Ask four suppliers what a first version of your product costs and you will get four numbers that do not agree. The spread is real, and little of it is about your product. It is about how each supplier plans to staff the work, how long they think it will take, and how much of your brief they quietly plan to build.
So this is a buyer's guide to the number. What the published ranges are measuring, the five decisions that move your figure more than any of them, what the word minimum is supposed to mean, and the case where the honest answer is to build nothing at all.
The short answer
The wider market publishes 150 to 350 dollars per hour for senior independent talent, and 20,000 to 150,000 dollars for a scoped build. We quote the same band on our own consultant cost page, and it is honest as far as it goes. A first version of a small product sits near the bottom of it. One that carries payments, several kinds of user and a compliance story sits near the top.
That range is wide enough to be useless on its own, which is the problem with every price table you will find on this question. Ranges describe suppliers. They cannot describe your product, because nobody has decided yet what your product is.
What the price bands are really measuring
Read the published tables closely and the axes are always the same three: complexity, team model and team location. Complexity is a word for how many features somebody agreed to build. Team model is whether you hire staff, contractors or an agency. Location is a proxy for salary. All three are properties of the arrangement you sign.
That is why two suppliers can quote the same brief a factor of five apart and both be telling the truth. One is pricing four people for four months. The other is pricing two people for three weeks and a hard refusal on everything outside the core. The second one is not cheaper. It is smaller.
Five things that move your number
These decisions change the figure before a line of code exists. Take them one at a time and write down your answer to each.
How many kinds of user it has
One kind of user means one set of screens, one permission rule and one path through the product. Three kinds is not three times the work. It is more, because every screen now needs a rule about who may see it, and every rule needs a test that proves it holds. When we co-built diil.ai the platform carried developers, agents and investors, and that separation was the largest structural decision in it. Tenants were isolated by subdomain with row level security in Supabase, rather than by permission checks scattered through the application. That choice is cheap on day one and painful to retrofit in month six.
How many systems it has to talk to
Count the boxes your product has to reach: the accounting system, the CRM, the calendar, the payment provider, the mail sender. Each connection is a piece of work with its own failure modes, and the number of connections grows faster than the number of boxes. If your list is long, read our note on build versus buy first, because part of that list is usually a product somebody already sells and maintains for you.
Whether money moves through it
A product that shows information and a product that takes money are different animals. Once Stripe or any other provider sits in the path, you inherit refunds, partial refunds, failed payments, invoices, tax treatment and a reconciliation story for your accountant. None of that is exotic. All of it is work, and it is the work that goes missing from optimistic quotes.
What happens to the data if it leaks
If the product holds names, phone numbers, health details or signed contracts, the build is not the only cost. Backups, access control, audit trails and a deletion route stop being nice to have. Ask where the data will sit and who can read it in production. If the answer is vague, the quote is missing a line, and you pay for that line later at a worse moment.
How much you refuse to cut
This one moves the number most and nobody asks about it. Every feature kept in the first version is bought twice: once to build it, and again to maintain it while you find out whether anyone wanted it. A supplier can discount an hourly rate by a fifth. Only you can remove a third of the scope.
What minimum is supposed to mean
Minimum viable does not mean cheap, and it does not mean unfinished. It means the smallest thing that produces a real answer to one real question. The question has to be written down before the scope is, because a first version with no question attached will grow until the budget stops it.
Here is a test that works. Name the decision you will make when the thing has been live for a month. If the honest answer is that you will see how it goes, you are not ready to build. If the answer is that you will know whether agencies will run their pipeline in it instead of a spreadsheet, you have a scope, and most of the feature list can go.
The cut list is where the money is. An admin panel three people can live without, because a database client does the same job. Reporting that a query answers. Onboarding flows for users who do not exist yet. A native app when a web page on a phone answers the same question.
Three shapes, and the honest price of each
The throwaway test
A page, a form, a manual process behind it, and you doing the work by hand for the first customers. It is not glamorous and it is often the correct build. Our seven day site launch is this shape: a real page in front of a real market before anyone commits to a product. If the form stays empty you have your answer for close to nothing.
The internal tool
One kind of user, your own staff, no public sign up and no payments. This is the cheapest shape that produces durable value, because the users are down the hall. You watch them work on Thursday and change it on Friday. It also fails quietly when nobody asked for it, the pattern we describe in why pilots do not reach production.
