This guide cuts through the confusion: the four types of AI agency, what they actually deliver, what UK pricing looks like, how to pick the right one, and the red flags to walk away from. But first: what does an AI agency do? This is what you must know.
An AI agency helps businesses identify, build, and deploy artificial intelligence solutions, from AI-powered marketing campaigns and ecommerce personalisation through to custom machine learning systems and workflow automation. The term has exploded into common use, but it covers a wildly inconsistent range of services. Some “AI agencies” are repackaged digital agencies adding ChatGPT to their workflow. Others are deep technical shops building custom models. Most UK businesses asking what does an AI agency actually do are caught somewhere between the two, knowing they need AI capability but unsure what to buy, who to trust, or how much to spend. This guide cuts through the confusion: the four types of AI agency, what they actually deliver, what UK pricing looks like, how to pick the right one, and the red flags to walk away from.
What an AI Agency Actually Is
An AI agency is a service business that helps companies adopt and deploy artificial intelligence. The remit varies enormously depending on the agency, but the work usually falls into one or more of: AI strategy and roadmap, custom AI build (machine learning models, chatbots, automation), integration of off-the-shelf AI tools, and ongoing optimisation and training.
Where it gets confusing is that the term is used loosely. A traditional digital marketing agency that started using ChatGPT to draft copy may now call itself an “AI agency”. So might a deep-tech consultancy that builds proprietary machine learning systems for enterprise clients. The work, the cost, and the outcomes are completely different. Knowing which type you’re talking to is the single most important first step.
The plain-English definition
An AI agency sells you AI capability you don’t have in-house. That capability might be strategic (helping you decide where to apply AI), creative (using AI to generate content faster), technical (building custom AI systems), or operational (automating workflows). Most established agencies do at least two of these, but few do all four well.
Why “AI agency” became a term
The label exists because demand for AI services outpaced the existing structure of marketing, dev, and consulting agencies. Buyers wanted a single point of contact for “we want to do something with AI”. Agencies repositioned to offer that, sometimes legitimately, sometimes by adding “AI” to existing service pages. The result is the noisy market UK businesses are trying to navigate.
AI Agency vs AI Agent: Two Completely Different Things
Search results for “AI agency” are polluted by content about AI agents, autonomous software entities that perform tasks. They’re not the same thing, and confusing them leads to expensive mistakes. Here’s the clean distinction.
| Term | What it is | What it does |
|---|---|---|
| AI agency | A service business (people, expertise, agency model) | Helps companies build, buy, or deploy AI solutions |
| AI agent | A piece of software powered by AI | Performs autonomous tasks on behalf of users |
An AI agency might *build* AI agents for you, but the agency itself is a service partner, not software. AWS defines an AI agent as “a software program that can interact with its environment, collect data, and use that data to perform self-directed tasks.” That definition has nothing to do with the question “what does an AI agency do?”.
If you’re trying to *adopt AI in your business* and need help working out where to start: you need an AI agency. If you’re trying to build software that runs autonomously to handle a workflow: you might need to build (or buy) an AI agent, often via an AI agency.
The 4 Types of AI Agency (and Which One You Actually Need)
Most “AI agency” market noise can be cleanly broken into four distinct models. Each does different work, charges different prices, and serves different needs. Understanding the difference saves UK businesses from buying the wrong type entirely.
Uses AI tools to deliver traditional marketing services faster, content production, ad optimisation, SEO, email, social. The output is marketing; AI is the engine. Best for brands wanting more output from existing budgets.
Applies AI specifically to online retail: personalisation, recommendations, generative engine optimisation, agentic commerce readiness, AI-driven CRO. Best for ecommerce brands needing AI applied to revenue-driving operations.
Builds workflow automations using AI plus tools like Zapier, n8n, Make, and custom scripts. Lead routing, support triage, data entry replacement, reporting dashboards. Best for ops-heavy businesses with clear repetitive processes.
Builds custom AI and machine learning systems, often from scratch or fine-tuned on proprietary data. Custom models, computer vision, NLP, predictive analytics. Most expensive and most specialist. Best for unique competitive advantage.
