Three articles ago. I talked about how you were prompting ChatGPT at the kitchen table and calling it automation. And that this isn’t using AI to its fullest. But how do you get an AI agent live inside your business?
Since then, we’ve chatted about the difference between a chatbot and an AI agent.
Discussed 6 specific AI agents that can do real work in businesses like yours. And last week, we looked “under the bonnet”.
The 3 parts of an AI agent: trigger, brain, hands, and why “I’m not technical” was never the barrier.
Which leaves one question.
The practical one.
The one that actually matters:
How do you get from reading about this to having one running in your business?
How to pick the first AI agent to build.
What happens at each stage, what you do versus what’s done for you, how long it takes, and what it costs.
I’m hoping by the end of this article, you’ll have everything you need to make a start.
First, the Mistake to Avoid
I’ll save you from the error I see most often.
An excited business owner reads about AI agents.
Sees the possibilities.
Decides to automate everything. Enquiries, quotes, invoices, bookings, reporting, the lot.
In one glorious project.
Then, 6 weeks later. He feels overwhelmed; nothing gets finished. Everyone’s lost faith, and “that AI stuff” gets filed next to the CRM that nobody ever filled in.
Start with one agent. One workflow. One clear win.
Get ONE AI agent live, no matter how “basic” it is.
Let it run for a few weeks.
For example, you and your team can see say follow-ups going out without anyone touching them.
See how it feels when a chunk of admin, well, stops being your “job”.
Then add the second.
You’ll start growing in confidence. Not only are the AI agents running. But also the fact that you can keep adding more into your business.
You’ll also see a growth in how many hours you save each week
How to Pick Your First AI Agent To Install
In part 2 of this series. I broke down 6 AI agents. But as I’ve mentioned. You don’t want to get all 6 up and running in one go.
How do you decide which one to start with?
Here’s the filter I use with every business I work with.
Your first agent should be:
High-frequency.
Something that happens weekly at least, better yet, daily.
An AI agent that runs twice a year improves nothing.
An agent that runs twenty times a week gives you feedback and a return on investment in less than a fortnight.
Rules-based.
You can describe the job in a few sentences.
“Quotes get chased at day three and day seven; stop if they reply.”
If explaining the task requires an hour and fourteen exceptions. It’s not the first one you should set up. It’s for further down the line. If ever (if it’s that complicated)
It’s annoying.
Pick something you or your team resents doing.
I’m going off script here.
But I remember ringing a lead. The gentleman complained that his staff keep saying to him, “I don’t want to do that”.
He wasn’t happy.
When that task disappears gets handled by an AI agent. Everyone notices; they are all happy about it. And everyone’s on board for what’s next.
Measurable.
Pick one with money attached.
Why this KPI?
Well, AI agents can save a shit ton of time, but that can be harder to gauge starting. But an uptake in revenue is easier to spot and then track.
Which is why, for most service businesses, the first AI agent is one of:
The quote follow-up agent
Because unchased quotes are revenue sitting in your sent folder. And the conversion history is easy to find, track and review.
The invoice chasing agent
Because getting paid faster is the most unambiguous win in business.
Nobody’s ever asked me to prove the value of money arriving sooner.
Both are high-frequency, both run on rules you already follow in your head. And both pay for themselves in a way you can see on a bank statement.
5 Phases To Get Your First Live AI Agent
I’m going to use the quote follow-up agent I’ve spoken about.
There are five distinct phases to get the agent live.
Here’s what they look like, start to finish.
Phase 1: Map (Week 1)
One call. That’s the start of it.
We book a call, done online, which takes around an hour or so.
And I get the rules, everything out of your head.
Because here’s the thing I said in Part 3: everything an AI agent does is a rule you already follow.
This call is where we document everything and produce a report for you.
For the quote follow-up AI agent:
- When do you consider a quote “gone quiet”? Three days? Five?
- What do your follow-ups actually say? (I’ll want to see real examples of emails you’ve sent — that’s how the agent learns your voice.)
- When do you stop chasing? After two nudges? Three?
- What should never be chased? (The mate you quoted as a favour. The big client with a slow finance department.)
- What happens when a client replies — who needs to know, and how fast?
It’s surprising how many steps make up this task. And how quickly you were able to get it out of your head.
You’ve been running this process for years. But you’ve never had the capacity to document it before.
My job is to ask the questions that surface these steps and the exceptions. The “oh, unless it’s” moments. Because those exceptions are where sloppy automation falls over. But professionally built AI agents don’t.
What you do: One call, and dig out a handful of past emails.
What we do: Turn it into a build specification in the form of an Automation Opportunities Report
Phase 2: Build (Week 1–2)
This is where we build the AI agent and connect it to your tools.
The trigger gets wired to your email and CRM, so it knows when quotes go out and when replies come in.
The brain gets configured with your rules and trained on the tone of your email.
The hands get connected, with security a priority. Through the official routes your software providers built for exactly this. Connecting to your email, your CRM, your calendar.
The guardrails go in here too.
Hard limits on what the agent can touch.
The approval steps we agreed. And the logging that records every action it ever takes.
What you do: Approve a couple of access permissions (takes minutes. It’s the same “allow this app” process you’ve done a dozen times).
What we do: Everything else.
Phase 3: Test (Week 2–3)
The AI agent goes live, with stabilisers on.
It runs against real-world data, watching real quotes, drafting real follow-ups.
But nothing happens without you or your team in the loop.
Every email it wants to send lands in your drafts folder, or comes to you for a one-tap approval.
