What AI and IT consultancy costs in the UK
Pricing in this market is opaque mostly because the work is genuinely variable, but the structure underneath it is not complicated. Here is how quotes are built, what makes them move, and where the cost you were not shown usually hides.
8 min read · Updated 6 September 2026 · AiXL Consultancy Ltd.
1. The four ways this work is priced
Almost every proposal you receive is one of these, or a hybrid.
- Fixed scope, fixed price. A defined outcome for a defined number. Best when the problem is well understood. The risk transfers to the supplier, so expect the price to carry a margin for that, and expect any change to be a change request.
- Time and materials. A day rate against an estimate. Flexible, honest about uncertainty, and dangerous without a cap and a weekly burn report.
- Retained capacity. An agreed number of days a month. Sensible once something is live and evolving, wasteful before you know what you want.
- Discovery first, then decide. A short paid piece of work that produces a costed plan. Usually the cheapest way to avoid a large mistake, and the one we recommend when the problem is still fuzzy.
Day rates in the UK vary by roughly a factor of five between an independent contractor and a global consultancy partner. Rate alone tells you very little: what matters is days multiplied by rate, and whether the days are being spent by someone who has solved this before.
2. What actually drives the number
In rough order of impact.
The state of your data. This is almost always the largest single variable. Clean, accessible, well-permissioned data can halve a project. Data spread across systems with no consistent identifier, or locked behind an API nobody has credentials for, can double it. Any estimate produced before someone has looked at the actual data is a guess.
How wrong it is allowed to be. A drafting assistant a human reviews is a different engineering problem from a system that writes to a ledger unattended. Evaluation, guardrails, approval steps and audit trails are most of the difference between a demo and something you can put in production.
Integration surface. Reading from one system is cheap. Writing back to three, each with its own auth model and change window, is where the days go.
Regulatory load. A DPIA, data residency requirements, retention rules and evidence for an auditor are real work. They are also much cheaper designed in at the start than retrofitted after a risk review, which is the argument for treating privacy and security as part of the build rather than a gate at the end.
How many people have to agree. Rarely on the invoice, frequently on the timeline.
3. The cost most quotes leave out
Build is a one-off. Running it is not, and the annual figure often surprises people more than the project did.
- Inference. Scales with usage, and usage grows if the thing is any good. Ask for a cost per transaction, not a monthly total, so you can model what success costs.
- Hosting and data stores. Vector indexes and document storage are cheap individually and add up quietly.
- Monitoring and evaluation. Quality drifts as your sources change. Detecting that needs an ongoing evaluation harness, not a one-time test.
- Human supervision. Somebody reviews the exceptions. That is a real, recurring cost and it belongs in the business case from the start.
- Model churn. Providers deprecate and reprice. Budget for a re-validation at least annually.
Ask for a twelve-month total cost of ownership alongside the build price. A supplier who cannot produce one has not run something in production for long.
4. Comparing two quotes that look nothing alike
Normalise before you compare. For each proposal, write down: what is delivered at the end, who does the work, what the twelve-month running cost is, what happens when scope changes, what you own, and what the exit looks like if you stop.
A quote that is materially cheaper is usually cheaper for a reason worth finding: a smaller scope, more junior staff, no evaluation harness, no handover, or an assumption about your data that has not been checked. None of those make it the wrong choice, but they should be a decision rather than a surprise.
5. How we price
We quote a fixed price against a fixed scope wherever the problem is understood well enough to make that honest, and we propose a short paid discovery when it is not. We give a twelve-month running cost with every build proposal. If we think the return does not justify the work, we say so before you commit, which occasionally costs us the engagement and has never cost us a client.
Tell us what you are trying to move and we will come back with a scope and a number.
6. Questions
How much does an AI consultant cost in the UK?
UK day rates span roughly a factor of five, from an independent contractor to a partner at a global firm. The rate matters less than the total: days multiplied by rate, plus the twelve-month cost of running whatever gets built. Ask for both figures, and ask who specifically will be doing the work at that rate.
Is fixed price or time and materials better?
Fixed price suits a well-understood problem and moves risk to the supplier, at the cost of a margin for that risk and a change-request process. Time and materials suits genuine uncertainty but needs a cap and a weekly burn report. When the problem is still fuzzy, a short paid discovery that produces a costed plan is usually cheaper than either.
What does it cost to run an AI system after it is built?
Model inference, hosting and data stores, monitoring and evaluation, human review of exceptions, and periodic re-validation as providers change models. Inference scales with usage, so ask for a cost per transaction rather than a monthly figure. Request a twelve-month total cost of ownership alongside any build price.
Why do quotes for the same project differ so much?
Usually scope, seniority or omission. A cheaper quote may cover less, use more junior engineers, exclude evaluation and handover, or assume your data is in better shape than it is. Normalise the proposals on deliverables, who does the work, twelve-month running cost, change process and what you own at the end, then compare.