Do You Need an AI Agency? An Honest Answer for UK Owners

I make my living telling business owners they do not need an AI agency. Most of my working week is spent showing someone how to build the thing themselves in Claude Code, so the obvious move here would be to tell you to keep your money and read my other posts. I am not going to do that, because it is not always true.

Some firms should hire an AI agency. Fewer than the agencies think, and far fewer than the adverts between YouTube videos suggest, but the number is not zero. What follows is the version I would give a friend who had just received a quote: what an AI agency actually sells, what the alternatives cost in the UK, the handful of situations where paying one is the right call, and a ten minute check you can run before you reply to that email.

Do you need an AI agency?

You need an AI agency when two things are true at the same time: the work you want automated is already named and repeated, and nobody inside the business can build or own the result. Scale pushes you further toward an agency. So does a process that touches customers at volume. If either of the two conditions is missing, you almost certainly do not need one yet.

That is the whole answer. The rest of this post is about working out which side of it you sit on, because the honest position for most UK firms with ten to fifty staff is a little uncomfortable. The first automation is usually small enough to do in house. The agency pitch is sized for the fifth.

What does an AI agency actually sell you?

An AI agency sells you a team’s time, packaged. Under that one label you will find four quite different business models, and knowing which one is on the table tells you most of what you need to know about the invoice.

1. Project builds. A paid discovery phase, then a fixed scope. The build itself is usually a Make or n8n scenario, a Zapier chain, a custom GPT or Claude project, or a voice agent sitting on Twilio or WhatsApp. One UK agency, AI Agency Plus, publishes its ranges: discovery at £500 to £2,000, a first automation at £1,500 to £5,000 over one to three weeks, and a departmental build at £5,000 to £20,000.
2. Retainers, often called care plans. The same page lists support at £350 to £3,000 a month. What you get for it is monitoring, plus fixes when Xero or HubSpot changes something. Small tweaks on request are usually included.
3. Productised subscriptions. An “AI receptionist” or “AI employee” at a flat monthly price. Softomate, for example, offers an AI phone assistant from £299 a month. You rent the output and never see the build.
4. White label resale. A Voiceflow bot or a GoHighLevel setup with the agency’s logo on it. The margin on the tool is thin. The real product is the configuration, which is worth knowing when you see the price.

Behind all four, the thing you genuinely pay a team for rather than a person is cover. If the one freelancer who built your invoice chaser is on a beach in Crete when it breaks, an agency has someone else who can log in on Monday morning. That has real value. It also has a real price: UK operator benchmarks put freelance day rates around £500 to £800 and agency day rates around £1,000 to £1,800. The gap is the account manager and the second pair of hands. Decide whether you want those before you pay for them.

What are the alternatives to an AI agency?

There are four realistic routes for a UK business, and most owners only ever hear about one of them.

A freelancer is one senior person with no account manager in between. Cheaper per day and often better per day, because you get the person who actually builds. The weakness is availability. When your Make scenario stops firing at 9am on a Tuesday, you are waiting on one person’s calendar.

An in house owner is not a new hire. It is the operations person you already employ, given half a day a week and a Claude Team seat at about $25 a month. The first automation takes them weeks. The fourth takes them an afternoon. If you intend to automate more than two or three things over the next year, this route compounds in a way no retainer ever will.

Building it yourself with Claude Code sounds like the developer option. It is not. You type what you want in plain English, something like “every morning, read the new PDF invoices in this folder, pull the supplier, amount and due date into this spreadsheet, and flag anything over £5,000”, and it writes and runs the script while you watch. The paid Claude plans start at $20 a month ($17 on an annual deal) and include Claude Code. The honest limit is your time and your patience with the first one. Getting owners past that first one is most of what I do, and you can read how I work.

Here is the figure that reframed this for me. In the government’s own AI Adoption Research, among UK firms using language AI, 71% buy off the shelf tools and only 9% outsource the development. Most British businesses that already use AI never hired anyone to build it.

When does hiring an AI agency make sense?

Hire an AI agency when the cost of the automation failing lands on a customer rather than on your afternoon, or when the build is genuine engineering rather than configuration. Four situations pass that test.

