How much does AI consultancy cost in the UK? A buyer's guide

Reviewed October 2026. This guide explains how the market prices AI consultancy and what each kind of engagement contains, so you can compare proposals on scope. It does not publish our rates; a scoped conversation does that.

Contents: Why there is no list price · The four pricing models · The five common engagements · What drives the number · Who is quoting · Reading a proposal · Questions to ask · What good value looks like · Reducing the cost without reducing the result · A buyer’s checklist

Anyone who has asked three AI consultancies for a price has had three incomparable answers: a day rate, a fixed fee for a “sprint”, and a retainer. None is wrong; they describe different things. This guide explains the models, what the common engagements contain, and how to compare quotes on what you receive rather than what the invoice is called.

Why there is no list price

AI consultancy is priced on four things that vary between every client: how much of the work is thinking (strategy, discovery, prioritisation) and how much is building; how many people are needed and of what seniority; how much of the client’s own time and data is available, which decides how long everything takes; and how much risk the consultancy carries (fixed-price work carries more, and costs more). A published rate card answers none of these, which is why reputable firms scope before they quote and why a quote without a scope attached should be read as a guess.

The four pricing models

Time and materials (day rates). You pay for days used. Transparent, flexible, and the right model for work whose shape is not yet known, such as the early part of discovery, or for embedded engineering alongside your own team. The risk is open-endedness; cap it with a budget ceiling and a weekly report of days used.

Fixed price for a defined scope. A fee for a defined deliverable: a discovery programme, a prioritised portfolio, a pilot with agreed success criteria. Predictable, and it shifts the delivery risk to the consultancy, which is why it costs more than the equivalent days. The right model once scope is known. Watch for scope so loosely written that the fixed price buys less than you thought.

Retainer. A monthly fee for ongoing access, usually a fixed number of days or an agreed service. Suits businesses that want a fractional AI lead or continuous engineering support. The risk is paying for capacity you do not use; review quarterly.

Outcome-based. A fee tied to a measured result, such as a share of cost saved. Rare in AI consultancy because attribution is hard and the client controls most of the variables. Where offered, it is usually a hybrid: a reduced fixed fee plus a success element. Treat it as a signal of confidence rather than a free option.

Most engagements combine models: time and materials for the first exploratory weeks, fixed price for the defined programme that follows, a retainer for what comes after.

The five common engagements

What each kind of engagement contains, how long it takes, and what you should get at the end. Durations are typical for a mid-market business with a responsive sponsor.

1. Leadership session or awareness programme. Half a day to two days. Levels the leadership team’s understanding of what the technology can and cannot do, surfaces the business’s own opportunities in the room, and produces the one-sentence purpose and a first list of candidates. Deliverable: a short report and a prioritised list. The lowest-cost engagement and often the one with the highest return per pound, because it prevents the wrong programme.

2. Readiness and discovery. Two to eight weeks depending on the size of the business and the number of functions involved. Structured divergent-then-convergent discovery with each function, a data-readiness check against the candidate use cases, consolidation into a scored and sequenced portfolio. Deliverable: the portfolio, with each opportunity’s value, feasibility, data position and pyramid layer, and a recommended first phase. Our Converge Methodology is this engagement; the Nidec case study is an example of it.

3. Strategy and roadmap. Often combined with discovery. Turns the portfolio into a sequenced plan with governance, budget phases and decision gates. Deliverable: a roadmap a board can approve and an operating model for running it.

4. Pilot or proof of value. Four to twelve weeks. Builds one opportunity to the point where its value is measured on real data with real users, with an evaluation set and an agreed production path. Deliverable: a working system, the measured result, and a scoped plan for production. The engagement where most AI spend goes wrong, because it is often bought without the production path.

5. Build and run. Three months upward. Production engineering of an agent, assistant or model, with integration, evaluation, escalation design, monitoring and handover or managed operation. Priced as a project, then a retainer for operation. Deliverable: a running system with its evaluation set, runbooks and a named owner on each side.

A sixth category, tool rollout and training, is priced per cohort or per programme and sits alongside the others.

