Services 01 Projects 02 About 03 Insights 04 Contact 05 Book a discovery call 06

AI consultancy ·

AI Automation Agency UK: How to Choose One, and the Questions That Expose the Weak Ones

There are now hundreds of UK agencies offering AI automation, most of them under two years old. Here is how to tell which ones will still be answering the phone when something breaks, and what to ask before you sign.

The barrier to calling yourself an AI automation agency is a Make account and a landing page. That is not cynicism; it is why choosing one is hard. The tools are the same everywhere. What differs is whether the agency understands your business before it touches a tool, whether it builds things that survive contact with real data, and whether it is still there in month six.

What the good ones do differently

  • They map before they build. A good agency's first deliverable is a picture of where your hours go and which tasks should be automated first, with numbers. If the first deliverable is a proposal, they are guessing.
  • They say what not to automate. A process that is broken, inconsistent or rare should not be automated yet. An agency that automates everything you ask for is selling hours, not outcomes.
  • They keep a person in the loop where it matters. Anything that sends money, files a return, or messages a client gets a human approval step. "Fully autonomous" is a sales line, not an architecture.
  • They build on the APIs, not the screens. Screen automation breaks when the vendor changes a button. API builds keep working.
  • They monitor what they built. You should get an answer to "how will I know if it stopped working?" that does not involve you noticing.
  • They can show you it running in their own business. An agency that automates for others but runs on spreadsheets and memory has not tested its own advice.

The pricing models, and what each one tells you

  • Per workflow, fixed price. The honest default for a defined build. Expect £2,500 to £15,000 depending on what it touches. Ask what is included after go-live.
  • Monthly retainer. Right when you want ongoing design and building; wrong when you want one thing done. Watch for agencies that make everything a retainer. Plenty of automation is build once.
  • Per-workflow subscription ("from £100 a month per workflow"). Attractive for the first three; expensive at twenty, and you own nothing.
  • Diagnostic first, then build. A paid audit (ours is £997) that produces the map, then a quoted build. This is the model that gives you the most information before the most money.
  • An installed system with a monthly service. Everything connected, a daily brief, a new automation each month. This is what an AI Operating System is, and it suits a business that wants the whole operation handled rather than one workflow.

Warning signs

  1. No published prices and no willingness to give a range on the first call.
  2. Case studies with no names and no numbers, or numbers that are suspiciously round.
  3. A single platform recommended before anyone has looked at your business.
  4. "Fully automated" as a selling point for anything involving money or clients.
  5. No answer to "who do I call when it breaks?".
  6. The agency cannot show you its own systems running.

Twelve questions for the first call

  1. What did you build last month and what does it do today?
  2. Can you show me one of your own automations running?
  3. How will you decide what to automate first in my business?
  4. What would you tell me not to automate?
  5. Where does a person approve things in what you'd build?
  6. Do you build on the APIs or on the screens?
  7. What happens when it fails, and how will I know?
  8. Who owns the code, the documentation and the data?
  9. Which AI providers do you use, on what tier, and do they train on my data?
  10. What does the first build cost, all in, and what is the monthly after?
  11. What did your last client say three months after go-live?
  12. If I stop working with you, what keeps running?

What a good first project looks like

Small, measurable, and on the path to the next one. The AI email responder we built for a premium dog-training business is the pattern: every enquiry that lands in the inbox is read, classified and answered with a drafted reply in the owner's voice, which the owner approves with one click. £3,000, a few weeks, and the owner stopped losing enquiries to a slow reply. It touched one system, had one human gate, and produced one number (enquiries answered the same day) that proved it worked. The second project, once the first was trusted, was bigger.

Compare that with the first project a weak agency proposes: a "full automation suite" across five systems, no single number to judge it by, and a retainer from day one. If the first project cannot be described in one sentence with one measurable outcome, make it smaller.

Red flags in the proposal itself

  • No section on what happens when it fails. Every automation fails eventually: an API changes, a token expires, a client sends something unexpected. A proposal that does not mention monitoring is describing a build that will stop silently.
  • Tools named before problems. "We will implement Make, n8n and OpenAI" is a shopping list. "We will cut the four hours a week your account managers spend on reporting" is a proposal.
  • Ownership missing or vague. If the proposal does not say you own the code, the documentation and the data, you do not.
  • Hours saved with no method. "Save 20 hours a week" without a before-measurement is a number somebody typed. Ask how it was measured.
  • AI provider unnamed. You need to know whose model reads your client data, on what tier, and whether it trains on it. "Our AI" is not an answer.

We answer all twelve on the AI consultancy page and on the discovery call, and we published what nobody tells you about buying AI automation for the parts agencies do not volunteer.

Dave Crawford, founder of Anco AI
Dave Crawford

Founder, Anco AI. Builds the systems this site talks about, and runs his own business on them.

Fair questions.

A defined build is typically £2,500 to £15,000 depending on what it touches; retainers run from about £3,500 a month; per-workflow subscriptions start around £100 a month each. Published prices are rare, which is why we publish ours.

It rarely matters for the build, which is delivered remotely either way. It matters for trust: being able to meet, and the agency being in your time zone when something breaks. Anco AI is in Manchester and works across the UK, US, Canada and Europe.

An agency builds what it is told; a consultancy decides what to build. The deciding is the hard part for a small business, so the useful firm does both. Ask whether the person who maps your week is the person who builds the system.

Keep reading.

Book a free discovery call.

Whether it's AIOS, consulting, software or automation, we'll map where the time and money are leaking and what would give you the most back. No obligation, you keep the map.

Book a discovery call →