OOmnistu

Blog / Startup Guides

Startup Guides

MVP vs prototype vs proof of concept: choose the next experiment

Choose a proof of concept, prototype or MVP around the uncertainty you need to test rather than treating them as mandatory build stages.

OOmnistu
Reviewed 2026-11-16 · 4 min read
Match the experiment to the blocker
Feasibility unclear? Proof of concept. User flow unclear? Prototype. Offer untested? MVP - which can be a manually delivered service.
Each answers one question
A PoC does not prove demand; a prototype does not prove willingness to pay; an MVP does not prove broad market fit.
Write the decision rule first
The suggested one-page brief ends with what result would support continuing, changing or stopping.

Comparison wording from the article's table, after Atlassian's MVP guide and Agile Alliance's definition linked in Sources.

A proof of concept asks whether a risky technical idea can work. A prototype explores how an experience might work. A minimum viable product tests a business or customer assumption through something people can actually use or respond to. Choose the next step by the uncertainty you need to reduce, not by which label sounds most impressive.

These terms are sometimes used differently across teams. Before commissioning work, agree what the proposed deliverable will test and what it will not prove.

A comparison you can use in a planning meeting

Atlassian's MVP guide separates technical feasibility, an experience prototype and a usable product for early feedback. A practical version of that distinction is:

Approach Main question Possible deliverable What it does not automatically prove
Proof of concept Can this technical dependency work? A narrow integration experiment Demand, usability or production readiness
Prototype Can people understand and complete this flow? Paper screens or a clickable design Reliable live delivery or willingness to pay
MVP Will a useful offer change real customer behaviour? A narrow product or a manually delivered service Broad market fit or a scalable business

The examples are illustrative. A team may need more than one experiment, but these are not compulsory steps in a universal sequence.

Proof of concept

Can this technical dependency work?

A narrow integration experiment. Does not prove demand, usability or production readiness.

Prototype

Can people understand and complete this flow?

Paper screens or a clickable design. Does not prove reliable live delivery or willingness to pay.

MVP

Will a useful offer change real customer behaviour?

A narrow product or a manually delivered service. Does not prove broad market fit or a scalable business.

Choose a proof of concept when feasibility is the blocker

Suppose a proposed inventory app depends on reading stock levels from a supplier system. The most important early question may be whether the system exposes the required data with suitable permissions and behaviour.

A small technical experiment could answer that more directly than designing a polished dashboard. Define the success condition first: which data must be obtained, under what conditions, and which failures would make the product impractical?

Also record the limits. A successful sandbox integration does not establish production access, commercial terms, performance at scale or the right to use all the data. Those may become separate questions after the experiment.

Do not call a proof of concept production-ready merely because it worked once.

Choose a prototype when the flow is unclear

If the technology is familiar but the user journey is uncertain, a prototype can let likely users try the main steps without first building the full service.

For a shared-equipment booking idea, that might mean choosing an item, seeing availability, requesting it and understanding the confirmation. Observe where participants hesitate or misunderstand the state.

A prototype can be low fidelity when the question does not require polished visuals. It can also be more realistic when details matter. The point is the question being tested, not the number of screens.

Be careful about the conclusion. Completing a simulated booking tells you something about comprehension and usability in that setting. It does not, on its own, show that people will change their existing habits or pay for the service.

An MVP does not require a fully automated backend

Agile Alliance's definition emphasizes learning with the least effort. It explicitly includes approaches such as a landing page or a service delivered manually behind the scenes.

That matters when a useful result can be tested before automation is justified. For example, a team might manually coordinate a narrow equipment-sharing service and learn whether people return, while clearly communicating how the pilot works.

Do not interpret "minimum" as permission to mislead participants, mishandle information or deliver something too unreliable to test the idea fairly. The offer must be suitable for the learning objective and the people using it.

Also separate evidence levels. A signup indicates interest in that offer under those conditions. It is not the same evidence as repeated use or a completed purchase. Pick the behaviour that matches the assumption you are testing.

Write a one-page experiment brief

Our suggested brief has six fields:

  1. Uncertainty: the one decision currently blocked.
  2. Audience: the people whose behaviour or feedback matters.
  3. Experiment: what they or the technical team will actually do.
  4. Evidence: what will be observed, not only what people say they like.
  5. Decision rule: what result would support continuing, changing or stopping.
  6. Limits: what the experiment cannot establish.

Set the rule before seeing the result, and label any numerical target as your own hypothesis rather than an industry benchmark. A small early test should lead to a better next decision, not a confident market-size claim.

Common mistakes

Building all three by habit. A project with routine technology may not need a separate technical proof of concept. Another project may need repeated experiments on different risks.

Treating a feature count as validation. A short backlog is not enough. The delivered slice needs to answer a customer or business question.

Stopping at positive comments. Observe relevant behaviour and its context. Polite enthusiasm is useful feedback, but it is weaker evidence than the action the business depends on.

Treating the first result as final. Atlassian's guide describes using feedback to continue, iterate, reposition or stop. Define who owns that decision before adding another release to the plan.

By Omnistu. Reviewed 16 November 2026. Examples and the experiment-brief structure are illustrative recommendations, not client results or validated market forecasts.

Sources

Previous
How to write a PRD: a practical template with a worked example