What an AI MVP Actually Costs — And Where the Money Goes

Three Quotes, One Idea, A 30x Spread

A founder sends the same two-page brief to three vendors. A retrieval-backed assistant for a niche professional audience, login, a paid tier, nothing exotic. The quotes come back at $12,000, $65,000 and $240,000.

Nothing about the idea changed between those emails. What changed was what each vendor silently decided the word MVP meant. One priced a demo. One priced a product. One priced a product plus six months of their bench.

After shipping 100+ applications and reviewing a lot of other people's quotes, I can tell you the spread is almost never about rates. It is about scope that nobody wrote down. This post writes it down.

The Expensive Quote Is Not A Scam

I want to be fair to the $240,000 number before I take it apart. A build at that price usually buys real things: a discovery phase, a design system, a dedicated team, SOC 2 paperwork, a roadmap past launch. For a funded company with a known market, some of that is worth paying for.

The problem is that most founders asking for an MVP are not at that stage. They need to find out, cheaply and quickly, whether anyone wants the thing. Paying for production hardening before you have proven demand is the same mistake as over-engineering before the use case is clear — just with a larger invoice.

The Four Cost Layers Of Every AI MVP

Every AI product I have scoped breaks into the same four layers. Vendors disagree on price because they are quoting different subsets of these layers without saying which.

A $12,000 quote is usually Layer 1 plus a thin Layer 2. A $240,000 quote is all four, plus a team retained to keep going. Neither vendor is lying. They are answering different questions.

Where the money really goes

When we build a full production MVP, the budget does not go where founders expect. The model and the prompts are a minority of the work. The majority is the plumbing and the hardening that make the model usable by strangers.

What The Quote Does Not Say

Three costs hide outside the headline number, and they decide whether a cheap quote stays cheap.

Open scope billed hourly

An hourly engagement with a loose brief is not a price. It is an opening bid. Every clarification becomes a change order, and the founder — who cannot judge whether a task should take four hours or forty — carries all the risk. More than once a founder has shown me a $15,000 MVP that closed at three times that, with nothing shipped that a user could log into.

The prototype that gets thrown away

Many shops build a disposable demo first, then start over for the real thing. You pay for the same screens twice. This is the quiet cost of prototypes that were never designed to survive contact with real users.

The bill that arrives after launch

Model API calls, vector storage, and hosting are a recurring cost that scales with usage. A vendor who never asks about your expected volume has not thought about your margins. If the AI costs $0.40 per user session and you charge $9 a month, the build price was never your real problem.

Our Numbers, In Public

I think the honest move is to publish prices, so here are ours. A 72-hour clickable prototype is a flat $1,950: a production-styled Next.js interface with the core AI journey working, ready to put in front of investors or users. If you go on to the full build, that entire fee is credited against it.

A production MVP starts at $8,500 and ships in three weeks: authentication, PostgreSQL with pgvector for retrieval, agent tool-calling with guardrails, Stripe billing, cloud deployment, and 30 days of post-launch fixes. Ongoing engineering after that is a separate monthly retainer, and native mobile apps are out of scope — we build web and PWA first.

The reason a fixed price is possible at all is that the plumbing and the hardening are not reinvented per client. The harness is the part that compounds, and after 100+ applications, ours is mature enough that the risk of a fixed scope sits with us, not with the founder.

The Three Questions To Ask Before You Pay

Whatever vendor you choose, these questions will tell you which of the four layers you are actually buying.

First: which layers are in the price? Ask them to name authentication, billing, deployment and monitoring explicitly. If any of those is “phase two”, your quote is for a demo.

Second: what happens to the prototype? If it is thrown away, you are paying for the UI twice. If it is credited or carried forward, the vendor has designed for continuity.

Third: what will it cost per user per month to run? A vendor who can answer within a reasonable range has built AI products before. One who cannot is about to learn on your budget.

Scope Is The Price

An AI MVP does not have a price. A scope has a price. The 30x spread between quotes is not three vendors disagreeing about value — it is three vendors answering a question the founder never pinned down.

Pin it down first. Decide which layers you need to learn what you need to learn, buy exactly those, and make the vendor carry the risk of delivering them. That is how a $240,000 question becomes a $8,500 one.

If you are deciding what to build and want a fixed number instead of an estimate, that is exactly the conversation we have with founders at Will of Dawn Labs. You can also book a 30-minute scoping call directly.

— Kaushal Malhotra
Founder, Will of Dawn Labs
willodawn.com/contact