How much does AI implementation cost in the UK.
The search is for a number. The honest answer is three different purchases wearing the same headline: a seat licence, a first workflow, and a system that runs in the business's own tools.
"How much does AI implementation cost?" has the same problem as "how much does a kitchen cost?". Without knowing whether you mean a kettle, a refit or a restaurant, any figure is a guess dressed up as an answer. So here is the split that actually helps a UK business decide.
How much does AI implementation cost in the UK?
The honest answer has layers. Trying it yourself costs roughly £19-20 a month for a personal assistant seat, or nothing on a free tier. For delivered work, Clerq publishes its prices: an assessment at £799, a process redesign at £2,500, a knowledge system at £2,000, one workflow build at £2,500 to £4,000, and full implementation at £4,000 to £8,000+. Where the price is a range, the figure is fixed in writing once the workflow, systems and sign-off are named.
A ChatGPT tab and a month-end close that posts to the ledger are not the same purchase. Confusing them is how businesses either overpay for a licence they never use, or under-scope a build and then call AI a failure.
Why most cost answers are useless
Search the phrase and the results quote everything from a few pounds a month to seven figures. Every one of those figures is true for somebody and wrong for you. The number means nothing until you know three things: who is paying, whether anything is being wired into the organisation's own systems, and who owns the exception when the output is wrong.
There is a second reason people struggle to price this: most of us have no reference point. DSIT's AI Adoption Research, published in January 2026, found that 60% of surveyed UK businesses reported limited AI skills, expertise and knowledge as a barrier to adoption. If you have never run the workflow, you cannot tell whether a price is fair, and that is exactly the condition in which vague pricing thrives. The wider picture is in our roundup of UK AI adoption statistics.
Three layers, not one number
Whatever the marketing calls its packages, implementation cost in a UK business resolves into three layers. Budget against the layer you are actually buying.
Trying it yourself
A seat licence, not an implementation. Free tiers of the mainstream assistants handle drafting, rewriting, summarising and explaining. A paid personal plan sits at roughly £19-20 a month, with cheaper entry tiers from about £5-7, and buys better models, longer documents, higher limits and file uploads. Check the vendor's current page before you budget: this is the fastest-moving number here. The real cost at this layer is still your time - learning to brief it, and checking every output. That is the experiment set out in how to use AI at work, and it is the right first spend.
A first bounded workflow
One recurring process, one owner, in tools you already have. Nothing posts to the ledger without a person. The spend is mostly people's time: a baseline of what the work costs today, a brief that an assistant can actually use, a check that catches the errors, and a week of running it in parallel with the old way. The software line is still a seat or two. The cost that matters is the hours, and whether those hours come back. How to pick that first workflow is in where to start with AI in your business.
A production system
The thing this URL is actually asking about. The workflow runs in the business's own tools, around existing approvals, with an audit trail and a named owner for exceptions. It is handed over working, not left as a slide deck. Cost here is baseline, build, integration, controls, review and the first months of tuning. It is not the model licence. A number published before those things are named is not a quote - it is marketing. A range is honest where a single number is not: Clerq publishes £2,500 to £4,000 for one workflow build and £4,000 to £8,000+ for a full implementation, with the exact figure fixed in writing once the workflow, systems and sign-off are named. The 90-day shape of a controlled build is in our implementation roadmap.
What actually moves the number
Two businesses can ask for "AI on month-end" and be an order of magnitude apart. Five things explain most of the gap.
- Does it write back to a system of record? A draft in a side window is cheap. A posting into Xero, Sage or a warehouse system is not, because the control and rollback work travels with it. See AI around the ledger, not instead of it.
- How many systems have to talk? One spreadsheet is a brief. Five systems with different identifiers is a reconciliation problem before it is an AI problem.
- How messy are the exceptions? High-volume, rule-stable work compresses. Judgement-heavy, one-off work does not, and pretending otherwise is how a build overruns.
- Who signs off? A workflow with a named reviewer and a written "working" test is cheaper than one that waits for a committee to decide after launch.
- Is the data already in one place? If the first month is spent finding the files, you are paying for housekeeping, not AI.
What a credible quote has to name
If you are being asked to approve a build, the quote is only useful if it can say these five things in writing, before anyone starts.
- The workflow - one process, not "AI for finance".
- The systems it will sit in, and what it is allowed to write.
- The controls, and who owns an exception.
- What "working" means, and the measure that will prove it.
- What is out of scope.
A quote that cannot name those five is not a quote. It is an invitation to discover the real cost later, on your time. How to turn that into a payback number you can defend is in AI ROI for UK SMEs.
The costs that never appear on the invoice
Three lines turn up reliably on implementations and on seat rollouts alike. Name them before you decide the spend is working.
- Checking time. Reviewing output is a permanent running cost. If a task takes longer to check than it took to do, it is not a candidate.
- Tool sprawl. Three overlapping subscriptions, each billed to a different card. Audit the list every few months and cancel the ones you cannot describe.
- Rework. Something plausible but wrong goes out unchecked, and an afternoon disappears undoing it. One of those wipes out a year of licence savings.
How to keep the spend honest
You do not need a procurement process to find out whether AI is useful. You do need a decision rule, or the charge quietly becomes permanent.
- A fortnight on the free tier. Two real tasks, times noted. If it does not obviously help there, a paid plan and a build will not rescue it.
- One workflow, one owner. Add a second only after the first has a baseline, a check and a week of parallel running.
- A review at 90 days. Say out loud what it saved, against the measure you wrote down at the start. If you cannot name the tasks and roughly the hours, stop.
Frequently asked questions
How much does AI implementation cost in the UK?
The honest answer has layers. Trying it yourself costs roughly £19-20 a month for a personal assistant seat, or nothing on a free tier. For delivered work, Clerq publishes its prices: an assessment at £799, a process redesign at £2,500, a knowledge system at £2,000, one workflow build at £2,500 to £4,000, and full implementation at £4,000 to £8,000+. Where the price is a range, the figure is fixed in writing once the workflow, systems and sign-off are named.
Is a ChatGPT subscription the same as implementation?
No. A subscription is a seat licence for one person in a browser tab. Implementation is a named workflow running in the business's own tools, with approvals, an audit trail and a human owner. They are different purchases wearing the same headline.
What does Clerq charge for implementation?
Clerq publishes its service menu: an assessment at £799, a process redesign at £2,500, a knowledge system at £2,000, one workflow build at £2,500 to £4,000, full implementation at £4,000 to £8,000+, and a concierge service at £750 to £1,500 a month for existing clients. Ranged figures are fixed in writing before any work starts.
What should I ask for in an implementation quote?
The workflow, the systems it will sit in, the controls and who owns exceptions, what working means and how it will be measured, and what is out of scope. If a quote cannot name those five things, it is not a quote.
What does it cost to try AI before implementing anything?
For one person on their own work, roughly nothing to about £25 a month. The free tier of a mainstream assistant covers a great deal of drafting, summarising and explaining. A paid personal plan is typically around £19-20 a month. Check the vendor's current pricing page before you budget.
Sources and how to read the prices
This article is editorial guidance, not a quotation. The personal-plan ranges describe publicly advertised consumer plans for mainstream AI assistants in the UK as at July 2026; vendors change these frequently, so check the current pricing page before you budget. Clerq's own service prices are published on the homepage menu and quoted above as at September 2026. The external figure referenced above is supported by the following source.
Want this scoped for your business?
Clerq builds working AI systems inside UK businesses - in your tools, around your controls, handed over working. A few sentences are enough. The founder will reply personally. Every price is on the homepage menu, and there is no obligation.