What drives the number

For a given engagement, these move the price most:

• Number of functions in scope. Discovery across three functions costs less than half of discovery across eight, because consolidation is where the value is and it scales well.
• Data position. Use cases whose data is reachable are cheap to pilot; use cases that need an extraction path first are not. A proposal that does not mention your data position has not looked.
• Integration depth. An assistant that reads documents is days; an agent that writes to your ERP, with the permissions and audit that implies, is months.
• Risk category. Anything near the EU AI Act’s high-risk list carries documentation, oversight design and testing that add real work.
• Seniority mix. Discovery and strategy are senior-heavy; build is engineer-heavy. A proposal should say who is on it.
• Your availability. The single biggest driver of elapsed time, and therefore of time-and-materials cost, is how quickly your people can be in the room and your data can be reached.
• Risk transfer. Fixed price costs more than the same days; you are buying certainty.

Who is quoting

The UK market has five kinds of supplier, and their prices reflect their cost bases rather than the value of the work.

Global consultancies and the big systems integrators. Large teams, strong at enterprise-scale programmes, priced for enterprise budgets. Often lead with a platform partnership.

Technology vendors’ professional-services arms. Priced to sell the platform; the consultancy is sometimes subsidised. Good value if you have already chosen the platform; not where to go for the choice.

Specialist AI consultancies. Senior-led, method-driven, usually platform-agnostic, priced between the integrators and the freelancers. Where most mid-market businesses land. We are one.

Digital agencies adding AI. Strong on customer-facing experiences, variable on data and engineering depth. Good for the customer-facing layer; check who does the engineering.

Freelancers and boutiques. Lowest rates, highest variance. Excellent for a well-defined build when you have found the right person; risky for discovery, which needs method and breadth.

Reading a proposal

Put the proposals side by side and compare these, not the totals.

Questions to ask

Ten questions that separate a scoped proposal from a guess:

1. What exactly will we hold in our hands at the end, and in what form?
2. Who will do the work, and how much of each person’s time?
3. What do you need from us, and when, for the timeline to hold?
4. What is not included that we will probably need?
5. How do you decide which opportunities to recommend, and can we see the method?
6. What happens if the pilot works? What if it does not?
7. Which platforms do you work with, and what is your relationship with them?
8. Who owns what you build, and what are the data terms?
9. Can we speak to a client for whom you did this kind of work?
10. What would you cut if we needed to spend a third less?

The last question is the most revealing. A consultancy that knows its own method can tell you immediately what is essential and what is nice to have.

What good value looks like

Good value is not the lowest total. It is a first engagement small enough to be a low-regret decision, with a deliverable that makes the second decision well-informed. The pattern that works: a leadership session or a short discovery first, a decision gate, then a pilot of the top opportunity with its production path scoped, then a build. Each step is funded on the evidence of the last. The pattern that fails: a large programme bought up front on a business case nobody can yet substantiate, or a cheap pilot bought with no thought for what follows.

A useful test: at the end of the first engagement, could you take its deliverable to a different supplier and have them act on it? If yes, the deliverable has value in itself. If the deliverable only makes sense as a prelude to the same firm’s next proposal, you have bought a sales process.

Reducing the cost without reducing the result

• Do the readiness work yourself. The twelve-question checklist is free and its answers shorten discovery.
• Bring your own domain experts, with time freed. Every hour of expert time you provide is an hour of consultancy time you do not buy, and the experts learn the method.
• Scope narrowly and sequence. Three functions this quarter beats eight functions badly.
• Choose first projects where the data is ready. The data position drives build cost more than anything else.
• Insist on an evaluation set from day one. It is cheaper to build at the start than to retrofit, and it is what makes the pilot’s result believable.
• Buy capability transfer. Ask that the engagement leave your people able to run the next cycle; it costs little extra and changes the economics of year two.
• Use the enterprise assistants you already licence. Much of the base layer of the pyramid is configuration and training, not consultancy.

A buyer’s checklist

Before you sign:

• The deliverables are named and described in a way a third party could act on.
• The team is named with seniority and days.
• Our obligations are written down and we can meet them.
• Assumptions and exclusions are listed; integration, licences, data extraction and hosting are addressed.
• The engagement has a decision gate and we know what we will decide.
• The production path for any pilot is scoped, even if not yet priced.
• IP, data and platform terms suit us.
• We have spoken to a reference.
• The first engagement is small enough that being wrong is affordable.

If you would like a scoped proposal rather than a price list, our AI strategy consultancy page describes the engagements above and the approach page shows the method they are built on. A scoping conversation costs nothing and usually takes an hour.