Prototype, proof of concept and MVP, side by side
The three are often used as if they meant the same thing. They do not, and the difference is not size. Each one is built for a different audience and thrown away, or kept, for a different reason.
| Prototype | Proof of concept | MVP | |
|---|---|---|---|
| The question it answers | Will people understand it? | Can the hard part be built at all? | Will people use it? |
| Who sees it | Test users, investors, the team | The team | Real users, on their own |
| What it is made of | Screens, often with nothing behind them | Only the risky part, on real inputs | One working flow, sign-in, an admin, kept data, deployment |
| What happens to it afterwards | Replaced by the real build | Usually thrown away; the answer is kept | Becomes the next version |
| How long it takes | Days | As long as one question takes | 14 days to 8–12 weeks with us |
| Price with us | Not a price we publish | $100 an hour for the AI question, no minimum | $5,800 to $120,000, fixed |
The first two columns describe general practice rather than anything we measured. The MVP column and both price rows are our own published figures, from the MVP, AI MVP and AI consulting pages.
What a prototype proves, and what it cannot
A prototype is a model of the product: a clickable design, or a front end running on made-up data. It answers whether people understand what the product is for and can find their way through it, and it answers that quickly, because nothing behind the screens has to work.
What it cannot answer is whether anyone will come back. Nobody signs in to a prototype, nobody’s data is kept, and nobody uses it without someone watching. A test user who says they would use it has told you they understood it, which is all a prototype can measure.
The trap is the one described in what an AI MVP costs: an impressive prototype is days of work, and the distance from it to something you can charge for — auth, tenancy, billing, error states, evaluation — is where the budget goes. A prototype that demos well is easily priced as if it were most of the product. It is the smallest part.
What a proof of concept proves
A proof of concept takes the one part nobody is sure can be done and does it, on real inputs, with nothing around it: no sign-in, no admin, no design. It proves feasibility and nothing else, which is why it is usually thrown away once the answer is in. The answer is what you keep.
Most products do not need one. If the risky part of your idea is whether users want it, a proof of concept answers a question you did not have. It earns its cost when there is a real technical unknown: a performance limit, an integration nobody has tried, a data set nobody has looked at.
An AI proof of concept is a data question first
In AI products the unknown is nearly always the same one: can a model answer correctly on your documents, often enough to be worth building around? That is decided by the data more than by the model. A proof of concept on the tidy examples proves very little. The useful version looks at a genuine random sample, defines what a correct answer is, and measures against it.
That is what our AI readiness assessment does, by the hour: whether your data can carry it, retrieval or fine-tuning, the cost of one answer at your volume, how you would know it works, the architecture and the budget. It is billed at $100 an hour with no minimum, and it ends with a yes, a no, or a smaller first step. A no before the build budget is spent is the cheapest useful result it can produce. How a correct answer is measured once there is a system to measure is in RAG evaluation before production.
What an MVP proves
An MVP is the first version that is a product: built to be used rather than demonstrated, by people who signed in, with their data kept. It proves the one thing the other two cannot, that people use it when nobody is watching.
That is why our 14-day MVP is four things and no more: one main user flow, email sign-in, a simple admin and deployment, with the code in your repository from the first day. Email sign-in is there so the people testing your idea are people you can write to. The admin is there so you can see what they do without asking an engineer to query the database. AI, payments and integrations with external systems are left out on purpose: each adds its own failure modes and its own week, and none of them is needed to learn whether people want the flow.
It also starts where a prototype never has to: with the data model, meaning the records and the rules the database enforces, because they are the part that is expensive to change once users exist.
Which one to build first
Start with the question you can least afford to get wrong:
- Can the hard part be done at all? Then a proof of concept comes first, or for an AI product an hour-billed readiness assessment. Building an MVP around a technical unknown is how a budget gets spent on the parts that were never in doubt.
- Will people understand it? Then a prototype, if the idea is unusual enough that users might not see what it is for. For a product built on a familiar pattern — book, order, submit, track — this step can usually be skipped.
- Will people use it? Then an MVP. If the product does not need AI to be useful, an MVP without AI tests that fastest and cheapest.
- Is AI the product? Then an AI MVP, where the data audit and the evaluation come before the screens. Its first step is the proof-of-concept question, asked on your data before the scope is fixed, so the two are not separate purchases.
The order can also run the other way. When the AI is not yet the product, our AI MVP development page recommends testing the product first, with an MVP without AI, and adding the model where users show you it helps.
What each one costs with us
We publish fixed prices for the MVP and the AI MVP, and an hourly rate for the questions that come before either. We do not publish a price for a prototype on its own.
| Build | Price | Time | What decides it |
|---|---|---|---|
| The 14-day MVP | $5,800 | 14 days | One main user flow, email sign-in, a simple admin, deployment |
| MVP without AI | Around $8,000 | 2–3 weeks | More than one flow, payments or an integration |
| Multi-tenant SaaS MVP | $40,000–$90,000 | 8–12 weeks | Many companies as customers, with tenancy in the schema |
| AI MVP | $40,000–$120,000 | 8–12 weeks | A model answering correctly is the product |
| AI readiness assessment | $100 an hour, no minimum | Until the answer is clear | One use case, on your own data |
What moves a build up that ladder is not the number of screens. A second flow, payments or an integration take the 14-day MVP to the standard one; many companies as customers take it to a multi-tenant SaaS; AI at the core takes it to an AI MVP. Each is a fixed price against a written scope, which comes back within 48 hours of a free 30-minute call, and every band is on the pricing page.