An MVP is not merely the smallest thing you can ship. It is a version that produces useful learning about an important hypothesis with the least reasonable effort, without ignoring essential obligations.
The term Minimum Viable Product (MVP) is associated with Eric Ries and validated learning: the goal is not a short list of features but observable behavior that informs the next decision [R01]. Thus, an app with three buttons is not automatically an MVP; a limited manually operated service might be one if it delivers value and tests the right question. Definitions vary across teams. This article uses MVP to mean a product version or experience delivered under explicit conditions to learn about customers and value, rather than merely a technical demonstration.
1. Distinguish a prototype, proof of concept, MVP, pilot, and product
| Form | Main question | What it does not prove on its own |
|---|---|---|
| Concept or sketch | Is the idea understood? | Real use or willingness to pay |
| Prototype | How does part of the idea work or feel? | Complete delivery or sustained viability |
| Technical proof of concept | Is a critical mechanism feasible? | Demand or acceptable experience |
| MVP | Do people obtain and recognize enough value for us to learn from their behavior? | Scaling, profitability, or definitive product–market fit |
| Pilot | How does the solution behave in a controlled setting and scale? | Identical performance in other contexts |
| First operational version | Can we deliver within defined commitments and controls? | No need for future improvements |
| Mature product | Does it reliably meet familiar needs in sustained operations? | That it will never change |
These labels are not mandatory steps every team follows in the same order. A pilot can serve as an MVP; a proof of concept may precede a prototype; an existing service may evolve without a separate “new MVP.” Pillar 3 already covers prototype representations and fidelity. Here we focus on scope and learning from a version people actually use.
If you still need a representation to explore the solution before delivery, see prototype types and their limitations. Prototype fidelity and MVP scope are different decisions.
2. Start with a pending decision
Do not begin with “Which features can we cut?” Begin with the decision: “Should we invest in automated confirmations?” “Does a single-channel booking service help?” “Can we fulfill a controlled production batch without critical defects?”
Translate the decision into a specific hypothesis: which people, in which context, what observable behavior, and over what period. Define what would support, weaken, or leave the hypothesis unresolved. The Pillar 3 guides on experimental design can help with recruitment, measures, and methodological limitations. The MVP creates evidence for the decision; it does not replace an experimental protocol.
If a positive result would change no decision or commitment, building may be premature. If a simpler method answers the question with sufficient evidence, an entire product may be an unnecessary expense.
3. Define core value and the smallest complete flow
Start with what the person needs to accomplish: “Make a booking and know it is confirmed,” not “Have a registration screen.” Map the flow from beginning to end: entry, action, confirmation, fulfillment, and recovery from failure. Leave out peripheral features, but do not make the experience a demo that abandons the user halfway through.
A booking MVP might need time selection, availability checking, recorded response, confirmation, and support when something fails. It could postpone advanced reports, multi-location management, and extensive customization. Manual operations can complete invisible parts of the experience if described truthfully and shown to be sustainable for the test.
For physical products, a complete flow includes appropriate safe use, instructions, delivery, and defect handling. For services, it includes real assistance, exceptions, and availability limits. “Complete” here means sufficient for the defined outcome, not every feature on a distant roadmap.
4. Separate the minimum, the optional, and the mandatory
Classify each element as necessary for core value, necessary for safety, quality, or obligations, necessary to run the test, useful but deferrable, or requiring more research. This is a working classification, not a regulatory taxonomy.
| Hypothetical element | Decision | Reason |
|---|---|---|
| Clear booking confirmation | Include | Without it, the outcome is incomplete |
| Access controls for personal data | Include | Unrelated customer information must be protected |
| Failure recovery and support contact | Include minimally | Users need a way forward when something fails |
| Advanced statistics | Defer | Does not test the central hypothesis |
| Integrations with ten platforms | Investigate | Adds cost without clear learning benefit |
The Pillar 4 MoSCoW framework can help agree scope, but an essential requirement does not become optional because funding is limited. If the trial involves regulated activity, payments, or sensitive information, investigate applicable requirements first. Do not market a prototype as a ready service when you cannot fulfill the promises.
