AI agency services: build, deploy and run, with a consultancy's governance
An AI agency is a firm that designs, builds and operates AI systems for its clients: workflow automations, AI agents, chatbots, voice agents and integrations with the software a business already uses. What the client buys is a working system in production, not a report. The AI Consultancy does that work for UK businesses, with three questions settled in the same engagement: whether the system is worth building, what UK GDPR requires of it, and who runs it after launch.
We are a consultancy, and the name is deliberate. Advice and delivery sit in one engagement, led by a named senior consultant from scoping to handover. Builds start from GBP 3,000 for an off-the-shelf chatbot and GBP 5,000 for an AI workflow automation or a straightforward AI implementation. Larger work follows our published stages: Discovery and Pilot from GBP 15,000 and Build and Embed from GBP 40,000, all ex VAT. We are an Anthropic Consulting Partner with Claude Certified Architect practitioners, based at 20 Wenlock Road, London, N1 7GU, in Hoxton, and working with clients across the UK.
If you already know what you want built, go straight to what we build. If you are not yet sure what to build, or whether AI is the right answer, read agency or consultancy first.
What is an AI agency?
The term is used by three different kinds of firm. A business looking for someone to build and run AI for it usually wants the second or third, and it is worth knowing which one you are talking to before you share a brief.
AI marketing agencies
Marketing firms that use AI tools to produce campaigns, content, advertising and search visibility for their clients. The AI sits inside their process; what you buy is marketing output, not a system you own.
AI automation agencies
Usually small teams building workflows on no-code platforms such as Zapier, Make and n8n, often sold as a monthly service. A good fit for well-defined, lower-risk processes. The depth of engineering, testing and data protection work varies widely from one firm to the next.
AI build and delivery firms
Engineering-led teams that design, build, integrate and operate AI systems against a business problem: agents, retrieval over company documents, voice and chat, and integrations with CRM, finance and operational systems. Some call themselves agencies, some studios, some consultancies.
The AI Consultancy works in the second and third categories: we build systems that run inside your operations. We are not a marketing agency.
What the work of an AI agency involves
- Scoping. One problem, one owner in the business, one measure of success and a baseline taken before anything is built.
- Data preparation. Finding the documents, records and system access the use case needs, and deciding what must stay out.
- Model and tool selection. Choosing the model, the platform and the hosting that fit the use case, the data and the budget.
- Build and evaluation. Prompts, tools, retrieval and integrations, tested against real tasks rather than demonstrations.
- Deployment. Moving from a working prototype to a production system with access controls, logging and monitoring.
- Running it. Support after launch, cost monitoring, and migration when the model provider retires the version the system was built on.
AI agency or AI consultancy: what is the difference?
The practical difference is what you are left with at the end. An agency leaves you with a system. A consultancy traditionally leaves you with a decision, a strategy or a roadmap, and someone else builds it. Each has a characteristic failure: an agency builds what it is briefed to build, including things that should not have been built, and a consultancy can produce a roadmap that never reaches production.
Our engagements are set up to avoid both: advice and build in one engagement, with the governance a consultancy would insist on and a working system at the end.
| Question | Typical AI agency | Typical AI consultancy | The AI Consultancy |
|---|---|---|---|
| What you buy | A working system | Advice, a strategy or a roadmap | Both, in one engagement |
| Who decides what gets built | Usually the client, in the brief | The consultancy, with the client | Agreed in a written, itemised proposal before build starts |
| Pricing | Per project or monthly retainer; rarely published | Day rate or project fee | Published fixed fees; day rate GBP 950 to GBP 1,500 where scope is open |
| UK GDPR and security | Varies by firm; often left to the client | Covered in the advice, not in a build | DPA, Article 30 entries and DPIA support in every engagement; security review on production builds |
| After go-live | A support contract, where one is offered | Not usually involved | On production builds: support window, runbook and monitoring, then your team or ours |
| Model and tool choice | Often the platform the agency knows best | Vendor-neutral in principle | Claude, OpenAI's latest GPT models or Gemini, chosen per use case |
Do you need an agency, a consultancy, or both?
The honest answer depends on how much of the decision is already made. Three sets of tests separate the cases.
You need build capacity when
- the use case is defined and someone in the business owns it;
- the data it needs exists and you can give access to it;
- success can be measured: hours saved a week, error rate, response time or throughput;
- the risk is contained, for example an internal process with a person approving anything consequential.
Where all four hold, a build can be scoped and quoted without a paid discovery stage: we do not bill for scoping calls or proposal writing.
You need advice first when
- you have several candidate use cases and no basis for choosing;
- the board wants a business case before approving spend;
- the data includes special category personal data, client confidential material, or sits under a sector regulator such as the FCA or SRA;
- a previous AI project stalled and nobody is sure why;
- you are choosing between vendors or platforms.