The hybrid model: AI-native ecommerce agencies
A growing fifth category sits at the intersection of types 2 and 4: agencies that combine deep ecommerce expertise (Shopify, BigCommerce, headless commerce) with AI capability (custom models, GEO, agentic commerce protocols). These agencies, like 5MS, exist because AI is no longer a separate workstream from ecommerce growth, it’s woven through every part of it. For UK ecommerce brands, this hybrid model usually delivers more value per pound than buying ecommerce and AI services separately.
Hiring a Type 1 AI marketing agency when what you actually need is a Type 4 AI development agency, or vice versa. The cost difference can be 5x or more, and the wrong type will not solve your actual problem regardless of how much you spend. Define what you need from AI before talking to anyone.
What an AI Agency Actually Delivers
Across the four types, AI agencies deliver around eight distinct services. Most agencies offer three to five of these; very few offer all of them well. Understanding what each one is, and isn’t, helps you scope what you actually need.
AI strategy and readiness assessment
The discovery work. The agency reviews your business, identifies where AI can deliver value, prioritises opportunities by ROI and feasibility, and produces a roadmap. Often the smartest first engagement, scoped at £5K to £25K depending on business complexity. Without this, every downstream investment risks landing in the wrong place.
AI-powered content and creative production
Using generative AI (text, image, video) to produce more marketing content faster. Blog articles, product descriptions, ad creative, social content, email sequences. The output is content, the differentiator is volume and speed. Most digital agencies now offer this; the real test is whether the content performs, not whether it was generated quickly.
Custom AI tool development
Building bespoke AI applications: recommendation engines, predictive analytics dashboards, computer vision tools, custom GPTs, internal knowledge assistants. Output is software you own (or have exclusive access to). Project-based pricing, typically £15K to £150K depending on complexity.
Workflow automation
Using AI plus automation platforms (Zapier, n8n, Make) to remove repetitive manual work. Automated lead qualification, support ticket triage, data entry replacement, automated reporting. Strong ROI when applied to genuinely repetitive processes; expensive theatre when applied to processes that should be redesigned, not automated.
Conversational AI and chatbots
Building AI-powered customer service, sales, and support assistants. Includes integration with platforms like Gorgias, Intercom Fin, Ada, or custom-built LLM workflows using RAG (retrieval-augmented generation). Strong fit for ecommerce stores with high support volume and clear FAQ patterns.
AI integration with existing systems
Connecting AI capability to the systems you already run, your CRM, ERP, ecommerce platform, marketing stack. The unsexy but critical work that determines whether AI investments actually deliver. Often the difference between AI that ships and AI that stalls in proof-of-concept forever.
Generative engine optimisation (GEO)
Optimising your content and product data so AI engines (ChatGPT, Perplexity, Gemini) cite your brand in answers. The natural successor to SEO as buyers move research into AI conversations. Read our full guide on generative engine optimisation for the detail.
Training, enablement, and ongoing support
The work that determines whether AI adoption sticks. Training internal teams to use AI tools well, building prompt libraries, defining workflows, ongoing optimisation as models evolve. Often delivered as a monthly retainer; the most consistently undervalued part of AI agency engagements.
Which Type of AI Agency Do You Actually Need?
Six questions to identify which of the four AI agency types is the right fit for your business. The recommendation reflects the patterns we see across hundreds of UK businesses scoping AI work.
Find your AI agency match
Answer for the primary outcome you want from AI investment. Takes 60 seconds.
Your recommended AI agency type
UK AI Agency Pricing and Engagement Models
AI agency pricing is one of the most opaque corners of professional services. Quotes for ostensibly similar projects can range 10x. Some of that is genuine quality variation; some is agencies pricing on what they think you’ll pay rather than what the work costs. These benchmarks reflect typical UK market ranges across the four agency types.
