How much does an AI consultant cost in the UK? (2026)

Start here

The short answer.

A first paid engagement with a UK AI consultant runs from about £999 to £10,000, and AGMM charges £2,000 for a fixed fourteen-day audit. Build projects on the same firms' pages run £10,000 to £150,000, and retainers £200 to £20,000 a month.

Those are the numbers six UK firms print on their own websites, re-read on 16 September 2026. Not a survey and not an average: the small minority of this market that commits to a figure in public. If only one of your three quotes arrived with a number attached before the sales call, you are not comparing three of the same thing.

The index

What do six UK firms actually charge?

A price index, not a review. It makes no claim about the quality of anyone's work, including AGMM's. Every figure was read from the page it links to, and every link was re-fetched on 16 September 2026.

Prices re-read on 16 September 2026; next check December 2026.

Every figure above is a firm's own published price, not survey data. Where this page quotes a claim about the whole market it names whose claim it is: the only one here is Hartz AI's, who write that UK consultant rates run £120 to £400 an hour. That is a vendor describing its own market, not a measurement.

Public sector buyers get more. The G-Cloud 14 supplier guidance makes a published rate card a condition of entry, and bans "prices from".

Pricing models

Day rate, fixed price or retainer: what does each do to your risk?

The model decides who pays when the work runs long, which is a bigger question than the headline number.

Day rate

You buy time. If the job takes twice as long you pay twice, and the vendor carries no delivery risk. That is why a day rate is usually the cheapest-looking option and often the most expensive one.

Fixed project price

The vendor carries the overrun and prices that risk in, so a fixed quote against a vague scope always looks expensive. It is only a fair comparison when all three quote the same written scope.

Retainer

You buy availability rather than a delivered thing. Published UK retainers run £200 a month for a pot of engineer time, £5,000 for an embedded lead, £20,000 for a programme. Right once systems are live; wrong as a first purchase.

Our own number

Why do we publish £2,000 instead of a day rate?

Because a day rate prices our time and you are not buying our time. You are buying a decision about where money is getting stuck, worth the same whether it takes us nine days or fourteen.

A day rate also puts both sides on opposite sides of the clock: every hour we save you costs us money, so the number that ought to be settled before the work starts stays open until the invoice. A fixed fee closes it, and if the work runs long that is our problem.

Publishing it costs us the ability to size a quote to whoever is asking, which is the real reason most firms do not. Fair trade for letting you compare us against five other firms before you speak to anybody.

Cheap and expensive diagnostics

What does a £2,000 audit contain that a £750 one does not?

Sometimes nothing, and it is worth saying plainly: a £750 audit from someone who knows your sector can beat £2,000 spent badly. The difference is scope, and scope is checkable.

The AGMM audit is 14 days, built around three sixty-minute sessions with the engineers who would scope the build, and it ends with four documents you own outright:

• The constraint map. A business-wide view of where capacity, cash and decisions are held up.
• The AI opportunity map. The viable opportunities, the dependencies underneath them and where a person must remain in control.
• The economic cases. Upside, cost, assumptions and risk for every opportunity worth considering.
• The ranked next step. A prioritised investment order and a custom proposal for the highest-value justified scope. You can use it with AGMM, take it elsewhere or stop.

The cheaper end typically buys one session and one report, which is a reasonable thing to buy. What it rarely buys is a business-wide view rather than the process you already suspected, an economic case per opportunity built from your own figures, and a build priced at the end. Compare those four, not the two numbers.

Full contents are on the AI constraint audit page, and the whole ladder is on what it costs.

Exclusions

What is never in the price.

Run this against every quote, including ours. The gaps between three quotes are usually here, not in the fee.

• Software and model costs. Licences, API usage and hosting are billed to you by the vendor, not marked up and resold. Ask every firm whether their figure includes these: the answer moves a comparison more than the fee does.
• Your team’s time. Three sixty-minute sessions, plus the hours spent pulling exports and answering questions. No consultancy carries that for you.
• The build. The audit prices the build. It does not include it, and the fee is not deducted from it.
• Data cleaning you already needed. Where records are inconsistent enough that no system could read them reliably, fixing that is its own piece of work, scoped separately rather than absorbed quietly.
• Training courses and change management. We hand over systems your team can run and show them how. A structured training programme is a different product, and other firms sell it better.

What moves the number

What drives a quote up?

Four things, in roughly this order. None is the technology, which is why a quote that moves when you name a different model was never priced on the work.

• How much of your data is machine-readable today. A process running through one system with consistent records is a smaller job than the same process running through a shared inbox, three spreadsheets and somebody’s memory. The largest mover, and measurable before anybody quotes.
• The integration surface. One system with a documented API is cheap. Four systems, one reached only by a nightly export, is not. Count the systems that have to agree and you have most of the price.
• Regulated sector requirements. Financial services, healthcare, legal and education carry audit trails, retention rules and approval steps that must be designed in rather than added later. Expect that to move a build price, not an audit fee.
• IR35 and how the work is contracted. A day rate bought through an intermediary sits inside the off-payroll rules, and the status determination is the client’s responsibility. A fixed-price contract for a defined deliverable carries a different administrative load. Price that difference.

