 decision rounds found that group hope pushed teams towards escalation more strongly than group fear pushed them towards termination. Close friendship within the founding team increased the risk.
The decision is rarely financial alone. Research on underperforming firms found that founders persist longer when they have fewer attractive alternatives, gain personal satisfaction from running the business or face high costs in changing careers. Loyalty to employees, creditors and family can turn persistence into a moral obligation, and those motives are understandable but they still do not create customer demand.
Shyp The on-demand shipping startup was once valued at $250 million. Its CEO, Kevin Gibbon, later acknowledged that the company kept pursuing consumers despite evidence that the strategy was wrong and maintained popular services that were not profitable.
Shyp finally focused on business customers and cut the rest. Its health improved within weeks, but the runway was already gone, and the company closed in 2018. The Shyp story shows how a late correction can be directionally right and still arrive too late.
Dopios Welcome Pickups founder Alexandros Trimis has described the early excitement around his first venture, Dopios. People praised the product, but the team had no convincing plan to turn that praise into revenue. The lesson is uncomfortable and common: approval is not adoption, and admiration is not a business model. Trimis discussed the experience with a Greek startup audience.
Stopping Is a Skill That Can Be Learned
The strongest evidence comes from randomized controlled trials led by Arnaldo Camuffo and colleagues. In a 2020 pilot involving 116 startups, founders were trained to treat their business ideas as hypotheses. They were less likely to continue projects that appeared promising but were not and more likely to terminate weak ones.
A 2024 replication involving 759 firms across four trials confirmed the effect on idea termination. It also found that the best results came from a small number of deliberate pivots rather than no pivots or constant changes of direction. Among the top quarter of revenue generators, trained startups earned about €28,000 more than the control group during the experiment.
A follow-up study with 382 entrepreneurs helps explain why. Trained founders became more accurate in judging the value of their projects and reduced their expectations earlier. Untrained founders eventually abandoned more projects, but they did so later and after spending more.
The trained group generated more new ideas and converted a higher share into ventures. Most founders eventually recognize a bad idea. The skill is recognizing it while enough time, money and energy remain to try something else.
Organizations can reinforce that behaviour. Astro Teller has described how X, Alphabet's moonshot factory, publicly recognizes teams that end their own projects and has awarded bonuses for doing so. The point is not to celebrate failure as theatre. It is to remove the incentive to conceal evidence until a project becomes impossible to defend.
MVP Kill Criteria
A founder should not invent the standard for success after seeing the result. Kill criteria work best when they are written before launch, tied to observable behaviour and paired with a date. No single metric can decide the fate of every digital product. Together, however, the following signals make it harder to rationalize a product that has lost its path.
Write the Criteria Before Launch
Itamar Simonson and Barry Staw compared methods for reducing commitment to losing decisions. People who defined minimum target levels in advance became more responsive to negative evidence. Annie Duke, author of “The Power of Knowing When to Walk Away”, calls these thresholds kill criteria: a measurable condition and a date, such as “if fewer than 20 of our first 100 users remain active after eight weeks, we stop.”
A pre-mortem can help the team find the right criteria. Imagine that the product has failed, then list the reasons. The original research on prospective hindsight found that treating the outcome as certain helped people generate about 30% more reasons. That does not make every reason correct. It exposes risks that optimism would otherwise keep vague.
Watch Whether Retention Flattens
Retention shows whether the product creates value repeatedly. A weak cohort curve keeps falling towards zero. A healthier curve drops as casual users leave and then begins to flatten. The exact level depends on the product, frequency of use and customer segment, so generic benchmarks should remain reference points rather than universal laws.
The shape of the curve matters because a product that continually loses every cohort has not yet found a durable reason to exist.
Chorus The social fitness app co-founded by former Twitter CEO Dick Costolo closed roughly eight months after launch because users generally left within four to eight weeks. The company still had money. Costolo chose not to hide churn behind a model that would profit from people paying for a service they no longer used. Reports on the shutdown make retention, rather than runway, the decisive evidence.
Remove Founder Effort From the Measurement
Early activity can be manufactured. Founders recruit friends, send reminders, create content, arrange introductions and personally rescue every stalled interaction.
That effort is sometimes necessary to start a market, but it can also conceal the absence of pull. Stop pushing for a defined period and observe what remains. If usage disappears as soon as the founders step back, the product may not have independent life.
WhereBerry After years working on products including YouTube, Chrome and Android, Nick Baum closed his events startup when he realized that users became active only when he created the activity himself.
He later used the contrast as a benchmark for StoryWorth, where customer behaviour felt different from the beginning. His account of the decision turns “organic usage” into something observable.
Count Costly Signals
Compliments, sign-ups and press coverage cost people very little, but payment, repeated use, referrals and the effort required to switch from an existing solution are stronger signals.
For SaaS, Jason Lemkin has proposed 10 paying customers with no prior connection to the founders as a meaningful early threshold. The number is a heuristic, not a law, but the principle travels: count behaviour that costs the customer something.
The Sean Ellis product-market fit survey can add context by asking active users how they would feel if the product disappeared. A result above 40% “very disappointed” has become a familiar benchmark, but it should not decide the case alone.
MeasuringU found no peer-reviewed validation of the threshold and calculated that a result of 40% from 50 respondents has a margin of error of about 13 percentage points. Use the survey to investigate why a product matters, not as a mechanical verdict.
Referly Danielle Morrill closed the original version of the Y Combinator startup about six months after Demo Day because its most important metric, revenue, was not growing at a promising rate. The company could have survived, but it would not become the business promised to investors. Her zombie startup test asked whether a company had ever achieved meaningful growth, even briefly, rather than whether it could remain technically alive.
