Straight answer: before you hire an AI agency, check what they’ve built for themselves (not just for clients), get a fixed handover date in writing, and ask what happens to your data and workflows the day the contract ends. Most people focus on the pitch deck and the case studies. The thing that decides whether you get value for money is what’s written into the contract about ownership, documentation, and exit.
Related reading: What Should You Look For in an AI Powered Customer Service Platform.
Related reading: How to Read an EPOS Sales Report and Use It to Grow Your Business.
Start by asking what they’ve built for their own business
I sat through a pitch in early 2024 from an agency that wanted £6,000 to build me an AI-powered content workflow. Beautiful deck. Logos of three household names as clients. Then I asked a simple question: what does your own agency use internally, day to day, that you built yourselves? Silence. Then a vague answer about “various tools we recommend.”
That silence told me more than the case studies did. An agency that sells AI implementation but hasn’t wired up its own operations with the tools it’s selling you is a marketing company wearing an AI badge. I have written before about the difference between people who talk about a thing and people who have done the thing, and the same test applies here. Ask to see their internal setup, even a screen share of their own automation, their own AI-assisted proposal process, their own client onboarding. If they can’t show you anything real, they’re reselling a workshop they read about.
Get specific about what “AI agency” means for your problem
“AI agency” now covers three very different businesses, and they’re not interchangeable:
- Agencies that build custom tools (chatbots, internal automations, data pipelines) using APIs from OpenAI, Anthropic, or Google.
- Agencies that implement off-the-shelf platforms (HubSpot’s AI features, Salesforce Einstein, Zapier automations) and configure them for you.
- Agencies that mostly do training and strategy, helping your team adopt AI tools that already exist without building anything custom.
You need to know which one you’re talking to before you compare quotes, because a £15,000 custom build and a £2,500 configuration job will both be called “AI implementation” on a proposal. I keep a running list of the tools most small businesses need in my AI tool stack guide for business owners, and honestly, for maybe six out of ten small businesses I talk to, the answer is configuration and training, not a custom build. A good agency will tell you that. A hungry one will sell you the custom build anyway.
The uncomfortable bit: they need you to stay dependent
Here’s the part nobody puts in their sales page. Most agencies are financially better off if you never become self-sufficient. A retainer that runs forever is worth more to them than a project that hands over cleanly in month three. This isn’t a conspiracy, it’s just how the incentives sit, and it means you have to build the exit into the deal from the start, not negotiate it later when you’ve already paid a deposit.
Ask for a documented handover date in the contract, not “ongoing support available.” Ask who owns the prompts, the workflows, the custom GPTs, and the automation scripts once you stop paying. I’ve seen small businesses in the Midlands and in London both get stuck paying £800 a month for “maintenance” on a chatbot that needed about two hours of upkeep a quarter, because nobody wrote down at the start what maintenance meant in hours or deliverables.
Questions to ask before you sign anything
I run through the same list with every agency now, whether it’s for my own business or when I’m advising a client who’s about to sign something. Ask these in the first call, not the third:
- Who on the team will do the work, not just who’s on the sales call?
- Can I speak to a client from a similar-sized business, not just your biggest logo?
- What does the project look like if it goes wrong in week three, what’s the process?
- Do I own the outputs (prompts, automations, custom models) outright when we stop working together?
- What’s the fixed handover date, and what documentation do I get?
- How do you measure success, in hours saved, leads generated, or revenue, and how will we track it?
- What happens to my customer data during and after the project?
If an agency gets defensive about question four (ownership) or question five (handover), that’s your answer. Walk.
Pricing in 2026, and what’s reasonable
UK small business AI agency work generally falls into three bands right now. A single automation build, say a customer service chatbot connected to your CRM or an email sorting workflow, typically runs £2,500 to £6,000 as a one-off project. A broader implementation, covering multiple workflows plus staff training, sits between £8,000 and £25,000 depending on how many systems need connecting. Ongoing retainers for monitoring and iteration run £500 to £2,000 a month, and that’s where you need the clearest scope, because “monitoring” can mean fifteen minutes or fifteen hours depending on who’s writing the invoice.
If a quote comes in wildly under these ranges, ask what’s being cut. If it comes in wildly over, ask why, and get a breakdown by deliverable rather than by day rate. I’ve turned down two proposals in the last year purely because the agency wouldn’t itemise the work, just gave me a total figure and a vague scope document. That’s not a pricing problem, it’s a transparency problem, and it tends to show up again later in the project.
What went wrong the time I got it wrong
In 2024 I brought in an agency to build a lead-scoring workflow for a client’s website, connecting form fills to an AI model that would rank enquiries by likelihood to close. £4,000, six-week timeline, sounded sensible. The build worked fine in testing. What nobody documented was how the scoring logic had been trained, so when we wanted to adjust it three months later after the client’s product mix changed, there was no record of the original prompts or the weighting decisions. We had to rebuild from scratch, at a similar cost, because the “documentation” the agency provided was a two-paragraph email.
The lesson wasn’t that AI agencies are bad. It’s that documentation is not optional, and you have to ask for it as a deliverable, named in the contract, with a format specified (a shared doc, a Notion page, whatever works, but something you can hand to someone else). I don’t sign anything now without that line item.