Pricing by engagement type
| Engagement | Typical UK price range | What it covers |
|---|---|---|
| AI strategy / discovery | £5,000 – £25,000 | Audit, opportunity mapping, prioritised roadmap |
| Pilot / MVP build | £10,000 – £40,000 | Single working solution proving value, 4-8 weeks |
| Core implementation | £25,000 – £120,000 | Multi-system AI rollout, 3-6 months |
| Custom AI / ML build | £40,000 – £250,000+ | Bespoke models, deep technical work |
| Monthly retainer (small) | £600 – £2,500/mo | Ongoing automation maintenance, light optimisation |
| Monthly retainer (full service) | £3,000 – £12,000/mo | Active optimisation, content, ongoing development |
| Day rate (consulting) | £800 – £2,500/day | Strategic input, senior expertise on demand |
Pricing models you’ll encounter
Fixed-scope project pricing
Set deliverables, set price. Best for well-defined work with clear outcomes. Advantage: predictable. Disadvantage: encourages agencies to scope conservatively and bill change requests aggressively.
Monthly retainer
Ongoing access to capacity for a fixed monthly fee. Best for continuous AI work, content, optimisation, automation maintenance. Pricing typically reflects guaranteed hours plus access.
Hybrid (setup + retainer + usage)
The default model. Initial setup fee builds the system; monthly retainer maintains and optimises; usage fees cover platform costs (OpenAI, Claude, Anthropic API charges). Most modern AI agencies in the UK structure work this way because it reflects how the underlying tools are actually billed.
Performance / outcome-based
Fee tied to a measurable outcome, leads generated, revenue attributable, hours saved. Increasingly common but requires clean attribution. Best for mature engagements where baselines are well established.
AI services run on platform APIs (OpenAI, Anthropic, Google) that charge per token or per request. A “fixed price” quote that doesn’t separate platform costs from execution is hiding ongoing variable expense. Always ask: is the API usage included, and what happens if usage scales? A clear answer saves nasty surprises in month three.
AI Agency vs Building an In-House AI Team
Every business considering serious AI investment eventually asks whether to hire an agency or build the capability internally. There isn’t a universal right answer, but there is a clear framework for making the call.
| Factor | AI agency | In-house team |
|---|---|---|
| Time to first output | 2-8 weeks | 3-6 months (recruit, onboard) |
| Up-front cost | Lower (project or retainer) | High (salaries, benefits, equipment) |
| Skill breadth | Multi-disciplinary by default | Limited to the people you hire |
| Long-term cost efficiency | Good for variable workloads | Better for sustained heavy use |
| Knowledge retention | Walks out the door if you leave | Stays in the business |
| Best for | Most SMEs, ecommerce, brands without an AI strategy yet | Enterprise, sustained heavy AI workloads, IP-sensitive work |
The hybrid that wins for most UK ecommerce brands
For most UK ecommerce businesses under £20M revenue, the right answer isn’t agency or in-house, it’s both, sequenced. Engage an AI agency to define the strategy, build the first systems, and deliver early wins. Use that work to clarify what AI capability you genuinely need long-term, then hire one focused in-house person (often an AI Product Manager or AI-fluent senior marketer) to own ongoing direction. The agency continues delivering execution; the in-house hire owns strategic continuity.
Want to scope an AI engagement properly?
5MS is a UK ecommerce agency with deep AI capability. We work with brands at every stage, from “we have no AI strategy” to “we need a custom model built”. Free 30-minute call to scope what’s right for your business.
How to Choose the Right AI Agency
Most AI agency selection processes are too short and rely on the wrong signals. These are the seven steps that consistently separate good selections from expensive mistakes.
Define the outcome before the brief
Write down, in plain English, what you want to be true 6 months after the engagement starts. “Reduce customer service ticket volume by 30%” is a clear outcome. “Implement AI” is not. Briefs without outcomes attract proposals without accountability.
Match agency type to outcome
Use the four-type framework above. Don’t brief a Type 1 (AI marketing agency) on a Type 4 (custom development) project, or vice versa. The mismatch shows up as missed deliverables and surprise scope changes.
Check the case studies for evidence, not theatre
Strong AI agencies show specific outcomes with numbers (“reduced support volume by 40%”, “lifted CVR 12%”). Weak ones show logos, awards, and vague claims. If every case study says “transformative” but no case study has a number in it, you’re looking at marketing dressed as evidence.