That is what our Readiness Sprint (two weeks, from GBP 3,500) is for. Half its fee is credited against a Discovery and Pilot or Build and Embed commissioned within 90 days, so advice first does not mean paying twice.
You may not need a build at all when
the work is drafting, summarising, research and analysis inside documents and tools your team already uses. A well-configured business AI workspace often covers that without custom software. Our Claude Enablement install sets one up in two or three on-site days, from GBP 4,500. It is worth ruling this out before paying for a build.
What we build
Each line below has its own service page with the method, cost bands and governance detail. Prices are the published entry points, ex VAT, and exclude running costs such as tool subscriptions and model usage.
Workflow automation
From GBP 5,000
n8n, Zapier and Make with a Claude or OpenAI step for classification, extraction or drafting, and bespoke applications on Google Cloud Run where volume, compliance or complexity need them. This is the work most people mean by an AI automation agency.
AI agents
Starter from GBP 15,000
Browser, internal workflow and customer-facing agents, with human approval on consequential actions, hard caps on steps and cost per run, and a full audit log on anything customer-facing.
AI chatbots
From GBP 3,000
Chatbots on the web, WhatsApp and Teams that answer from your own knowledge base and hand over to a person on a defined escalation path.
Voice AI
From GBP 5,000
Inbound and outbound call agents on ElevenLabs and Twilio, with intent classification and handoff to a member of staff.
LLM integrations and AI applications
From GBP 5,000
Claude, OpenAI's latest GPT models or Gemini integrated with your CRM, finance platform and document stores, evaluated against real tasks before go-live. Claude-specific builds run through Claude Implementation.
Legacy system replacement
From GBP 8,000
Access databases, spreadsheets run as operational systems, discontinued SaaS and ageing bespoke applications, replaced with AI-built software under senior engineering review.
On-premises AI
From GBP 2,500 plus from GBP 250 a month
A local AI agent on your own hardware for professionals whose client data cannot be sent to a cloud model: solicitors, IFAs, accountants, family offices and private clinicians.
Claude workspace install
From GBP 4,500
When the answer is not a build: a fixed-price, on-site install of a governed Claude workspace for your team, with a 30-day adoption window inside the price.
How an engagement runs: scope, pilot, build, run
We classify every build at the outset as a demo, a prototype, a pilot or a production system. Each carries a different cost and a different level of hardening, so agreeing the level first is what keeps the price and the result aligned. Smaller builds are quoted as a single fixed fee; larger ones move through our published packages.
1. Scope
A free scoping call, then a written, itemised proposal setting out the scope, the maturity level, the deliverables and the price. We do not bill for scoping or proposal writing. Once work starts, any change of scope is quoted in writing before the extra work begins, and the fee moves down as well as up.
2. Pilot: Discovery and Pilot, from GBP 15,000
Four to eight weeks. A solution design covering architecture, data flow, vendor selection and the UK regulatory position, and a working prototype on real or representative data, deployed to a UK or EU cloud region. The prototype is functional software, not a slide deck. It comes with technical documentation and a deployment plan with cost projections, so the production decision is made on evidence.
3. Build and deploy: Build and Embed, from GBP 40,000
Eight to sixteen weeks. The production build, integration with the systems it needs (CRM, ERP, accounting, ticketing, telephony, email and your cloud platform), the UK GDPR documentation, a security review, and onboarding for the staff who will use it, with role-specific workshops.
4. Run: after go-live
A 30-day post-launch support window, a written runbook, monitoring and cost monitoring, and a defined point of contact. The documentation is written so an in-house team or another supplier can take the system on. If you would rather we kept it running, managed support is available as a separate retainer. For ongoing AI leadership across the business, our Fractional Chief AI Officer retainer starts from GBP 3,000 a month.
Full terms, including milestone invoicing and what is included in every engagement, are on our pricing page.
Governance inside the build, not added afterwards
UK GDPR regulates what you do with personal data, whichever supplier builds the system. You stay the controller, so the build has to leave you able to show what the system does with that data and why. That is why the documentation is part of the deliverable in every engagement, not an extra.
- Data protection paperwork, in every engagement. A Data Processing Agreement where personal data is involved, Article 30 record of processing entries where required, and DPIA support where the use case triggers one.
- Security review, on production builds. Build and Embed includes a review of authentication, authorisation, secret management, logging and incident response procedures.
- Residency, documented. Where data is stored and where model inference runs are recorded for each component, rather than assumed. The current position for Claude and ChatGPT is set out on our AI Implementation page.
- Human approval on consequential actions. Our default rules for AI agents: writes to financial, HR or legal systems of record need a person to approve them, and any action that fails twice is escalated to a person rather than retried a third time.
