Most AI implementation shortlists are built from a search, a directory listing and two recommendations, and the three firms that end up in the room turn out to be doing different jobs. One writes a strategy. One sells licences and configuration. One writes production code. They quote different numbers for what sounds like the same project, and the buyer ends up comparing prices for work that is not comparable.
This post does two things. It sets out ten UK consultancies a mid-market B2B buyer would realistically consider, with what each firm publishes about itself, and it gives you the method for cutting that to three and testing them properly.
Last checked: 12 Sep 2026. Every row below was read on the firm's own website on that date. Where a firm does not publish a price, the table says "not published" rather than guessing. Prices are as each firm states them, in pounds, and several are explicitly quoted excluding VAT.
Ten UK AI implementation consultancies, as they describe themselves
| Firm | Best fit | Published prices | Delivery model | Credentials, as stated | Watch-outs |
|---|---|---|---|---|---|
| Fifty One Degrees (51d.co) | Mid-market firms wanting an agent in production. States it is "built specifically for the mid-market, typically £5m–£250m in turnover (or the dollar equivalent), across the UK and US" | Yes. "A proof of concept typically costs under £15,000 and ships in 2–4 weeks." "Claude implementation packages run from £25,000 (Claude Ready) to £50,000–£70,000 (Claude Integrated)" | Build. States an "Embed Over Advise model" whose deliverable is "a production system in your environment, not a strategy deck" | OpenAI Select Partner; Attio Expert Partner | The published ladder is Claude-shaped. If you want a model-neutral recommendation, ask how the choice is made before you scope |
| OpenKit | Regulated and process-heavy organisations that want an audit first. Names education, legal, healthcare, financial services, manufacturing, recruitment, insurance and public sector | Yes. AI Audit "from £10,000"; AI Charter add-on "£3,000"; Embedded AI Lead "from £5,000 a month". "Fees exclude VAT" | Both. Audit and roadmap, then an embedded lead who builds "the automations and connectors" | ISO 27001 and ISO 9001 (certified Nov 2024); Cyber Essentials | The build capacity arrives through a monthly embedded role rather than a fixed-scope project. Agree deliverables and an exit point, or the retainer becomes the project |
| Scott Logic | Large, regulated engineering programmes. Heritage stated as "capital markets and investment banks", plus government and "Energy, Commodities & Utilities" | Not published | Both. Architecture and tech strategy alongside software engineering and delivery | "six offices across the UK"; states a heritage in "highly regulated industries" | Built for enterprise programme scale. A mid-market buyer should expect a scoping conversation before any number, and should check the size of the smallest engagement the firm takes |
| CGI UK (formerly BJSS) | Enterprise and public-sector transformation across a very wide sector list, including defence, health and care, central and local government | Not published | Both. Advisory, business consulting, agile digital services, data analytics and managed IT services | States "94,000 professionals" and describes itself as "among the largest IT and business consulting services firms in the world" | bjss.com now returns a 301 redirect to CGI's UK site (checked 12 Sep 2026). If BJSS is on your list from an older recommendation, confirm which entity contracts, which team delivers, and whether the people you met are still on it |
| AND Digital | Digital product and platform builds where you also want your own team upskilled. States it works with "multi-national businesses right through to smaller scale-ups" | Not published | Both. A "Guide, Build, Equip" model, with Equip being capability transfer to your people | 1,600 employees in 25 "clubs" of roughly 90; partnerships with AWS, Google Cloud, Microsoft Azure, Snowflake and Databricks; ISO certified | The capability-transfer element is real work and real cost. Decide up front whether you are buying a system, a trained team, or both, because the price differs |
| Faculty | Applied AI and data science at scale, particularly public sector, defence, energy and health. Names 13 sectors | Not published | Both, weighted to applied AI and data science rather than business-systems integration | Homepage carries a quote from OpenAI's chief executive on the firm's red-teaming contribution. Case studies name NHS demand forecasting, NESO network stability and Dstl | The public reference work is government and enterprise scale. Ask specifically what a mid-market commercial engagement looks like and who staffs it |