Talk to the people who’ll do the work
Many agencies sell with senior people and deliver with juniors. Ask explicitly: who runs my account day-to-day, what’s their experience, and can I meet them before signing? Quality of the actual delivery team predicts outcomes more than quality of the pitch deck.
Ask about platform costs explicitly
“Does your quote include OpenAI / Anthropic / Google API costs, and what happens to my bill if usage scales?” The answer reveals how transparent the agency is, and how thoughtful they’ve been about ongoing economics.
Get references from completed engagements
Live engagements love their agency; finished engagements know whether it actually worked. Ask for references from clients whose projects ended at least 6 months ago, with permission to ask honestly about what worked and what didn’t.
Start with a paid pilot before committing to a long contract
A 4-6 week paid pilot tells you more than any pitch. You learn how they work, how they communicate, how they handle changes. Saved pain on a bad fit is worth multiples of the pilot cost.
10 Red Flags When Hiring an AI Agency
Each item below is a pattern we’ve seen repeatedly when reviewing failed AI engagements. Individually, any one is a caution flag. Two or more, and you should walk away regardless of how good the quote looks.
- “AI” added to every service line on the website with no specific capability described
- No case studies with numerical outcomes, only logos and vague claims
- Vague pricing, “starts from £X” with no scoping conversation
- Senior team in pitch, junior team in delivery, with no clarity on who’ll actually do the work
- Refusal to break out platform / API costs from agency fees
- “AI will solve everything” framing instead of clear scope and trade-offs
- No mention of evaluation, monitoring, or how AI quality will be measured ongoing
- Promises of specific ROI numbers (“3x ROAS guaranteed”) without baseline data
- No published thinking, no real depth on AI in their content, talks, or open work
- Pressure to sign a 12-month contract before a pilot has demonstrated value
“The most expensive AI mistakes we see in UK ecommerce aren’t bad models or wrong tooling. They’re businesses that bought the wrong type of agency for the outcome they wanted, then spent six months trying to reshape the engagement instead of admitting the mismatch and starting again. Match the agency type to the outcome before anything else.”
Paraphrased from UK ecommerce AI engagement patterns
What an AI Agency Does for Ecommerce Brands
For ecommerce, AI is no longer a single feature, it’s woven through every part of the operation. A specialist AI ecommerce agency typically delivers across these areas, often as connected work rather than isolated services.
Core ecommerce AI services
AI-driven personalisation and recommendation
Product recommendations that actually convert, dynamic homepage merchandising, personalised email sequences. Lifts AOV and CVR meaningfully when implemented well; performs no better than rule-based systems when implemented badly.
Generative engine optimisation (GEO)
Optimising product data, schema, and content so AI engines (ChatGPT, Perplexity, Gemini) cite your store when shoppers ask “best [thing] in the UK”. The successor to traditional SEO and one of the most strategically important areas right now. Read our full guide to generative engine optimisation for the deep dive.
Agentic commerce readiness
Preparing your store for AI agents that buy on shoppers’ behalf, ChatGPT Instant Checkout, Google’s Universal Commerce Protocol, and similar systems. Schema, real-time inventory, structured product data, protocol integration. Brands without this work risk being invisible to AI shoppers entirely. Our full agentic commerce guide covers the technical foundation.
AI-powered customer service
Conversational AI for support, returns, post-purchase questions. Built using platforms like Gorgias, Intercom Fin, Ada, or custom RAG-based assistants. Handled well, displaces 30-60% of repetitive support volume. Handled badly, frustrates customers and damages CSAT.
AI for merchandising and demand forecasting
Predictive models for stock levels, seasonal demand, new-product success scoring. Particularly valuable for stock-sensitive categories (fashion, food, fast-moving consumer goods). Strong ROI when paired with clean inventory data; expensive guesswork without it.
AI-driven CRO and experimentation
Using AI to identify CRO opportunities, generate test variants, and predict winners faster than traditional A/B testing alone. Compresses experimentation cycles from months to weeks for stores running enough traffic.