5. Bound who, where, how much, and under what conditions
An MVP is not only a feature reduction. It may also limit users, locations, volume, hours, variety, duration, or channels. A five-restaurant manual operation may be enough to investigate the reservation flow, but it cannot establish that serving one hundred restaurants will cost the same or maintain quality.
Prepare a record with the hypothesis, eligible users, end-to-end scope, resources, support, necessary data, measures, stopping conditions, evidence limitations, and owner. A small sample can expose serious problems but cannot automatically estimate population-wide rates. There is no universal number of MVP participants.
The UK government's beta guidance illustrates limited-group rollout and learning before expansion [R02]. It is an example of gradual delivery, not a mandatory scale for every business.
6. Check that the version supports a useful test
Ask five questions:
- Can a person achieve and recognize a real outcome?
- Can this version reveal the behavior that matters for the hypothesis?
- Can results be interpreted as supportive, unfavorable, or inconclusive?
- Have relevant safety, quality, and compliance conditions been addressed?
- Is the test narrow enough that the team can change course before a larger commitment?
If the first fails, you may have only a functional fragment. If the second fails, investment may not address the current uncertainty. If the third fails, the version may work but produce uninterpretable data. If the fourth fails, redesign or stop the trial. If the fifth fails, the cost of learning may be unreasonable.
Do not rely solely on stated opinions. Observe use, repeat behavior, task completion, costs, errors, and real commitments when appropriate. A waitlist may suggest interest in an offer but does not prove retention or willingness to pay if no transaction took place.
Once the MVP has a defined flow and learning question, plan solution tests to verify requirements, record failures, and distinguish observed use from favorable opinions.
7. Decide what happens after the evidence
The MVP belongs to a cycle. Preserve the original protocol and metrics; record incidents and unfavorable outcomes; then decide to expand, fix, change the proposition, investigate again, or stop. The Pillar 3 Design–Test Loop covers the broader iteration logic.
Do not change three components simultaneously and attribute an improvement to one of them. Document what changed and which alternative explanations remain. A new version may require fresh quality checks as well as value testing. Do not declare “product–market fit” because ten people completed a task once.
Worked example: three scopes for a restaurant booking tool
A fictional team wants to learn whether hosts gain meaningful value from recording and confirming booking changes.
Option A—structured manual reception. Two restaurants use a simple form while a person confirms changes by phone; the team logs updates, time, and errors. This could provide evidence about the flow and human workload. It does not demonstrate self-service app adoption or higher-scale cost.
Option B—minimum digital flow. Hosts record, change, and confirm bookings through an interface. The team observes whether employees outside the design group complete tasks and return to use it. Appropriate access, data integrity, and support controls are needed. This does not prove business profitability.
Option C—full platform. Advanced analytics, multiple locations, payments, and integrations might help later, but they introduce uncertainty and costs unrelated to the immediate question.
The team provisionally selects B because it needs to observe independent interaction. It specifies eligible users, change cases, testing period, issue logging, and stopping conditions. If B fails due to confusing confirmation, it fixes the flow before scaling. If it works, the next question might concern capacity or payment behavior. This is an instructional choice, not a universal recommendation.
If the decision is to continue with changes, manage versions and releases without erasing the conditions of the first trial. Comparison requires knowing what was delivered and to whom at each stage.
Common mistakes and limitations
Calling a noninteractive sketch an MVP; confusing minimum with poor quality; building disconnected features; measuring downloads or enthusiasm alone; recruiting unrepresentative users without disclosure; promising nonexistent automation; neglecting support and recovery; omitting safety requirements; treating a free pilot as evidence of payment; declaring a single successful session proof of complete validation.
An MVP is not permission to breach requirements. Nor must every existing product or improvement be rebooted as an MVP. The method is valuable when material uncertainty justifies bounded learning.
What you should have when finished
An MVP scope record documenting the pending decision, hypothesis, core outcome, smallest complete flow, justified inclusions and exclusions, obligations, users and limits, expected evidence, stopping conditions, and next decisions. A short feature list alone does not specify what you will learn.
References
- [R01] Eric Ries, published by Lean Startup Co. What Is an MVP? https://leanstartup.co/resources/articles/what-is-an-mvp/
- [R02] UK Government Digital Service. How the beta phase works. https://www.gov.uk/service-manual/agile-delivery/how-the-beta-phase-works