Readiness matters first. DSIT's AI Adoption Research, published in January 2026, reports that among UK organisations planning to adopt AI, 34% feel ready, 33% are unsure and 32% say they are not, and that the most commonly cited barriers are a lack of identified need and limited AI skills.

Getting comparable quotes

How do you brief three vendors so the quotes are comparable?

Six steps, about an hour. They remove the advantage a vendor gets from a loose brief, including ours.

1. Write the process down as it runs today, including the step people complain about. One page is enough.
2. Name the outcome, not the technology. "Quotes out the same day" beats "an AI agent".
3. Give all three the same systems list, volumes and access constraints.
4. Ask for a fixed price against that written scope, and ask what happens to it when the scope changes.
5. Ask who does the work, and whether the person on the call writes the code.
6. Ask what you own at the end, in writing, and whether you keep it if you stop after stage one.

If two quotes land within ten per cent of each other on the same written scope, the number is probably about right. If they are three times apart, one has understood something the other has not: ask each what they think the hard part is.

The conflict nobody discloses

Why is the audit fee not credited against a build?

Because the moment the fee comes off the build, the audit has a financial reason to recommend one. A credited diagnostic is a discount voucher with a report attached: the firm is only really paid if it finds work for itself, and every reader of that report knows it.

AGMM's £2,000 is not deducted from anything. We are paid the same whether the recommendation is a build, a cheaper process fix or nothing at all, which is the only arrangement that makes "we do not think you should build this" worth reading.

Evidence

One number you can go and check.

Every firm here, AGMM included, asks you to believe a description of work you cannot see. Here is one piece of ours that is open: a real issued job, a first-floor build-over in Stockport, 63 measured lines, £98,214.72, produced by a quoting system we built and published in full.

It is not a case study and claims no saving. It is an artefact with an address: read all 63 lines of it. Ask each vendor you brief for something equivalent.

Noise

What to ignore in this market.

• Percentage claims with no baseline. "40% faster" means nothing without the before figure and who measured it.
• "Best AI consultancy" lists. Several are paid placements, and at least one ranking for the term is a referral site taking a fee when you engage a listed firm. That order is a commercial arrangement, not a signal.
• Case studies with no client and no artefact. "A leading UK manufacturer" is not evidence. Something you can open is.
• Free strategy sessions that arrive at a build. An unpaid diagnostic must convert to pay for itself, so the recommendation is decided before the meeting.

When we are the wrong answer

Use somebody else in these cases.

• MQLFlow. Buyers who want to purchase time by the day rather than a scoped outcome.
• Prime AI Solutions. Buyers who want the cheapest first diagnostic, with the savings figure underwritten rather than measured.
• OpenKit. Buyers who need a larger delivery team than AGMM runs.
• Fifty One Degrees. Buyers whose procurement requires named client references with published metrics, which AGMM does not yet have.
• Helium42. Mid-market and enterprise programmes beyond a single constraint.
• Nobody, yet. If the constraint is a pricing decision, a staffing decision or a process nobody has written down, software on top makes the problem faster rather than smaller.

Considered and left out

Who was considered and left out, and why?

A list that only shows what the compiler wanted you to see is not a comparison. These were considered and excluded, with the reason in each case. The first two were in the table until 16 September 2026, and came out when their figures stopped verifying.

• Elevate AI. In this table until 16 September 2026 with pilots from £3,000. Re-read that day, elevateai.co.uk/pricing carries no figures at all: fee, deliverables and schedule are now agreed before work begins. A firm that has withdrawn its price cannot sit in a table sorted by published price, so the row came out rather than the figure being carried on trust.
• KORIX. In this table until 16 September 2026 at £10,000 to £80,000. Their site serves a JavaScript shell: the HTML a screen reader or a crawler receives carries no prices, so there is nothing at that address to check. The row came out for that reason alone.
• Faculty. Enterprise and public sector scale. Publishes no price and is not competing for the same buyer as anyone else in the table.
• Deloitte, PwC, Accenture, EY. Published day rates start around £2,000 and typical programmes run £60,000 to £300,000 and up. A different purchase entirely.
• Aucta AI, Northloop, Automazen, ORYX, Fracas Digital, Softomate. All publish a comparison article naming other agencies, and none publishes its own starting price. They cannot be placed in a table sorted by published price.
• The no-code and template tier. Firms assembling n8n or Make templates for under about £3,000. A real option for a small one-off automation and a genuinely different product from a scoped build.
• TopTenAIAgents.co.uk. Not an agency. A referral site operated by Hello Leads Ltd which takes a fee when a reader engages a listed firm, so its ordering is not a price signal.