The real product
Several kinds of user, tenants that must not see each other, money moving through it, and people outside your company depending on it. This is the shape the top of the published range describes, and it is also where speed is possible when the scope is honest. On diil.ai the agencies quoted 80,000 to 150,000 dollars and four to six months. The version that reached production shipped in 21 days with six modules, and it now carries more than 100 million dollars of transactions across more than 50 agencies in 12 countries. The saving did not come from cheaper people. It came from deciding what the six modules were and refusing the seventh.
Four questions that change a quote
- What would you cut if the budget were half this? A good supplier answers in one breath, because they already ranked the list. One who cannot has estimated the work, not scoped it.
- Which parts are you buying rather than building? Payments, mail, authentication and hosting should be bought. If a quote contains a bespoke version of any of them, ask why.
- Who owns the code and the accounts at the end? Everything should end up in your accounts, in your name. Cheap to insist on at the start, expensive to fix at the end.
- What is the monthly bill after launch? Hosting, database, mail and monitoring. If nobody has added it up, add it up yourself before signing.
The costs that never appear in the quote
- Running costs. Hosting, database, mail, payments and monitoring are monthly and permanent.
- Maintenance. Dependencies move, providers change their APIs, browsers change. A budget with no maintenance line describes a product that gets rewritten sooner than it should be.
- The second month. The first version tells you something. Acting on what it tells you is another piece of work, and it decides whether the first piece was worth doing.
- Your own time. Every build needs someone on your side who answers questions within a day. Where nobody has that time the calendar stretches, and the calendar is what you are paying for.
Two habits cut the running bill more than any negotiation: boring infrastructure and fewer moving parts. PostgreSQL behind a managed host, one deployment target such as Vercel, workflow automation in n8n or Make instead of bespoke glue code, and a short set of end to end tests in Playwright so a change on Monday does not cost you Tuesday.
The cheapest first version is the one you do not build
Sometimes the software already exists. If a product on the market does most of what you described, buy it and commission only the gap it leaves. We wrote the test for that call in when to build a custom app. Custom is right when the thing you do differently is the thing the software has to do. Custom is wrong when the thing you do differently is your taste in screens.
The other case is smaller than software altogether. For a Tartu roadside operator the work was a site rebuilt as 24 pages, one per real search, with zero third party scripts and 16 requests in total. For an electrical contractor it was a six page brochure grown into 125 URLs in three languages, with the 13 year old domain kept and no ranking page lost in the migration. Neither was a product build. Both moved the number a product build was supposed to move. The case studies we publish also say what each build did not prove.
What we charge
Our audit is free, takes about 30 minutes, and ends with a written list of where your business leaks money, ranked by monthly cost. Build and run starts at 600 euros per month plus VAT, and the number is quoted after the audit rather than before, because the honest figure depends on how many kinds of user you have, how many systems sit in the chain, and how much of the list you are willing to cut. The method is on the consultancy page.
Questions people ask about MVP cost
How much does an MVP cost?
The wider market publishes 150 to 350 dollars per hour for senior independent talent and 20,000 to 150,000 dollars for a scoped build. Where you land inside that band is set by five things: how many kinds of user the product has, how many systems it must talk to, whether money moves through it, how sensitive the data is, and how much of your feature list you cut before anyone starts.
Why do quotes for the same brief differ so much?
Because they are not quotes for the same amount of work. Published price tables sort by complexity, team model and team location, and none of those is a property of your product. One supplier prices four people for four months. Another prices two people for three weeks plus a refusal on everything outside the core. The second is not cheaper, it is smaller.
Can I build a first version for a few thousand euros?
Yes, when the shape is a throwaway test: a page, a form, and you doing the work by hand behind it. That answers a demand question honestly and costs close to nothing. What a few thousand euros does not buy is several kinds of user, payments in the path and data you would be sorry to lose. Pick the shape first, then the budget.
How long should a first version take?
Weeks, not quarters, when the scope is honest. We co-built diil.ai from concept to production in 21 days with six modules, against agency quotes of four to six months. Speed came from deciding what the six modules were and refusing the seventh, not from cheaper people or longer days.
Should I build or buy?
Buy whenever a product on the market does most of what you described, then commission only the gap. Building is right when the thing you do differently is the thing the software itself has to do.
If you want your own number instead of a band, the fastest route is not a longer brief. It is a shorter one. Write down the single question your first version has to answer, cut everything that does not serve it, then ask for prices. If you would like that done with you, book the free audit. Thirty minutes, your real constraints, and a straight answer on whether the build is worth doing at all.