Questions to ask before hiring an AI automation agency or consultant in the UK

What does an AI automation agency or consultant actually do?

They work out which of your repeat jobs software can take over, build the workflow, and keep it running. In practice that means connecting the tools you already use (email, accounts, a spreadsheet, a booking system), adding AI where something needs reading or writing, and setting up checks so a person sees anything unusual.

The words overlap. A consultant usually advises and sometimes builds. An agency usually builds and sometimes advises. For a small business the label matters less than a simpler question: who will do the work, and what will you be left with when they have finished?

A disclosure before the rest. I run FTW Digital, which does this work for small UK businesses, so I am not a neutral party. The last section shows how I would answer the same questions, so you can judge my answers against the checklist.

It also helps to know you may not need anyone. If one job is eating your week and a feature in software you already pay for covers it, switch that on first. If you are not sure where to start, a short paid review of your week can answer that before you commit to a build. I cover how to choose a first job in the AI for accountants guide, and the same logic applies to most small firms.

What should you ask before you hire an AI automation agency?

Ask about six things: who does the work, who owns the result, where your data goes, what it costs over a year, what happens when something breaks and how mistakes are caught. The table gives each a question, a reassuring answer and a warning sign.

None of this needs technical knowledge. It needs someone willing to say the answers out loud and then write them down.

Who will actually do the work?

Ask for the name of the person who will build it and the person who will fix it. They may be the same person, which in a small business is often better. It is a problem only if the pitch is delivered by one person and the work by someone you never meet.

• Who builds it, and who fixes it?
• Will anyone outside the agency see my accounts or documents?
• What happens if that person is ill or busy?
• Can I speak to a past client?

Some agencies use subcontractors or offshore teams. That is not wrong, but it raises a data question as well as a quality one, so ask whether anyone outside the agency will see your accounts or documents. If they will, ask which of them, from where and under what written terms.

Ask what happens if the person is ill, on holiday or busy. A small provider cannot promise cover they do not have. What you can ask for is documentation clear enough that another competent person could pick the work up, and your own copy of the instructions and access.

References are the usual test, and they are fair to ask for. If a provider cannot offer one, ask why. A new business may have a good reason, such as client confidentiality, and a straight answer beats an invented testimonial.

Who owns what is built, and can you leave?

Ask whose account the automation runs in. If it runs in an account in your name, you can see it, pay for it directly and move it to someone else. If it runs in the agency's account, you are renting it, which can be fine as long as you know that and have agreed what happens if you stop paying or they stop trading.

• Whose account does it run in?
• Who pays the platform and AI subscriptions?
• What do I get at the end?
• What does leaving cost?

Most small-business automation is built on a workflow platform such as n8n, Make or Zapier, plus an AI service from a provider such as Claude or OpenAI. The platform and AI subscriptions are separate from the agency's fee. Ask who pays them, whose name they are in, and whether they are included in the quote or billed on top.

Then ask what you get at the end. A useful handover includes the workflows themselves, a plain-English description of what each step does, a list of every login and connection involved, and what to do if something stops. Ask to see an example of the handover document before you sign, with another business's details removed.

Finally, ask about lock-in directly. Is there a minimum term, a notice period, a fee to export your workflows or a charge to hand over the logins? A clear provider will answer all four in one sentence.

Where does your data go, and who is responsible for it?

Ask for a plain list of every service that will receive your data, what each receives and where it is processed. That should include the workflow platform, the AI provider and any hosting, and say whether any of it leaves the UK.

If the work involves personal data about your customers, clients or staff, the agency will usually be your processor: it handles the data on your instructions, while you stay the controller. The ICO says a written contract is required whenever a controller uses a processor, and that UK GDPR sets what it must include (ICO, accessed 26 October 2026).

The ICO lists terms such as acting only on your documented instructions, keeping data secure, using sub-processors only with your prior authorisation, helping with breaches and deleting or returning the data at the end. It also says to use processors that give "sufficient guarantees". Ask whether the agency will sign such a contract, and read it first.

A processor must also tell the controller about a breach without undue delay, and transfers outside the UK need the controller's authorisation. Ask what the agency's breach process is. Full detail belongs in a separate data protection guide, so this is only a pointer. For private AI that keeps data in a controlled set-up, see private AI.

