How much does it cost to build software in 2026? What to budget

Budget £30,000 to £300,000 for a software build. Budget £10,000 to £50,000 if what you need is a process you already run being done by agents instead of people. Those are fixed sums, agreed before anyone writes a line of code, and they do not move unless you change what you asked for. We never quote a day rate, and that part is worth explaining, because the day rate is where most software budgets go wrong.

If you have been searching for this number, you already know the problem. Type the question into Bing and one of the suggested completions is the same question with “Reddit” on the end. People are asking strangers on a forum, because the companies who actually know keep answering “it depends” and quoting a range from nothing to half a million. Nobody can budget against that.

What to budget

Three lines cover most of what people arrive asking for. They are ranges rather than a menu. What you actually pay depends on scope that neither of us can see until we have looked properly. Treat these as the number to put in a budget, not the number that ends up on an invoice.

Start with a discovery, £5,000 to £20,000. It is the only way we know to scope a build honestly. You own the output whether you proceed with us or not, so if you walk away, you walk away with a usable spec you can hand to another team. When the discovery is scoping a build, we take the fee off the build price. When it turns out to be advisory work with no build behind it, it is priced standalone and we say so before you commit.

Then the build itself, fixed price, agreed before anyone writes code.

• Micro build, £30,000 to £60,000, two to five weeks. One workflow, a dashboard, a small internal tool, or a single-feature MVP.
• Standard build, £60,000 to £120,000, five to ten weeks. That covers a full MVP, a multi-workflow internal tool, or a data product.
• Platform build, £120,000 to £300,000. Three to five months for a SaaS product, an AI-powered tool, or a multi-user system.
• Enterprise build, from £300,000. Four to six months at least, and there is usually real compliance, integration or scale work sitting in here.

We do not take every build. The ones that suit us are greenfield or well defined, where AI-native delivery makes a real difference rather than a marginal one.

Not every build is a product, though. Sometimes you have a process that already works, run by people, and you want agents doing most of it instead. Invoice matching. Chasing purchase orders. Pulling five systems together into one weekly report. Budget £10,000 to £50,000 for that. It is the cheapest useful thing most companies buy from us, and a fair few should be asking for it instead of a platform.

The third line is hardening, and it has become a common way for people to arrive. You vibe-coded a prototype, it proved the idea fast, and now it has to hold up with real users and real data. Secrets and API keys sit in the source. The database will not survive fifty people writing at once. When something breaks at 2am, you hear about it from a customer rather than from your own tools. We wrote up the six places these prototypes actually break, with a checklist you can run yourself. Answer no to eight of the questions on it and you are looking at a rebuild. Answer no to one or two and it is an afternoon. £20,000 to £75,000 covers almost everything that lands in between, and every job starts with a fixed-price audit whose fee comes off the work.

Every one of those prices includes our AI costs. Token spend, tooling seats, model bills, all of it sits inside the number. We do not meter it and we do not pass it through. The moment a meter appears you start watching the meter instead of the software, and we pick up a quiet incentive to burn more tokens rather than fewer.

Why the day rate is the wrong unit

A day rate prices the input. You buy days of a person’s attention and hope enough of them add up to a product. It punishes speed. It rewards the wrong people. The economics only work if the work takes longer than you wanted it to.

For twenty years that was mostly a philosophical objection, because the days genuinely were the cost. Six people over roughly six months, at a blended UK agency rate of around £600 a day per head, is about £300,000. The price described something real about how the work got done.

That relationship has come apart. Three senior people working with AI agents now ship what the six-person squad shipped, in a fraction of the calendar time. When two people can produce what six to ten used to, the day-rate model breaks. If we billed you by the day we would charge you less for a better result. That is a strange way to run a business. Pricing the outcome instead makes the efficiency our problem, and our reward, and the number you were quoted is the number you pay. It is why we abandoned rate cards entirely when we started the studio.

So ask a prospective supplier what happens to your invoice if the build takes half as long as expected. If the honest answer is that it goes down, and so does the amount of software you get, you are buying days rather than products.

Why two quotes are never the same thing

Comparing prices across a shortlist is harder than it looks, because the numbers are rarely describing the same deal. Four questions sort most of it out.

Is the figure fixed, or is it a starting price? Plenty of quotes are indicative, and indicative numbers go up. Ask what happens to it if the work takes longer than planned. Ours is a ceiling, and it only moves if you change what you asked for.

Then ask who is doing the work, and where. Many quotes assume a mostly offshore team with one onshore lead on the calls. That is a legitimate way to run a business and it produces a lower headline number. It is not what you are buying from us. We are fully onshore.

Seniority is the part people forget to check. Every Forward Deployed Builder in the LevelFive network has a minimum of ten years of experience. No exceptions, no graduate programmes, and some of them have twenty. The person you meet in the pitch is the person who does the work. A blended day rate can hide a lot of juniors.

Last, check what sits inside the number. Ours covers design, our AI costs, and the discovery fee where it credits back.

Against a traditional time-and-materials consultancy doing the same scope, we come out cheaper, and we will show you the working. Against someone pricing low to win the work, we will not, and nor will anybody else. That number usually finds its way back up later through change requests.

None of that makes us the cheap option. Fast, cheap, good, pick two was never a law of software. It described one particular way of making it, a room of people typing and coordinating, where going faster meant adding people and adding people cost you quality. Change the shape of the team and the rule stops holding. One senior builder took a regulated building-safety product from a standing start to production in under two weeks. It cost roughly a quarter of a traditional engineering build. In testing it reached the same conclusion as the client’s own experts 95% of the time. You are buying a better outcome, sooner, and the lower price follows from how the work gets done rather than from a discount on it. The longer argument is here.

What actually moves the number

Scope is the obvious answer and it matters less than people think. Three things genuinely change our price.

Integrations, because connecting to a system somebody else controls is the one part of delivery that AI has barely improved. Compliance, because a regulated build carries evidence, audit and review work that happens at human speed. Data, because if yours is messy, undocumented, or spread across four systems that disagree with each other, cleaning it up is the project and the software is the easy part.

The volume of code barely moves the number any more. That used to be the whole estimate.

What AI changed, and what it did not

AI has collapsed the cost of producing software. A prototype that needed £20,000 and a small team two years ago now costs about the same as a monthly subscription. That is useful for testing whether an idea has legs.

It has not touched the cost of knowing what to build. That is now the binding constraint on almost every project we see, and it is why we insist on a paid discovery before quoting a build. Anyone who gives you a fixed price on a first call is guessing, and you pay for the guess later through change requests.

The labels do not matter

Dev shop. Software house. Digital agency. Digital product studio. IT services provider. Systems integrator. Buyers ask us which one we are, usually because they are trying to work out what a fair price looks like for that category.

The label tells you nothing. Sellers pick these names for themselves and they go in and out of fashion every few years. Two companies using the same label will quote you numbers a factor of five apart. The four questions above separate a shortlist faster than any category name.

Where that leaves you

Allocate £10,000 to £50,000 to automate a process you already run, and £20,000 to £75,000 to get a working prototype ready for real users. Allocate £30,000 to £60,000 for something small and contained, £60,000 to £120,000 for a real MVP, and £120,000 upwards for a platform. Put a discovery in front of all of it, and ask whether the fee credits against the build. Treat a supplier who wants to skip the discovery as a supplier who intends to renegotiate later.

If those numbers fit what you had in mind, start a conversation. If they do not, tell us the budget you have and we will tell you honestly whether it buys the thing you want. Sometimes it does not, and hearing that in week one is worth more than a proposal that pretends otherwise.