There is no single price. A custom AI build is priced by the job: its complexity, the systems it touches, the state of your data and the checks the output needs. Published UK agency guides suggest roughly £21,000 to £400,000 or more. We quote per project, after a paid mapping phase.
What drives the price of a custom AI build?
Four things set the price of a custom build: how complex the job is, how many systems it has to talk to, the state of your data, and how many checks the output needs before a person can trust it. Everything else is detail.
Complexity. A system that drafts one document from a clean template is a small build. A system that applies hundreds of your specific rules, handles the exceptions, and knows when to hand a case to a person is a bigger one. The question that matters is not "how clever is the AI" but "how many decisions does this job contain, and how many of them are yours alone".
Integrations. Every system the build has to connect to adds work: your CRM, your accounting software, your job boards, your document store. Modern software that's designed to be connected to keeps this cheap. Old software with no proper way in can double a quote on its own, because someone has to build the way in.
Data state. If the information the system needs lives in one or two tidy places, the build moves fast. If it's scattered across spreadsheets, inboxes and PDFs in fourteen formats, a chunk of the budget goes on getting the data into a usable state before the interesting work starts. This is the cost item most owners don't see coming.
Checks. When a wrong answer is cheap, you need few guardrails. When a wrong answer means a mis-paid employee or a compliance breach, the build needs verification steps, an audit trail and clear rules for when a person takes over. A payroll checking system we shipped spends a good share of its design not on the checking itself but on proving every check happened: it runs in the client's own cloud and keeps a full audit trail. That rigour costs money and is worth every penny in the jobs that need it.
What does the market charge for custom AI?
Published UK agency pricing guides put custom AI software development at roughly £21,000 to £400,000 or more, depending on complexity. The same guides price a basic AI integration at £3,500 to £21,000, and advanced components at £21,000 to £35,000 each.
Treat those numbers with care. They come from a development agency's own pricing guide, not from independent research, and there is no government or analyst source for what small-business-scale AI builds cost. They are market figures, not our prices, and they're useful for exactly one thing: a sanity check. If someone quotes you £5,000 for a system that touches payroll, or £300,000 for a document formatter, the range tells you a question needs asking.
The subscription alternative has a cost too, and it's easy to undercount because it never stops. We've set out that comparison properly in our guide to build versus buy.
How does our engagement shape price it?
We price in three parts: a paid mapping phase at a fixed fee, a build quoted per project and delivered in two-week sprints, and optional monthly support you can cancel any time. No number goes on this page, and that's deliberate: anyone who prints a price before seeing your week is guessing, and you'd be paying for the guess.
The mapping phase exists to kill the guesswork. We sit inside the job, count the hours it eats, find the rules and the exceptions, and come back with a plan and a firm quote. If the quote doesn't make sense against the hours saved, you walk away with the plan and owe nothing more. The sprints exist so you see working software every fortnight rather than a slide deck at the end, and the support is optional because the system is yours: we've written about what that ownership means in who owns the AI you pay for.
When is building not worth the money?
A custom build is not worth it when the job is generic, the hours are small, or your data isn't ready, and a builder who won't say so to your face is the wrong builder. By some estimates, more than 80% of AI projects fail, twice the failure rate of comparable non-AI IT projects, and a 2025 survey by S&P Global found 42% of companies had abandoned most of their AI initiatives, up from 17% a year earlier. Most of those failures were projects that should never have been started.
The honest tests: if a £30-a-month tool already does 90% of the job, buy the tool, and see where off-the-shelf tools hit their limits before you assume you're the exception. If the job eats two hours a month, no build price on earth pays back. And if nobody in the business can say where the source data lives, fix that first: it's cheaper than discovering it mid-project.
| Makes a build cheaper | Makes it dearer |
|---|---|
| One well-defined job with clear rules | A vague remit to "add AI" across the business |
| Data already tidy, in one or two systems | Data scattered across spreadsheets, inboxes and PDFs |
| Modern software that's built to be connected to | Old systems with no proper way in |
| A person reviews the output before it's used | Output must be right on its own, every time |
| Low cost of a wrong answer | Regulated or high-stakes work needing audit trails |
How do you get a straight number for your business?
You get a straight number by doing the arithmetic on a real job, and that starts with counting what the job costs you now. One team we work with loses eight to ten days, every single month, building one report by hand. Put a day rate against that and the ceiling for a sensible build price falls out of simple multiplication: what the job costs a year, what a system would give back, how many months to break even.
That's the whole method, and it's what a free Impact Call does: you tell us where the week goes, we run that arithmetic on one real job in your business, and you leave with a one-page opportunity map. Sometimes the map says "buy a tool" or "don't touch this one", and we'll say so, because a build that doesn't pay back is bad for exactly one party's reputation: ours.
Sources
Figures and claims in this guide draw on our own delivery work and the sources below. We only publish numbers we can stand behind.
- Appinventiv, "How Much Does AI Software Cost in the UK?", published 25 June 2026, accessed 8 July 2026: appinventiv.com/blog/how-much-ai-software-cost-in-uk. A vendor pricing guide, cited as an indicative market range only.
- RAND Corporation, "The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed", published 13 August 2024, accessed 8 July 2026: rand.org/pubs/research_reports/RRA2680-1.html.
- S&P Global Market Intelligence, 451 Research "Voice of the Enterprise: AI & Machine Learning, Use Cases 2025", published 2025, accessed 8 July 2026: spglobal.com/market-intelligence.
- AI Nativ.es delivery experience, 2026: the payroll checking build and the reporting build described are real client projects, told without names until we have written permission to use them.