Put a Clock on the Search
Product-market fit can take time. In a study of 24 successful B2B companies, the median period from a working product to the first feeling of fit was roughly 9 to 18 months. That sample contains only winners, so it cannot tell us how long failed companies should keep searching.
It does show why the deadline must reflect the product, buying cycle, regulatory environment and available capital. A team should set review dates in advance and define what evidence must exist by each one. Without a clock, a temporary search can become a permanent zombie state.
Ask Whether a Credible Future Remains
Sunk costs belong to the past. The decision concerns the future. Aaron Harris proposed four questions for founders considering shutdown:
- Do you still have concrete ideas for growth?
- Can you pursue that growth profitably?
- Do you want to run the company that would result?
- Do you want to do it with your current co-founders?
The useful comparison is not continuing versus quitting in the abstract. It is continuing this product versus a specific alternative use of the team's remaining time and capital.
An outside view helps because founders are the people least able to evaluate their own product dispassionately. Use sales trends, cohort behaviour, customer interviews and economic data.
Then ask a board member, mentor or advisor who agreed to the criteria before launch to assess the same evidence. Research on de-escalation found that judging people by the quality of their decision process rather than the eventual outcome reduced the urge to defend losing investments.
Product Closure Criteria at a Glance
| Signal | What to Measure | When It Points to Closure |
|---|---|---|
| Pre-set criteria | A measurable state and review date agreed before launch | The date arrives and the state is absent |
| Retention | Cohort retention over an appropriate period | Every cohort keeps declining towards zero |
| Independent usage | Activity when founder intervention is reduced | Usage disappears when the team stops pushing |
| Costly signals | Payment, repeat use, referrals and switching effort | Interest remains cheap and non-committal |
| Time | Evidence required at scheduled decision points | The search continues without new evidence or a credible test |
| Future path | Specific, affordable growth ideas and founder commitment | No credible path remains, even to a deliberate pivot |
| Outside view | Assessment by a neutral advisor who knows the criteria | An independent reviewer sees a zombie product |
When Not to Close the Product
Stopping too early is also a mistake. The Camuffo trials found that the strongest performers made a few considered pivots. Teams should reopen research when they discover that they recruited the wrong segment, misunderstood the context, tested a weak value proposition or built an MVP that customers could not realistically adopt. Negative evidence from a badly designed test is evidence about the test as well as the product.
A pivot must still be honest about what ended. If the team changes the customer, the problem and the value proposition, the previous product has effectively been killed even if the company and some technology survive.
Persistence deserves respect when each cycle produces new evidence and a more credible path. Repeating the same test while moving the success threshold does not.
A product also does not need venture-scale growth to justify its existence. A profitable company that customers value and founders want to run may be a good business.
The relevant question is whether the product can meet the goals and obligations that define its company, not whether it resembles a famous outlier. Stories such as Airbnb are memorable because persistence worked under exceptional conditions. They should not be used to make evidence optional.
How to Close a Product Well
A timely decision preserves choices. Aaron Harris recommends making the shutdown clearly and with enough money left to help employees find new work. Customers need notice, data export and a realistic transition plan. Investors need an honest account of the evidence and the remaining capital.
Founders who wait until the bank balance makes the decision lose the ability to treat those groups well.
The public explanation serves a different purpose from the private record. A study of 118 public closure statements found that entrepreneurs emphasized effort, achievement and lessons while rarely blaming external forces.
That is understandable reputation management. The internal post-mortem should be less polished. Record which assumptions failed, what the team observed, when the evidence became strong enough and why the decision still took as long as it did.
Research on founders in the disk-drive industry found that people who attributed an initial failure to their own decisions and changed their behaviour later built more successful careers. Those who blamed external factors repeated unsuccessful patterns.
Failure by itself teaches very little:
Harvard economists found that previously failed venture-backed founders succeeded at roughly the same rate as first-time founders, about 20% versus 18%, while previously successful founders reached 30%. Learning requires a change in judgment or behaviour.
In Greece, a 2022 diaNEOsis survey of 267 startup founders found that 22.5% had already experienced at least one failed business. The subject is not marginal. Founders need a language for ending products before the only possible explanation is that the cash disappeared.
The Decision Is About the Future
A product can be beautifully designed, technically sound and built by people who worked with intelligence and care. It can still be the wrong product. The MVP exists to expose that possibility before the company spends everything defending it.
The courage to close a product is not contempt for the work already done. It is the refusal to demand that customers validate a founder's identity, loyalty or sunk costs. When the research has been revisited, the tests have been fair and the evidence still says no, stopping is the final responsibility of the product team.
Selected Sources
- Gould and Lewis: Designing for Usability Key Principles and What Designers Think
- Human Factors and Ergonomics Society: What Is Human Factors and Ergonomics
- NIST: Human Centered Design
- ISO: ISO 9241 210 Human Centred Design for Interactive Systems
- CB Insights: The Top Reasons Startups Fail
- University of Delaware: Escalation of Commitment Research
- Bayes Business School: Friendship and Startup Escalation
- Strategic Management Journal: A Scientific Approach to Entrepreneurial Decision Making
- SDA Bocconi: The Science of Navigating Business Uncertainty
- Research Policy: Scientific Decision Making and Longer Term Outcomes
- Journal of Applied Psychology: Deescalation Strategies
- JCX Corporate Behavioural Science: The Premortem
- Sequoia Capital: Retention
- Lenny's Newsletter: Finding Product Market Fit in B2B
- MeasuringU: The Product Market Fit Item
- Y Combinator: Shutting Down
- Danielle Morrill: Zombie Startups
- Aalto University: How Entrepreneurs Present Business Closure
- Strategic Management Journal: Attribution and Entrepreneurial Learning
- Strategy and Business: Success Breeds Success in Startups
- diaNEOsis: Startups in Greece