Integrated AI marketing execution
AI-accelerated content production, ad creative testing, generative imagery for product variants, AI-driven email personalisation. The bread-and-butter work that AI ecommerce agencies blend with the more technical capabilities above.
An ecommerce-specialist AI agency that bundles these services together usually delivers more value per pound than buying ecommerce work from one agency and AI work from another. The reason: most of these services depend on the same underlying foundations (clean product data, schema markup, real-time inventory, well-instrumented analytics). One team building all of that once costs less than two teams duplicating it.
What an AI Agency Engagement Actually Looks Like
Knowing what to expect across the engagement removes the most common source of friction: misaligned expectations on timing and milestones. Here’s the typical shape of a serious AI agency engagement.
Discovery (Weeks 1-2)
Workshops, system reviews, opportunity mapping, success metric agreement. Often delivered as a paid discovery sprint that produces the project roadmap.
Design and architecture (Weeks 2-4)
Solution design, data model, integration plan, prompt frameworks, model selection. The phase where decisions are made that determine cost, performance, and maintainability of everything downstream.
Pilot or MVP build (Weeks 3-8)
First working version of the AI solution. Typically scoped tight to prove value fast. The deliverable is something you can actually use, not a prototype that needs another six months to ship.
Iteration and optimisation (Weeks 6-16)
Based on real-world usage, refinements, edge case handling, and gradual scope expansion. Often where the bulk of the work happens; AI systems benefit hugely from real-data feedback loops.
Embedding and training (Weeks 8-20)
Internal teams trained on the new capability, processes updated, documentation written, ownership handed over. The phase that determines whether AI investment delivers ongoing value or becomes shelfware.
Ongoing optimisation (Month 6+)
Retainer-based or project-based ongoing work. Model improvements as new versions release, expanded use cases, integration with adjacent systems. The compound returns of AI investment usually show up here, not at launch.
Well-scoped AI engagements typically show measurable ROI within 3-6 months of starting. Faster claims usually involve trivial use cases (a chatbot answering FAQs); slower returns usually mean scope creep, data issues, or wrong agency type. If you’re 6 months in without measurable outcomes, something needs renegotiating.
What an AI Agency Does: The Short Answer
An AI agency helps businesses adopt artificial intelligence by delivering one or more of: AI strategy and roadmaps, AI-powered content production, custom AI development, workflow automation, conversational AI, system integration, generative engine optimisation, and ongoing training. AI agencies fall into four main types: AI marketing agencies (output focus), AI ecommerce agencies (revenue focus), AI automation agencies (operations focus), and AI development agencies (custom build focus). UK pricing typically ranges from £5K for discovery work through to £250K+ for substantial custom builds. The single most important step is matching the agency type to the outcome you want before scoping anything else.
The 10-step action list:
- Define the outcome in plain English before writing a brief.
- Identify which of the 4 agency types matches the outcome.
- Use the calculator above to confirm your match.
- Set a budget range aligned to UK market benchmarks.
- Shortlist 3-5 agencies that genuinely fit your type.
- Check case studies for numbers , not just logos and awards.
- Meet the delivery team before signing anything.
- Run the 10-point red flag check on every shortlist agency.
- Start with a paid pilot before committing to a long contract.
- Plan for ongoing optimisation , AI investment compounds with iteration.
Looking for an AI ecommerce agency in the UK?
5MS combines deep ecommerce expertise with AI capability across personalisation, generative engine optimisation, agentic commerce readiness, and custom AI development. We’re an AI-native UK agency built specifically for ecommerce brands. Book a free 30-minute call to scope what’s right for your business.
What Does an AI Agency Do: Frequently Asked Questions
Sources & further reading
- McKinsey: AI agents and enterprise value forecasts
- AWS: What are AI agents?
- Google Cloud: AI agents definition and types
- BCG: AI Agents and business impact
- Stripe: Agentic commerce documentation
- Digital Agency Network: AI agency pricing benchmarks
- Use All Five: AI agency typology framework
- Zapier and n8n: workflow automation platform documentation
- Schema.org: structured data specifications
This guide is updated quarterly with refreshed pricing benchmarks, agency type definitions, and shifts in the AI services landscape.