You read what it wrote. Tweak anything that doesn’t sound right. And we can use those tweaks to refine the agent.
This phase does two jobs at once.
It catches those edge cases that are difficult to surface, while the cost of a mistake is zero.
And — just as important — it builds your trust.
You watch it make good decisions, day after day, on real scenarios.
By the end of a week or two of approving drafts that needed no changes. Most owners are telling me to take the stabilisers off.
What you do: Ten seconds to review each draft, a few times a day, for a week or two.
What we do: Watch the agent, tighten the rules, refine the tone from your feedback.
Phase 4: Live (Week 3)
We take the stabilisers off.
At whatever pace you’re comfortable with.
Routine first follow-ups go out automatically, but second nudges still need your nod.
Then, all follow-ups are handled by the agent. But you get a daily summary of what the agent did. The level of oversight is a dial, not a switch, and it’s your hand on the dial.
From this point, the agent does the job without you.
Quotes get chased on schedule, in your voice, every time. And the first you hear about most of them is a client replying.
What you do: Read the replies.
What we do: Keep an eye on the agent and logs while it beds in.
Phase 5: Run (Ongoing)
Here’s the part most people never think about at the start.
And it’s the part that decides whether your agent still works in a year.
Your new AI agent connects to software that’s built by other people.
Your email provider, your CRM, your accounts package.
Those tools change constantly: updates, new versions. Changes to the settings, features moved around.
When they change, anything connected to them needs to move too.
An AI agent nobody’s watching doesn’t fail and then scream and shout about it
It fails silently.
The connection breaks on a Tuesday; no error lands in your inbox. The quotes don’t get chased, and it all goes quiet.
You find out three weeks later when you wonder why fewer emails are coming, and fewer and fewer jobs are coming in.
By then, how many quotes went cold?
If your average job is £4,800 and two of them died of silence. That’s the best part of £10,000 gone, invisibly, while everything looked fine.
So every AI agent we put live gets monitored, backed up, and we make sure it stays running.
WE know it’s broken before you do
Usually before it’s had the chance to cost you anything.
When your tools change, the AI agent gets updated to match. When your business changes. A new service, new pricing, a new team member who needs looping in.
The AI agent changes with your business.
What you do: Nothing. You get a random email explaining the issue, what it meant, and what changed.
What we do: Everything in that note.
The Timeline, All Together
Week 1: Map — one call, rules out of your head
Week 1–2: Build — agent built and connected
Week 2–3: Test — running with your approval on everything
Week 3: Live — stabilisers off, your pace
Ongoing: Run — monitored, kept working, moving with your business
From first call to a live AI agent in two to three weeks.
Most of that is us doing the work, not you.
Your total time in: one mapping call, a few days of ten-second approvals. And then reading replies from clients you didn’t chase.
What It Costs For A Live AI Agent
I’m going to be open and transparent here.
No hidden costs or fees.
Or trying to get you to commit to something without understanding the investment. Both upfront and monthly.
The build.
A first AI agent, mapped out, built, tested, live. Sits in our Foundation tier. Roughly £3,000–£5,000.
The price depends on how many tools it connects to. And how many rules and exceptions your process carries. The more complicated it is, the more expensive the build will be.
We require a 50% deposit up front to start the work. Then the remaining 50% when the AI agent goes live.
The running.
This is the monitoring, the fixes, the updates when your tools change. The refinements when your business does.
The build price includes running costs while we build and test the AI agent. So there are no hidden costs, and you won’t get a surprise invoice before you’ve even seen the agent in action.
Once we hand over the AI agent and it’s live. Starting from the 1st of the next month. It’s £150 a month for as long as the agent’s earning its keep.
Is £150 a month worth it?
Is that the right question to ask?
What happens and how much does it cost if the AI agent stops working for an entire month? And you haven’t even realised?
Go back to those two £4,800 quotes that died because the broken agent didn’t send the follow-ups.
That’s the comparison.
One prevented failure, funds years of running costs.
And as you add agents, it scales the same way.
Each one is £150 a month. Each one hands back 2–3 hours a week. And the second build is usually faster than the first because your tools are already connected.
Compare it to hiring costs
Recruitment fees, salary, National Insurance going the way it’s going.
Holidays, sick days, and the six months before someone’s trained up.
An AI agent doing ten hours a week of rules-based admin for the price of a few hours of human time isn’t a close decision.
What Actually Changes
Strip out the technology and here’s what you’re buying.
The quote you sent on Monday gets chased on Thursday whether you remembered or not.
The invoice that hit day seven gets nudged while you were on site.
That enquiry that landed at 8 pm got answered at 8:04 pm. Your weekly report was in your inbox before you woke up on Monday.
And the 9 pm kitchen table session
The one this entire series started with — gets shorter.
In 2 or 3 months, you’re shutting your laptop at 7 pm
Then it stops happening at all.
Not because the work stopped existing.
But because it stopped being yours.
That’s the actual product. Not the software.
Getting you time and life back
Next Steps: Get An AI Agent Live
I hope you’ve stayed the course and read all four parts.
You now know the difference between a chatbot and an AI agent.
You’ve seen how 6 different AI agents work inside service-based businesses like yours.
You know what’s under the bonnet, how an AI agent works. And now you know the route from where you are now to getting one live.
If you’re still reading, you’re not wondering whether — you’re wondering which one first.
That’s exactly the conversation to have.
Book a call, bring your messiest workflows. And we’ll map out which agent earns its place in your business first.
Two to three weeks from that call. You have an AI agent in your business that is chasing, replying, logging and remembering.
And the best part, it isn’t you.