| SoftBlues | Mid-market firms automating high-volume processes. States its aim is AI "in production, not in a slide deck" | Not published. States "Production in about 90 days, fixed price" and "Money back if the proof of concept fails", with figures quoted after a discovery call | Both. Strategy and roadmap in weeks 1–3, deployment weeks 4–10 | "Registered Anthropic Partner Network member"; Google Cloud Partner; "Registered Microsoft partner". London registered address (71–75 Shelton Street, WC2H 9JQ) | A money-back proof-of-concept guarantee is only as good as the written success criteria. Get the pass/fail test into the contract before you rely on it |
| The AI Consultancy | UK SMEs, scale-ups and mid-market firms wanting a graded ladder from diagnostic to build | Yes, in detail. "Readiness Sprint, from £3,500"; "Discovery and Pilot, from £15,000"; "Build and Embed, from £40,000"; day rate "£950 to £1,500"; fractional CAIO from £3,000 per month | Both. Diagnostic, then prototype, then "full production build, integration with existing systems" | States Anthropic Consulting Partner status and AWS, Google Cloud and Nvidia certifications. London office at 20 Wenlock Road, N1 7GU | The published ranges are wide (Build and Embed is quoted from £40,000 to £250,000 depending on scale). Pin the scope to a fixed figure before comparing it to anyone else's |
| Nexa Cognition | Mid-market and enterprise firms whose AI work sits on top of HubSpot, RevOps and marketing operations | Not published. Offers a quote calculator | Both. HubSpot implementation, CRM optimisation, integrations and migrations alongside AI services | "Accredited in the top 1% of global HubSpot partners"; HubSpot Elite; CRM Implementation, Solutions Architecture and Onboarding accreditations; "100+ Digital Experts"; five global offices including London, Birmingham and Dubai | The service list spans AI, CRM and digital marketing. Ask which team does the engineering and what share of the firm they are |
| SpotDev | UK B2B companies from around £3m revenue upwards that have outgrown configuration and no-code workarounds, particularly where the AI has to act inside HubSpot and the systems around it | Yes. AI and Data Readiness Assessment "Starts at £5,000 with 20% credited to an implementation project"; AI Foundations £8,000; Custom Agents £20,000; AI Transformation £45,000 | Build, with a paid diagnostic in front of it. An AI and digital transformation consultancy that builds real software | HubSpot Diamond Solutions Partner; Custom Integration Accredited; Onboarding Accredited; OpenAI Select Partner; a Claude Registered Partner; Cyber Essentials Plus; 5.0 across 38 reviews on the HubSpot Solutions Directory | Deepest where the work touches HubSpot and connected business systems. We do not run a research-grade data science practice, so a bespoke model-training programme is not our strongest fit. The team is fully remote, so there is no office to visit |
Two notes on what is not in the table. Aiken House publishes a widely cited consultancy comparison, but the firm itself describes itself as "Pittsburgh's only Venture Studio", so it is not a UK shortlist candidate. And several firms that appear on "top AI consultancies" listicles publish no prices at all, which is not a fault, but it does mean a directory ranking tells you nothing about budget fit.
Define the metric before you pick the model
The most expensive mistake in AI buying is starting with the technology. A shortlist conversation that opens with which model a firm prefers will produce a proposal built around that model, and you will have no way of telling whether it worked.
Write down the metric first, in one sentence, with a number and an owner. "Reduce the time from quote sent to quote chased from six days to one, measured on the deal record, owned by the sales director." That sentence does four things at once: it names a business outcome, it names where the evidence will come from, it puts a person's name on it, and it gives every firm the same brief to quote against.
Then let the firms argue about the model. A consultancy that can explain why a cheaper model handles the routine volume and a frontier model handles the hard cases, and can show you the cost difference, is telling you something useful. A consultancy that leads with a vendor logo is telling you about its commercial arrangements. Our own view on how that choice should be made is set out on our AI implementation page, and in more detail in Claude versus ChatGPT for business.
The procurement test: six questions
Ask all six of every firm on the shortlist, in writing, and compare the answers side by side. They are deliberately awkward.
- Who writes the code, and where do they sit? Name the individuals who will be on your project, their employer, and whether any part of the build is subcontracted. Ask for this before the proposal, not after.
- What share of the delivery team is engineering? A firm that is 80% consultants and 20% engineers is a strategy practice with a build function attached, which is a legitimate model but a different purchase. Get the number.
- Can I speak to a client whose system is in production now? Not a pilot, not a proof of concept, not an award entry. A named reference running the thing daily, who will take a call. This single question removes more firms from a shortlist than any other.