Check how they talk about failure and iteration
Anyone can show you a slide with a 40% efficiency gain. Ask what didn’t work on their last three projects. An agency that’s honest will tell you a chatbot flopped because the client’s customers preferred phone calls, or a content workflow needed three rounds of retraining before it sounded like the brand. That kind of honesty in a sales conversation is rare and it’s the strongest signal I’ve found that you’re dealing with people who do the work rather than people who sell the work. It’s the same instinct David Ogilwy built his agency on, tell the client the truth even when it’s not flattering, and it still separates the agencies worth paying from the ones chasing a logo for their homepage. I wrote more about that approach in my piece on business lessons from David Ogilvy, and it holds up just as well for AI agencies as it did for advertising.
Size the agency to your problem, not your budget
A ten-person agency with enterprise clients is not going to give a £3,000 project the attention it needs, no matter how nicely they talk on the call. Junior staff will run it, senior staff will check in twice. A one or two person specialist, or a solo AI consultant working directly with small businesses, will often deliver more for less because you’re the priority client rather than the smallest one on the roster. Ask directly who your account will be handled by day to day, and get their name, not just the founder’s.
Test them on something small first
Before any six-week engagement, ask for a paid pilot: one workflow, one week, a fixed low fee (£300 to £750 is common). This tells you three things fast: how they communicate under a deadline, whether the output is usable or needs heavy rework, and whether the relationship feels collaborative or like you’re being sold to at every check-in call. I’ve never regretted insisting on a pilot. I have regretted skipping one.
Watch how they explain AI limits, not just AI wins
Larry Page built Google on the principle that the best products solve real problems rather than chase trends, and that instinct is worth borrowing when you’re vetting an agency. Anyone who tells you AI will fix everything, with no caveats about hallucination, data quality, or the need for human review, either doesn’t understand the technology or is choosing not to mention its limits because it’s easier to sell you the dream. I go into more of that thinking in my notes on business lessons from Larry Page, and the same discipline, building for the real problem rather than the flashy demo, is exactly what separates agencies worth your money from the ones optimising for the sale.
Match the agency to your team’s readiness, not just the tech
The best AI agency in the world will not fix a team that refuses to change how it works. Before you hire anyone, be honest about whether your staff will use what gets built. I’ve seen a beautifully built AI proposal-writing tool sit unused for four months because nobody trained the sales team on it, and the agency had already moved to their next client. Ask upfront: does the fee include training sessions for my staff, and how many, and what happens if adoption is slow three weeks in? If the answer is “that’s not really our department,” you need a different agency, or you need to bring in a dedicated AI implementation coach alongside the build to handle the human side separately.
Don’t skip the boring parts of due diligence
Check Companies House for how long the agency has existed, not how long their website copy implies. Ask for two references you can call, not just written testimonials. Read their case studies for specific numbers, not adjectives, if a case study says “significantly improved efficiency” with no percentage or hours figure attached, be sceptical. Claude Hopkins built an entire career proving that vague claims sell nothing and specific, testable numbers sell everything, and that principle is worth remembering when you’re reading agency marketing rather than writing your own. I cover more of his approach in my piece on business lessons from Claude Hopkins, and it’s a useful filter for reading any agency’s homepage with a more critical eye.
Know when an agency is the wrong hire entirely
Sometimes the honest answer is that you don’t need an agency at all. If your problem is one workflow, say sorting inbound leads or drafting first-pass social copy, a few hours with a consultant or even a well-structured course might solve it for a fraction of the cost. Agencies make sense when you’ve got multiple systems to connect, real data to secure, or a team that needs structured training over weeks rather than a single afternoon. If you’re weighing formal routes into this kind of strategic thinking generally, it’s worth a look at how I broke down the actual outcomes of postgraduate business study in my piece on what MBA programmes really deliver, because the same “what will this get me” question applies to hiring an agency as it does to a £20,000 degree.
Bringing it together
Choosing an AI agency well comes down to a small number of unglamorous checks: what they’ve built for themselves, whether they’ll document the work in a format you can hand to someone else, whether ownership and handover are written into the contract rather than promised on a call, and whether the pricing is itemised rather than a single vague number. None of that shows up on a polished pitch deck. All of it shows up three months later, when the contract’s up and you find out whether you got a partner or a subscription you can’t quite cancel.
Frequently asked questions
How much does an AI agency cost in the UK in 2026?
Single automation builds typically run £2,500 to £6,000, broader multi-workflow implementations with training run £8,000 to £25,000, and ongoing retainers sit between £500 and £2,000 a month. Get every quote itemised by deliverable, not given as a single total.
What’s the difference between hiring an AI agency and an AI consultant?
An agency typically builds and manages ongoing systems with a team behind the work, while a consultant usually works directly with you on strategy, training, and smaller implementations, often at a lower cost and with more direct access to the person doing the thinking.
How long should an AI implementation project take?
A single automation typically takes two to six weeks from kickoff to working handover. Multi-system implementations with staff training generally run eight to sixteen weeks. Anything quoted at “a few days” for a genuine custom build should raise questions about what’s being skipped.
Should I ask an AI agency for references before signing?
Yes, and ask for two you can call directly rather than written testimonials on their site. Ask the reference specifically about documentation quality, communication under deadline pressure, and what happened after the contract ended, since that’s where most agency relationships quietly fall apart.
Sources worth reading
Related reading: Are Online Transcription Courses Worth It for Beginners? and How Do You Decide Between Full-Time and Part-Time Remote Work.
For the bigger picture, see my full guide to AI marketing.
