There has never been more choice of AI suppliers in the UK. Established software houses have added AI to their services, marketing agencies now sell chatbots, new specialist studios appear every month, and a long tail of freelancers can wire a language model into almost anything. That's good news for buyers, but it makes the choice harder. Most proposals look similar, most demos are impressive, and the differences only show up months later, when the system meets real customers and real data.
This guide is the checklist we would want a business to use on any AI agency, including us. It covers how to frame the project before you talk to anyone, when an agency is the right kind of supplier at all, the questions that separate experienced teams from confident ones, and the red flags worth walking away from.
Start with the problem, not the agency
The most useful thing you can do before contacting any AI agency is write one page about the problem. Not "we want an AI chatbot", but "we get 300 enquiries a week, half of them ask the same ten questions, and replies take a day because the team is busy". A page like that does three things. It lets agencies quote against the same brief. It gives you a way to judge their answers. And it makes it obvious when someone is selling you a product rather than solving your problem.
Your page should cover:
- The job: what happens today, who does it and how long it takes.
- The cost of the problem: lost enquiries, staff hours, slow responses, errors.
- What success looks like: a number you could measure in three months.
- The data involved: where it lives, who owns it and whether it includes personal or sensitive information.
- The constraints: budget range, deadline, regulators, systems that can't change.
If you can't fill in the page yet, that is a sign you need advice before a build. A short, fixed-price AI consultancy engagement exists for exactly this: ranking your ideas by value and effort before you commit money to any of them.
Agency vs freelancer vs in-house: which do you need?
Before choosing an AI agency, check that an agency is the right shape of supplier. Each option suits a different kind of project.
| AI agency | Freelancer | In-house hire | |
|---|---|---|---|
| Best for | Defined projects that need several skills and must ship in weeks | Narrow, well-specified tasks you can manage yourself | A long-running AI roadmap that is core to the business |
| Skills covered | Strategy, engineering, integration, design and support together | Usually one or two specialisms | Whatever you can recruit, one hire at a time |
| Time to start | Once scope is agreed | Often quick, subject to availability | Recruitment plus notice period |
| Main risk | Paying for more process than a small job needs | Single point of failure if they move on or fall ill | One person expected to cover every skill |
| Who manages it | The agency, with your sign-off | You | You, plus technical leadership |
These aren't exclusive. A common pattern is to use an agency for the first one or two systems, learn what AI does for your business, and then hire in-house once there is a roadmap to justify the role. If you're weighing that decision in detail, our guide to an in-house AI hire vs an AI agency compares the costs line by line.
The AI agency checklist: what a good one should show you
Use this list while you read proposals and sit through pitches. A strong agency will cover most of it without being asked.
- Live systems, not just demos. Ask to see something they built that real users rely on today, and how it's monitored.
- Engineers in the room. The people pitching should include someone who will build your system, or you should meet them before signing.
- A clear discovery stage. Bespoke AI needs the workflow mapped and feasibility tested before a reliable price exists.
- A written scope and price. What will be delivered, by when, for how much, and how changes are handled.
- Honest limits. They should tell you which parts of your idea AI won't do well.
- A data map. Every supplier and AI provider that will touch your data, where it's processed and whether it can be used for training.
- Human oversight by design. For anything that affects customers or staff, a named person reviews or can override the AI.
- Testing before launch. How they'll check answers are accurate, safe and on-brand with your real questions, not just happy-path demos.
- Running costs explained. Model usage, hosting and support, separate from the build fee, with an estimate at your volumes.
- A way out. Prompts, configuration, data and workflows documented so another supplier could take over.
Questions to ask an AI agency before you sign
These questions are worth asking every agency on your shortlist. The answers matter less than how specific they are: experienced teams answer with examples and trade-offs; inexperienced ones answer with adjectives.
"What would you not use AI for in our business?"
A good agency will name something. If every part of your process is apparently perfect for AI, you are talking to a salesperson.
"Which model or platform would you use, and why?"
Listen for reasoning about cost, accuracy, speed and data handling, and whether they would switch models if a better or cheaper one appears. An agency that only ever uses one vendor may be choosing for its own convenience.
"How will you know it's working?"
You want a measurable target agreed up front, a test set built from your real cases, and monitoring once it's live. "The client was happy" isn't a measure.
"What happens when the AI gets something wrong?"
It will, sometimes. The answer should cover how errors are caught, how a person takes over, and how the system is corrected afterwards.
"What will this cost to run each month?"
A build fee is only part of the cost. Ask for model usage, hosting and support costs at your expected volume, and whether those accounts will be in your name or theirs.
