What an MVP Actually Costs in 2026
Our own delivered range, labelled as ours, with the line items behind it — plus why every published MVP price you can find comes from someone selling MVPs.
Contents8 sections
Ask five agencies what an MVP costs and you get five numbers with no methodology between them. That is not evasion so much as structure: the only people who publish MVP prices are the people selling MVPs, which makes every figure in the search results a positioning statement. This page publishes ours anyway, with the line items, and is explicit about which parts are evidence and which are our own experience.
What does it cost?
What a first build costs us to deliver, by scope
Fixed-price range, USD, over a 30–90 day window
- Prototype — one flow, not for real users$12K
A demo. Do not run a business on it
- Focused MVP — one workflow, real users$42K
Typical range $25K–$60K
- MVP with payments and roles$70K
Money and permissions are the expensive parts
- Two-sided or multi-tenant product$95K
Two user types is close to two products
Midpoints of our own ranges. The spread inside each band is almost entirely about how settled the scope was on day one, not about the feature count.
Codeable delivered-project range — our own figures, not industry research
What about in the UK?
Same scopes, and two differences that matter to the total rather than to the price. Our bands are set in dollars because that is where most of our work is contracted, so the sterling figures below are a conversion of the same ranges rather than a separately measured UK dataset — we would rather say that than present a converted number as a second source.
| Scope | USD | GBP (approx.) |
|---|---|---|
| Prototype — not for real users | $10K–$15K | £8K–£12K |
| Focused MVP — one workflow, real users | $25K–$60K | £20K–£48K |
| MVP with payments and roles | $55K–$85K | £44K–£67K |
| Two-sided or multi-tenant product | $80K–$120K | £63K–£95K |
Two things that change the UK total, not the UK price
- VAT. A UK-registered supplier charges it and a VAT-registered buyer reclaims it, so it is cash flow rather than cost — but it is 20% of the invoice sitting on your balance until you do, and first-time founders routinely budget the ex-VAT number.
- The contractor comparison moves. If your alternative to an agency is a day-rate contractor, employment status changes who carries the risk and can make the contractor the more expensive option for identical work. The real cost of an engineer in 2026 covers the current position.
What are you actually paying for?
Not typing. The proportions below are ours, from our own delivery, and the useful thing about them is how small the coding share is — which is why "can you just build it cheaper" usually means "can you skip the parts that stop it failing".
Writing the code is under half of a first build
Share of effort on a typical focused MVP
- Engineering45%
- Design and product decisions18%
- Testing and QA14%
- Infrastructure, deployment, environments13%
- Project management and communication10%
Where an agency hides project management tells you something. As a line item it is visible and negotiable; folded into the rate it is neither.
Codeable delivery data — our own project mix, not an industry benchmark
What turns a fixed price into a change order?
Four things, and each is answerable before the contract exists. This is the conversation most founders skip because it feels adversarial — it is the opposite, because a partner with clear answers here has thought about the failure case.
The four that actually move the number
An integration nobody has read the documentation for
"Connects to your CRM" is not a scope line. Which system, which records, whose test environment, and who owns the credentials? An integration against an unread API is the most common overrun we see, and it is usually discovered in week three.
Permissions described in one sentence
"Users and admins" becomes five roles the first time somebody draws a real workflow. Roles are nearly free to decide up front and expensive to retrofit, because they touch every query in the system.
Content and data that does not exist yet
The build finishes and waits three weeks for copy, a product catalogue, or an export from the old system. Nobody did anything wrong and the timeline still slipped, and a slipped timeline is a cost.
A decision-maker who has not seen the scope
A co-founder or investor reviews it late and reopens settled questions. Get everyone who can say no into the scoping conversation before it is priced.
What is the expensive mistake?
Building the wrong thing well. CB Insights analysed 431 venture-backed shutdowns since 2023: 70% ran out of capital, and the root causes underneath were poor product-market fit at 43%, bad timing at 29% and unsustainable unit economics at 19%. Companies are counted under more than one cause, and the dataset is venture-backed shutdowns specifically — but the direction is unambiguous and none of it is an engineering failure.
Where we are the wrong call
- Under about $10,000. We do not take scopes that small, and at that level you should be looking at a freelancer or building it yourself first.
- You have not validated anything. Build the cheapest thing that answers the question. That is sometimes a landing page, and it is occasionally a spreadsheet.
- You need it in three weeks. Our published window starts at 30 days and compressing it means cutting scope, not adding people.
- You are optimising purely on rate. Several firms will come in under us, and with a settled spec and someone technical managing it, that can be the right trade.
Common questions
Why is your number higher than the ones I found online?
Usually because the cheaper figure describes less. The common omissions are testing, deployment and environments, and any real design work — each of which is genuine work that someone does eventually, normally you, later, under time pressure. Normalise what is included before comparing totals and most of the gap explains itself.
Should I just build it myself with an AI tool first?
Sometimes, genuinely. If the goal is to see whether anyone wants it, a tool-built version in a fortnight is excellent value and we would rather you did that than spend $40,000 finding out. The trap is the one afterwards: it works, people pay, and nobody can safely change it — which is what our AI-app rescue guide is about.
What is the cheapest thing that is still worth building?
One workflow, one type of user, real traffic and some way of measuring whether it worked. That is a genuinely useful product and it is the shape we price at the lower end of our range. Breadth before reliability is how first builds end up shelved.
Does fixed price mean you cut corners to protect margin?
It is the right thing to be suspicious of. The protection is that the scope is written down and the change mechanism is in the contract, so the incentive is to scope honestly at the start rather than to under-deliver at the end. Ask any fixed-price firm what happens when they have underestimated — the answer tells you which way the incentive actually points.
What should I budget for after launch?
Plan for it as a line item from the start. The common convention is 15–20% of the build cost per year, though that figure is an industry habit rather than a research finding. What is better attributed is that maintenance is the majority of a system's lifetime cost — published estimates run from two-thirds to over 90%.
The short of it
Our range is $25K–$60K for a focused first build, it is our figure rather than a market fact, and the number matters less than what is inside it. Pin the scope, get the integrations and permissions named before pricing, and spend the saved money on finding out whether anyone wants the thing.