- Automated decisions about people. Where a system would take a significant decision about someone with no meaningful human involvement, the safeguards in Articles 22A to 22D of UK GDPR, inserted by the Data (Use and Access) Act 2025, apply. Our guide to the Data (Use and Access) Act 2025 covers what they require.
- Cost caps. Agents run with hard limits on steps, tool calls and cost per run, set in the first week of the build, because usage-based pricing scales with adoption.
Delivery evidence
Two builds from our case studies, and the installs behind our Claude Enablement packages. We publish client names only where the client has agreed to it, and figures only where they are documented.
MoverAI: built, then extended
A video-based AI property survey platform for Master Removers Group. Customers or surveyors record a short walkthrough video of each room, and a Google Gemini multimodal model produces a structured inventory with cubic footage estimates, room-by-room breakdowns and packing box calculations. In a trial with 13 office-based users, 7 reported accurate or near-accurate volume estimates on first use. The platform was later extended with QuoteFlow, which qualifies inbound calls and sends the caller a survey link by SMS. The engagement began in June 2025 and continued beyond the first build.
Read the MoverAI case studyOne AI platform for an 11-site chain of dental surgeries
A cloud-native AI platform on AWS for a multi-chain of UK dental surgeries growing from one flagship practice to eleven: chat and voice patient communication (Amazon Lex, Connect and Polly), serverless appointment confirmations and payment reminders, and clinical AI support with DICOM image storage, deployed in the AWS London region. Total infrastructure cost under GBP 1,750 a month.
Read the multi-site dental case studySince June 2026 we have also delivered on-site Claude workspace installs for firms in pharma recruitment, engineering staffing, professional services, wine production, security, resource finance and paper products. Those are described on the Claude Enablement page, and the full set of engagements is on our case studies page.
Anthropic Consulting Partner
Claude delivery with Claude Certified Architect practitioners
Certifications
AWS, Google Cloud and Nvidia
Published pricing
Fixed fees for every stage, ex VAT, on our pricing page
UK delivery
London-based, working with clients across the UK
How to choose an AI agency: eight questions to ask, including us
Searching for the best AI agency returns lists, directory badges and logo walls. None of those tells you whether a particular firm will build the right thing and leave it in a state you can live with. These questions do.
- What state will the system be in at the end? A demonstration proves an idea; a production system has to survive real users, real data and a bad day. Ask which one the quote is for, and get the answer in the contract.
- Who owns the code, the prompts, the configuration and the data? Ask whether the model accounts and cloud resources will be held in your name or the supplier's, and have the answer written into the contract before build starts, when it costs nothing to settle.
- What will it cost to run, not just to build? Model usage and tool subscriptions scale with adoption. Ask for a monthly running-cost estimate alongside the build quote.
- What happens when the model changes? Providers retire model versions on published schedules. Ask who migrates the system and whether that is inside the fee.
- How will we know it is working? Agree a measurable success criterion, and take a baseline, before the build starts. Without one, nobody can say afterwards whether the money was well spent.
- What is automated, and what needs a person to approve it? Ask to see the list. If there is not one, the question has not been considered.
- Who supports it after launch, and could someone else take it over? Ask for the response times and a sample of the handover documentation. If only the supplier can keep a system running, you have bought a dependency as well as a system.
- Will you tell us not to build something? Ask for an example of a use case the firm advised against. The case studies that matter are the ones with figures you can check, for clients like you.
For a fuller procurement checklist, see our guides to AI vendor due diligence and whether to build or buy.
Frequently asked questions
What is an AI agency?+
What does an AI agency do?+
Is The AI Consultancy an AI agency?+
How much does an AI agency cost in the UK?+
Do we need an AI agent or a workflow automation?+
Which AI models and tools do you build with?+
Can you take over an AI system another agency built?+
What happens after the system goes live?+
Do you only work with London businesses?+
Related services and reading
- The AI Consultancy: who we are, and the AI consultancy work that sits alongside our builds.
- AI Strategy Consulting: for when the question is what to build, not how.
- AI Workflow Automation: our AI automation agency work, from no-code to bespoke.
- How much does AI implementation cost in the UK?: build and running costs, with the working shown.
- Agentic AI for UK businesses: where agents earn their cost and where a workflow is enough.
- AI App Production Clinic: engineering review for applications built with AI coding tools.
Readiness Sprint from GBP 3,500 · Claude Enablement from GBP 4,500 · Discovery and Pilot from GBP 15,000 · Build and Embed from GBP 40,000 · Day rate GBP 950 to GBP 1,500. All prices exclude VAT.
Book a free scoping call
Tell us what you want built, which systems it touches and how you would measure it. We will tell you whether it is worth building, at what maturity level, and which route fits, from a fixed-price build to a Discovery and Pilot. Pricing follows in writing.