"Who will actually do the work?"
Ask for names and roles. If the build is subcontracted, you should know who to, and where they are.
"What do we own at the end?"
Get a straight answer on code, prompts, configuration, data and documentation before you sign, not when you want to change supplier.
Red flags when hiring an AI agency
- Guaranteed accuracy or ROI before they've seen your data. Nobody can promise a number without testing on your cases.
- A detailed fixed quote after one short call. Either the scope is padded to cover unknowns, or the unknowns will become change requests.
- Vague answers on data. "It's all secure" is not a data map. You need to know which providers process what, and where.
- No live examples. Prototypes are easy now. Systems that keep working for months are the evidence that matters.
- Everything is a chatbot. If every problem gets the same product, the agency is fitting your business to its offer.
- Long retainers up front. Pressure to sign a twelve-month contract before a first system has shipped puts all the risk on you.
- Undocumented prompts and workflows. If only the agency understands how your system works, you're locked in.
- No mention of regulation. If you work in financial services, health, legal or with children's data and nobody raises the rules, they haven't thought about them.
Understanding how AI agencies price their work
AI agency pricing in the UK usually falls into three shapes: fixed-price products with a setup fee and monthly charge (common for chatbots and voice agents), fixed-price projects agreed after a discovery stage, and retainers for ongoing development. Day rates are common for advice but leave the total open-ended.
Whichever model you're offered, compare quotes on the same basis: what's delivered, who does it, how changes are priced, and what it costs to run. Model and hosting costs should be visible rather than buried in a markup.
To see real figures rather than ranges from an article, our AI consultancy cost guide lists our published prices for assessments, strategy, proofs of concept and day rates, and our AI chatbot cost guide does the same for chatbots. Our full AI development pricing page covers discovery, builds and ongoing partnerships.
How to run a fair selection
- Shortlist two or three agencies. More than that and you'll spend more time in pitches than the project saves.
- Send everyone the same one-page brief. It makes proposals comparable and shows who read it.
- Score against your checklist, not the demo. Demos are designed to impress; the checklist reflects what you need.
- Ask for a reference or a live system. Ideally something similar in size or sector.
- Start small. A paid discovery stage or proof of concept tests the relationship before a larger commitment. If it goes badly, you've lost weeks, not a year.
A first project that automates one painful job, such as the admin after an enquiry arrives, is often a better test than an ambitious platform. Our AI automation work usually starts there for exactly that reason.
Where we fit, honestly
We're a small UK AI development agency based in Manchester, working with businesses across the country. We advise and build, so the people you meet at the start are the people who ship your system. Bespoke work starts with a free scoping call and a fixed-price Discovery Sprint that's credited back if you build with us. Model and hosting costs are billed at cost on your own accounts.
We're not the right choice for everything. If you need a large team on site every day for a multi-year transformation programme, a bigger consultancy will suit you better. If you want one team to work out where AI pays off and then build it, you can see the products we've built and run and put every question above to us.
Choosing an AI agency: frequently asked questions
How do I choose an AI agency in the UK?
Start by writing down the business problem and how you'll measure success. Shortlist two or three agencies, send them the same brief, and compare what they'd deliver, who would do the work, how they handle your data and what it costs to run once live. Ask to see systems they've built and still operate.
Should I use an AI agency, a freelancer or hire in-house?
An agency suits a defined project that needs several skills at once and has to ship in weeks. A freelancer suits a narrow, well-specified task where you can manage the work yourself. An in-house hire suits a long-running AI roadmap with technical leadership already in place to support them.
Is it normal to pay for discovery before an AI build?
Yes. A reliable fixed price for bespoke AI work needs someone to map your workflow and test feasibility first. Many agencies charge for that discovery stage, and some credit it against the build. Be wary of a detailed fixed quote given after one short call.
What are the biggest red flags when hiring an AI agency?
Guaranteed accuracy or ROI figures before they've seen your data, vague answers about where your data goes, no examples of live systems, a refusal to document prompts and workflows, and pressure to sign a long retainer before anything has been delivered.
Who owns the work when an AI agency builds something for us?
That depends on the contract, so ask before you sign. At minimum you should be able to take your data, prompts, configuration and documentation to another supplier, and model and hosting accounts should ideally sit in your name.
Conclusion
Choosing an AI agency in the UK comes down to a few things you can check: a clear problem on your side, live systems and engineers on theirs, honest limits, a transparent data map, visible running costs and a way out. Start small, compare like with like, and treat confident promises without evidence as the warning they are. If you'd like us to be one of the agencies on your shortlist, book a free scoping call and bring your one-page brief.
