What Does AI Actually Cost a UK Business?

It is the question everyone wants answered before the second meeting and almost nobody will answer in the first: what does it actually cost to get AI doing something useful in a business of a few dozen or a few hundred people? Here is a straight answer, including the parts that usually get left out of the quote.

Why nobody gives you a number

Partly because it genuinely varies, and partly because the traditional model is paid by the day. If the engagement is priced in consultant-days, an accurate estimate is against the supplier interest and a vague one costs them nothing. The result is a discovery phase that produces a roadmap and an invoice, and a client no closer to a working system.

The variation is real, though. A tool that drafts internal summaries from documents you already hold is a fundamentally different job from one that makes or influences decisions about people or money, because the second has to be explainable, auditable and monitored in a way the first does not.

The three cost bands we see in practice

Small, contained automations

A single well-defined job: extracting structured data from a recurring document type, triaging inbound enquiries, drafting first-pass responses that a human approves. The data usually already exists in one or two systems. These are weeks rather than months of work, and they are the right place to start if you have never done this before, because they prove the value before you commit to anything larger.

Systems that change how a team works

Shortlisting, case handling, bid production, quality assurance across a whole workflow. Multiple data sources, real integration work, an interface people actually use every day, and human review designed into the process rather than bolted on. Months rather than weeks. This is where most of the genuine return sits, and also where most of the failures happen when nobody owns the operational side afterwards.

Systems with regulatory or safety weight

Anything influencing decisions about employment, credit, eligibility, health or safety. Everything from the second band, plus explainability, bias monitoring, full audit trails and documentation designed in from the start. It costs more because it has to survive being questioned, and building it any other way is a false economy.

The costs that do not appear in the quote

The build is rarely the expensive part. What catches people out is everything around it.

Data preparation is the big one. Almost every business believes its data is worse than it is, and almost every business underestimates how much work goes into making it consistent enough to rely on. Integration is the second: connecting to a CRM, a finance system and a document store is usually more effort than the model work itself.

Then there is the running cost, which never stops. Hosting and inference, monitoring, and — most importantly — someone improving the thing as your business changes. A system that was right in January and untouched by November is quietly getting worse. If your supplier hands over a build and walks away, that upkeep becomes a job for a team you probably do not have.

And finally, change. The best system in the world returns nothing if people carry on doing it the old way. Budget for the rollout, the training and the fortnight where everything feels slower.

A more useful question than what does it cost

Ask instead: what is the current thing costing? Take one job you would like AI to handle. Count the hours it consumes each week, the salary cost of those hours, and the opportunities you turn down because those hours are spoken for. That number is your ceiling for what solving it is worth, and it is usually much larger than people expect once the turned-down work is included.

Then ask what happens if it goes wrong. A drafting assistant that produces a poor paragraph costs a few minutes. A system that quietly mis-ranks candidates or misprices risk costs considerably more. That answer tells you which of the three bands you are in, and therefore roughly what you should expect to spend.

How we price it

We quote a fixed price for a defined outcome, agreed before the work starts, and a monthly fee to run and keep improving it. If the build turns out harder than we expected, that is ours to absorb. We do that because it forces us to understand the problem properly before quoting, and because it puts the risk with the party best placed to manage it.

Before any of that, we will tell you free of charge whether we think it will pay off at all. If the honest answer is no, you will hear it from us before you have spent anything.

The practical next step

Pick one job. Not a strategy, not a roadmap — one job you would happily never do again. Bring it to a free thirty-minute AI-readiness teardown and we will take it apart with you: whether AI is the answer, where it would start, what it would take and what it would be worth. No slides.

Related reading

• The Dashboard Trap
• The Data Puddle That Works
• Machine Learning Beyond the Chatbot
• Your Data is Scattered Across a Dozen Systems

How we can help

AI Applied is a UK software studio building production AI systems. See our AI services, including AI readiness assessment, AI implementation, machine learning development, data engineering, AI automation and AI governance and compliance. Or just get in touch.