The work faces customers at volume. A voice agent handling 300 appointment calls a day for a group of dental practices is not a weekend project. If it mishears a date, a patient turns up to a locked door. You want someone whose job is to watch it.

It crosses several systems that were never designed to talk. An ERP, a warehouse scanner, a Shopify store and Xero, with stock levels that have to agree across all four. That is real integration work with real edge cases, and a team that has done it before will save you months of discovery.

Nobody inside can own it, and you have decided that is fine. Some owners genuinely do not want an automation skill in the business, in the same way they do not want a payroll skill. They want the outcome, delivered, like a utility. That is a legitimate choice as long as you make it on purpose.

You need it live in eight weeks. A seasonal peak is coming, the current process will not survive it, and the learning curve is a luxury you do not have this quarter.

The same logic tells you when a retainer is right, which is something most advice gets wrong by treating every retainer as a trap. A retainer on a chatbot that takes bookings at 2am is insurance. A retainer on a script that renames PDFs is a subscription to someone else’s convenience. Only the first kind is worth a monthly fee.

When is an AI agency the wrong buy?

An AI agency is the wrong buy when you cannot yet finish the sentence “every week we spend about X hours doing Y by hand”. No agency can finish it for you, and the good ones will send you away until you can. The ONS found in July 2026 that even among UK businesses using AI, only one in ten use it extensively, and the pattern I see behind that number is firms that bought a tool before they had a task. If you want a quick way to put a number on the hours first, my free AI Opportunity Scorecard takes two minutes and will tell you whether there is enough there to be worth anyone’s fee.

It is also the wrong buy when the job is configuration dressed up as development. Connecting a web form to a CRM and sending a templated follow up is a two hour job in Zapier. Some agencies will quote it as a bespoke build, because the label “AI” lets them.

There is a reason so many agencies are suddenly pitching you. The “AI automation agency” model was popularised in 2023 by Liam Ottley, a UK founder whose YouTube channel has around 816,000 subscribers and whose free online community has passed 280,000 members. A large share of the agencies now emailing UK firms were founded after watching those videos. Nothing wrong with being new. Plenty wrong with learning on your invoice.

Five red flags that are specific to agencies, as opposed to consultants in general:

• The demo you are shown is the same demo every prospect gets. Ask to see something built for a business nothing like yours.
• The founder sells and someone on Upwork delivers. Ask, by name, who will log into your accounts and build this.
• No price until you have sat through a “strategy call”.
• Hours saved are promised before anyone has seen your inbox.
• The case studies have no company names in them.

What should you ask about leaving before you sign?

Ask what you would be holding if the agency closed tomorrow, because small agencies do close, and acquired ones change their pricing. The answer should be a handover pack, delivered as part of the first build rather than promised at the end of the relationship.

A proper pack contains the exported workflow files (a Make blueprint or an n8n export is a single JSON file), every prompt written out in a document you can open, a list of each API key and where it lives, the monthly running cost of every tool involved, and a short screen recording of someone fixing a failure. If an agency hesitates at the screen recording, you have learned something useful about how often things fail.

Also ask for a named second person. Not a team, a name. The whole premise of paying for a team is that there is one.

How do you decide in ten minutes?

Score your situation against the six questions below, then read off the route. It is deliberately blunt. If you cannot answer the first question, stop, because none of the others matter yet.

Two notes on reading the result. A high score says an agency is worth a conversation, not that you should sign the first proposal. And a retainer only earns its place if the overnight question was a yes.

Where I would start

Before you talk to any agency, including me, write one page. The task, the hours it eats each week, the systems it touches, and the name of the person who will own it afterwards. If that page takes you an hour to write, you have learned that you are not ready to buy anything yet. If it takes ten minutes, send it to whoever you are considering and watch how they respond. Anyone who quotes against that page, instead of trying to book you onto a strategy call, deserves a second conversation.

My own view, having sat on both sides of these conversations, is that most owners who search for an AI agency need a first automation and a bit of confidence, not a team. That is what the AI Opportunity Audit is for: a fixed fee £500 session of 60 minutes where we go through your processes, pick the place where automation pays back fastest, build a real one live in Claude Code while you watch, and you leave with a prioritised roadmap. It starts with a free 15 minute call, and if that call tells me you are a four pointer who genuinely needs a team, I will say so and point you at the questions above.