- Will you give me a fixed price and a fixed date? If the answer is a day rate and an estimate, you are carrying the delivery risk. That may be acceptable for genuinely open-ended research. It is not acceptable for a defined workflow.
- Who owns the intellectual property, the prompts and the trained configuration? Get it in the contract. Ask specifically about prompts, agent definitions and any fine-tuned artefacts, because these are often silent in a standard software IP clause.
- What happens after launch, and what does it cost? Model versions change, vendors retire endpoints, and an agent that worked in September can behave differently in December. Ask what the support arrangement covers, what it costs per month, and what counts as a fix rather than a new piece of work.
A scoring sheet you can actually use
Score each firm out of five on each line, multiply by the weight, and total. The weights below suit a mid-market buyer with a defined workflow problem. Change them if your situation differs, but change them before you see the proposals, not after.
| Criterion | Weight | What a five looks like |
|---|---|---|
| Production evidence | 25% | A named, contactable reference running a comparable system in production today |
| Engineering depth | 20% | Named in-house engineers on your project, with a stated engineering share of the delivery team |
| Commercial clarity | 15% | Fixed price, fixed date, written scope, and a change-control process you understand |
| Fit with your systems | 15% | Demonstrated work in your CRM, ERP or data platform, not a generic integration claim |
| Governance and security | 15% | Named certifications, a DPA, a data-residency answer, and a human approval step where the AI acts on something that matters |
| Post-launch support | 10% | A priced support arrangement with a response time and a definition of what it covers |
Run three firms through it, not five. Five briefings cost you more internal time than the difference between the third and fifth choice is worth.
Buy a two-week paid discovery, not a project
The sensible first engagement is a short, paid, fixed-price discovery with a written deliverable, bought from your preferred firm after the scoring exercise. Two weeks is usually enough. You are buying four things: a mapped current-state process, a technical assessment of the data and systems the work depends on, a fixed-price proposal for the build, and an honest answer on whether the project is worth doing at all.
Pay for it. A free scoping exercise is a sales activity, and it will be scoped to produce a proposal rather than a judgement. A paid one you can cancel after, having kept the document. Several firms on the list above credit some or all of the discovery fee against the build if you proceed, which is the arrangement to ask for.
Two weeks of paid discovery also tests the firm in the only way that matters before you are committed: you find out how they run a deadline, how they write, whether they ask your finance team the awkward data question, and whether the engineer you were promised turns up.
If you want a starting point that is not a sales call, our diagnostics set out what we look at first, and the HubSpot agency pages cover the systems side where most of this work actually lands. When you are ready for a number, request a quote.
Common questions about shortlisting an AI consultancy
How many firms should be on the shortlist?
Three. Send all three the same one-page brief with the metric, the systems involved and the deadline, and compare the responses against a scoring sheet you wrote before you read them. Five briefings cost more internal time than the extra choice is worth.
Why do so few UK AI consultancies publish prices?
Of the ten firms checked on 12 Sep 2026, four publish figures (Fifty One Degrees, OpenKit, The AI Consultancy and SpotDev) and six do not. Firms selling enterprise programmes generally price per engagement; firms selling repeatable mid-market work are more likely to publish a ladder. Absence of a published price is not a red flag, but it does mean you cannot filter on budget without a call.
Should I hire a strategy firm or a build firm?
If you already know the workflow you want to fix, hire a firm whose deliverable is a working system, and check the engineering share of the delivery team. If you genuinely do not know where AI helps, a paid audit or readiness assessment is the cheaper first purchase. Buying strategy when you needed a build is how organisations end up with a roadmap and no change.
What should a first AI engagement cost?
Based on published prices on 12 Sep 2026, a diagnostic or readiness assessment runs from about £3,500 to £10,000, a proof of concept or pilot from about £15,000, and a first production build from roughly £20,000 to £70,000 depending on how many systems it touches. Figures above about £90,000 generally indicate enterprise scope rather than a single workflow.
Who should own the AI project internally?
A named business owner who owns the metric, not IT and not a steering group. The owner needs authority over the process being changed, because most of the difficulty in these projects is process and data quality rather than the model.
How do I check a consultancy's partner claims?
Ask which programme, which tier and when it was awarded, then check it against the vendor's own directory where one exists. Partner language varies between vendors and between tiers within the same vendor, so a badge on a website is a starting point for a question rather than an answer.
Stay Updated with Our Latest Insights
Get expert HubSpot tips and integration strategies delivered to your inbox.
