What good AI implementation actually costs, honestly

Most people who ask what AI implementation costs get a non-answer: it depends, let's talk. It does depend, but not so much that I cannot give you real numbers. So here they are, along with where the money actually goes and how to tell when a quote has been padded.

The actual ranges

I price projects as fixed fees, not day rates, and they fall into three bands. An AI Business Transformation, where I rework a core process end to end, is typically £15,000 to £30,000. A Custom AI Product Build, a bespoke tool or feature built into your business, is usually £20,000 to £40,000. And Implementation Days, for focused ad-hoc work on a problem you already understand, are £2,500 to £3,500 a day. Those are the numbers. If your project is a normal one for a business turning over between one and twenty million pounds, it almost certainly sits inside that range.

Why a fixed price, not a day rate

A day rate makes the cost open-ended and quietly puts the risk on you. The longer it takes, the more you pay, so the incentives point the wrong way. A fixed price moves that risk to me, where it belongs. You get one number and a delivery date before any work starts, and you can budget with certainty. It also forces me to scope the work properly up front, which is where most of the value of the first conversation actually sits.

A fair quote is mostly the work. An inflated one is mostly the overhead.

Where the money actually goes

Not where most people expect. The cost is rarely the AI itself, because the model is cheap. The cost is judgement and the last mile. It goes on understanding your business well enough to pick the right thing to build in the first place, on the unglamorous engineering that turns a promising demo into something dependable, on the integration with the systems you already run, and on getting your people to actually use it. That last mile is most of the work and most of the value, which is exactly why it is the first thing a cheaper quote quietly leaves out.

What moves the price within the range

A handful of things. How much integration the work needs with your existing systems. How ready your data is, against how much of it has to be cleaned or organised before anything can be built. How costly a mistake would be, because a higher cost of error means more careful design and more human checks. And how well-defined the problem is when we start, because a vague brief costs more than a sharp one: part of the fee goes on making it sharp.

How to tell when a quote is inflated

Watch for a few things. A day rate with no ceiling, so the total can drift upward with no natural stop. A senior person selling the work and a junior team you never meet delivering it, which is the Handoff Tax, and you pay it in both money and quality. A large "discovery phase" that produces a document rather than working software. And padding that comes straight from an organisation chart: account managers, project managers and layers of oversight that exist to run the agency, not to build your thing. Strip those away and you are left with the actual work, which is what you should be paying for.

When it costs less, and when it costs more

Less: a single, well-understood problem with clean inputs and low risk can often be an Implementation Day or two rather than a full project. More: genuinely large programmes, heavy regulatory constraints, or messy data that needs real work before anything can be built on top of it. I will tell you honestly which of these you are, because quoting you into the wrong band helps neither of us.

The honest summary

For most AI implementation in a mid-sized UK business, the real answer is fifteen to forty thousand pounds, fixed, with the money going on judgement and the last mile rather than the model. If a quote is a long way above that, ask where the extra is going. If it is a long way below, ask what is being left out. Either way, you should be able to see exactly what you are paying for. That clarity is the whole point of a fixed price, and it is a fair test to hold any quote up to, including mine.