Ask about AI training too. Providers often treat personal and business plans differently, and the terms change, so ask the agency to confirm the current terms in writing. The AI for accountants guide has a table of where the main providers stood in October 2026.

One more question catches the commonest leak. Ask how staff will be stopped from pasting customer details into a personal AI account. Even a short written rule is a good answer.

What does it cost, and what keeps costing?

Ask for the cost in three lines: build (one-off), running costs (platform and AI subscriptions) and support (someone watching and fixing it). A quote that rolls them together makes providers hard to compare.

Ask whether the build price is fixed or an estimate, and what counts as a change. Day rates can be fair too, but ask for a cap so a discovery phase has an end.

I have not quoted an average market price here. Published "best agencies" lists are usually written by the agencies, so treat their figures as marketing. Get the same brief priced by two or three providers and compare.

Then ask what would make the cost go up. Typical causes are an app changing its connection, volumes growing, or a second job being added to the first.

Last, ask what return to expect and how it would be worked out. A sound answer starts with hours or errors you can measure now, such as time spent keying in receipts, not a promised percentage.

What happens after it goes live?

Software that connects to other software needs looking after. Apps change how they connect, logins expire and a document arrives in a format nobody expected. None of this is unusual, and none of it is a sign of a bad build. It is why you need a clear answer to "what happens in month four?" before you start.

• Who finds out first when it fails?
• What does the support fee cover?
• What happens if I decline support?

Who monitors it? Does the provider find out when something fails, or do you find out when a customer complains? Get the response time in writing as a target rather than a guarantee, because nobody can promise a fix time for a fault they have not seen.

The support fee should cover monitoring, fixing faults, keeping up with changes to connected apps and a little time for tweaks. Find out what falls outside that, such as a new job, and whether it is priced before work starts.

If you decline support, you should still have the documents and logins to fix things yourself, or the option of a one-off paid fix. You should never be forced into support just to keep your own system running.

How will it handle mistakes and edge cases?

Ask what the AI is allowed to do on its own and what has to wait for a person. AI is good at reading and drafting. It can also be confidently wrong, so anything that goes to a customer, touches money or changes a record should start with approval, and loosen only when you have seen it work.

Ask to see it working on your own examples before it goes live. Demo data is always tidy. Your real receipts, emails and enquiries are not, and the way a system copes with a blurry photo or an odd request tells you far more than a polished demo.

Can you see what it did and why, for example in a log of each action? Can you switch it off in a minute? What happens to a job it was halfway through? These matter more than accuracy claims, because every system makes some errors and the question is how quickly you spot them.

A useful last question is what the provider thinks should not be automated. A good one has a list. If the answer is "anything", you are talking to a salesperson.

What are the warning signs when choosing a UK AI automation agency?

These sit outside the six questions. Watch for promises of specific savings before anyone has looked at your business, pressure to sign quickly, and a tool-first pitch where the platform is chosen before they hear what your week involves.

Be careful with "best AI automation agency" lists. Many are published by agencies, and some place their own business near the top. That does not make them useless, but check who wrote the list, whether they disclose it and what the ranking is based on. A list with no stated method is advertising.

Be wary of anyone who offers legal, tax or regulatory assurance they are not qualified to give. An AI provider can tell you how your data moves and can document it. Whether that meets your professional obligations is a question for your regulator, professional body or adviser.

• A fixed saving or return promised before they have seen your processes.
• No named person, no references and no explanation for either.
• Everything runs on their platform with no exit terms.
• The only price is "it depends", with no ranges or example lines.
• No answer on where your data goes beyond "it is secure".

How would FTW Digital answer these questions?

Here is how I would answer the checklist, limited to what I can state as fact today. Ask me anything else on the call.

If you want somebody to look at your week before you spend on a build, the AI audit is a written plan showing what AI and automation could save. For the full range, see AI automation services and pricing. The home page covers AI consultancy in the UK.

If you want somebody to look at your week before you spend on a build, the AI audit is a written plan ranking what could be automated by the hours it could save. If you would rather talk first, the free 20-minute call covers the same ground. The pricing page shows how the costs are laid out.