Search for an AI agency in the UK and you'll find hundreds of firms describing themselves in much the same words. Some build software. Some resell a platform. Some run workshops and write strategies. Some do a bit of everything. The label doesn't tell you much, so the questions you ask matter.
This guide is for small and mid-sized businesses looking for help putting AI agents to work on real business systems: the accounts, the CRM, payments, support and the shared inboxes. The questions work whether you're talking to an AI agency, an AI consultancy or a freelancer, in London or anywhere else in the UK.
What they actually do
1. Do they build, resell or advise?
Ask plainly what you'll have at the end. A strategy document? A licence for someone else's platform? Working agents connected to your systems that your team uses every week? All three can be legitimate, but they're very different purchases. If the answer is a platform they resell, ask who supports it, and what happens if the agency and the platform part ways.
2. What will they build first, and why?
A good agency starts from how your business runs, not from a demo. Expect questions about which teams do the most routine work, which systems hold what, where things get dropped between people, and where the risks are. Be wary of a firm that proposes a long list of agents before it has looked at your systems, or promises a fixed percentage of work automated before discovery.
3. Can they show you the work, not just the slides?
Ask to see something running: an agent doing a real job, the rules it works to, and the record of what it did. If everything they can show is a prototype or a video, ask why.
Your systems and your data
4. Which systems can they connect, and how?
Get specific. Name your accounting system, CRM, payment provider, help desk, project tool, chat app and email, and ask how each one will be connected. Agents work best on systems with good APIs, such as Xero, QuickBooks, HubSpot, Pipedrive, Stripe, Zendesk, Asana, Google Workspace or Microsoft 365. If a key system has no API, or only a desktop version, you want to hear that now rather than in week three.
5. Where does your data go?
You should know which AI providers the agency uses, where data is processed and stored, whether any of it is used to train models, and how long it's kept. Ask for the list of subprocessors.
6. How do they handle UK GDPR?
If the agency will process personal data on your behalf, which it almost certainly will if agents read your email or CRM, it should be acting as your processor under a written data processing agreement. Ask for theirs. Ask how they handle international transfers if any provider sits outside the UK, and what happens to your data when the contract ends. If you handle special category data, such as health information, raise it at the start.
7. What can each agent reach, and who decides?
Agents act on what they read, so a hidden instruction in an email or a document could try to steer one, and connecting systems widens what it could reach. Look for:
- least access: each agent reaches only the systems and actions its job needs
- named callers: you decide who can ask each agent for what
- approvals: anything outside your rules, such as a refund over a limit, waits for a person
- an audit log: every run recorded with who asked, what it touched and what it cost
Who owns what
8. Who owns the agents and the know-how?
Ask what you keep if you stop working together: the instructions and rules each agent runs on, the configuration, the documentation, your data. Some arrangements leave everything inside the agency's platform. That can be fine if it's clear and the terms are fair, but you should know before you sign.
9. Will your team be able to run it?
The best outcome is a capability that stays inside your business. Ask what training is included, who on your side needs to take part, and whether your managers will be able to give and remove access, read the log, approve held actions and adjust agents without calling the agency every time.
How they price it
10. What are all the costs?
AI work usually has three kinds of cost, and you should see all three in writing:
- Set-up: discovery, connecting systems, building and testing the first agents, training. Usually a one-off fee or a fixed block of time.
- Ongoing: hosting, maintenance, keeping connections working, support and changes. Often a monthly subscription.
- AI usage: what the AI provider charges for each run. Ask whether it's passed through at cost, marked up or bundled, and how you'll see it.
As one example, we charge £4,000 plus VAT for a 30-hour set-up, then a monthly subscription for the managed service, with AI usage shown run by run. Other firms structure it differently. What matters is that nothing is hidden. Our pricing page shows what that looks like in practice.
11. What's excluded?
Ask what isn't covered: licences for the systems you connect, custom integrations, a wider rollout, changes after go-live. A clear list of exclusions is a good sign.
12. Can you start small?
A sensible first engagement is bounded: one team, one or two agents, a fixed amount of time, a clear review at the end. Be cautious about long contracts before you've seen anything working on your own data.
After go-live
13. What happens when something changes?
Systems change their APIs, processes change, people join and leave. Ask who notices when a connection breaks, how quickly it's fixed, and how changes to an agent are requested and made.
14. How will you know it's working?
Agree up front what success looks like: hours saved on a named job, fewer errors, faster responses. Ask how the agency will measure it with you, and what happens if the numbers don't add up.
Red flags
- a fixed automation percentage promised before anyone has looked at your systems
- no clear answer on where your data goes, or no data processing agreement
- agents that act with broad access and no approvals or log
- pricing that leaves out AI usage or ongoing costs
- no training or handover, so everything depends on the agency
- pressure to sign a long contract before a first piece of work is delivered
A short checklist
Before you sign, you should be able to answer yes to each of these:
- We know exactly what will be built first, and why.
- We know how each of our systems will be connected.
- We know where our data goes, and there's a data processing agreement.
- We know what each agent can reach, what needs approval, and where to see what it did.
- We know every cost: set-up, ongoing and AI usage.
- We know what we keep if we stop working together.
- Our team will be trained to run it.
If you'd like to put these questions to us, book a strategy call. We're an AI agent agency based in London, working with businesses across the UK, and our implementation page sets out exactly what the first engagement covers. If you're still deciding whether you need agents at all, start with ChatGPT, Copilot or AI agents.