Resources

Planning

MVP vs prototype vs proof of concept: what should you build first?

A proof of concept, a prototype and an MVP answer different questions. Here's how to tell which one your AI idea needs right now.

By the Jerrax team · · 6 min read

“MVP”, “prototype” and “proof of concept” get used as if they mean the same thing. They don’t. Each answers a different question, and building the wrong one is one of the most common ways to spend money without learning anything.

Three different questions

  • Proof of concept (PoC): Can this work at all? It tests technical feasibility — usually the riskiest assumption — and nothing else.
  • Prototype: How will it work for the user? It shows the experience — screens, flow, interactions — usually without a real backend.
  • Minimum viable product (MVP): Will people actually use it? It’s a real, working product that real users can sign up for and rely on, built around one core job.

Side by side

Proof of conceptPrototypeMVP
AnswersIs it technically possible?Does the experience make sense?Do people use it — and pay?
AudienceYour teamUsers, stakeholders, investorsReal users
What’s realThe risky technical partThe interfaceEverything in the core workflow
Thrown away?UsuallyUsuallyNo — it’s the foundation you grow

What changes when the product uses AI

With AI products, the proof of concept matters more than usual, because the riskiest question is often is the model good enough on my data? You can answer that cheaply: take a few dozen real examples — documents, tickets, emails — run them through a model with a clear instruction, and judge the output honestly. If the results are close to useful, you’re ready to build. If they’re not, you’ve saved yourself a product nobody can rely on.

An AI MVP goes further than that test. It wraps the AI step in everything a real user needs: sign-in, a place to put their input, stored results, sensible handling when the model gets something wrong, and limits so usage costs stay under control.

Which should you build first?

  1. Not sure the AI can do the job? Start with a proof of concept on real examples.
  2. Need buy-in for how it should look and flow? A clickable prototype is enough — don’t pay for a backend yet.
  3. Confident it works and know the flow? Build the MVP and put it in front of users.

A good test: if you can describe the core workflow in one sentence and you’ve seen the AI handle your real examples, you’re ready for an MVP.

Common mistakes

  • Calling a prototype an MVP. A clickable design can’t tell you whether people will keep using the product.
  • Skipping the feasibility test. Building screens around an AI step that turns out to be unreliable is the most expensive way to find out.
  • Over-building the MVP. “Minimum” is the important word. One workflow, done properly, teaches you more than five done halfway.

Next step

When you’re ready for the MVP, the next job is keeping it small enough to ship. Read how to scope an AI MVP you can launch in 14 days, or see how a Jerrax build works.