An AI automation agency that does the work, with a person still signing off.
We find the business processes that eat your team’s week, build the automations that take them over, and keep a human approval step wherever a mistake would cost you. Built in your accounts, documented, and yours to keep.
- Problem
- 2 people rewriting more than 100 adverts a week by hand.
- Built
- AI, automations and job data agents doing it end to end.
An AI automation agency designs, builds and maintains systems that take repetitive work off people, with AI doing the steps that used to need someone to read or write something. We are marketers who build: we map the process, measure it, automate it in your own accounts, and keep a person signing off wherever a mistake would cost you.
Most automation projects fail for a dull reason. Someone automated a process nobody had written down, measured nothing beforehand, and left no one in charge when it broke. We start with the process and the baseline, and we only call it done when the hours saved show up in your numbers.
A worked example: 300+ job adverts a week
At a national healthcare provider, 2 people were rewriting more than 100 job adverts a week by hand. It left them no time to think about whether any of it was working.
The fix wasn’t one tool. It was a system of AI, automations and job data agents that writes the adverts and publishes them. It went from 100 a week by hand to more than 300 a week, written and published automatically. The team got their week back.
What makes a build like this work
The process written down before anything is automated, a fixed brief for the AI to write to, and a person who can stop or change any output. Those 3 things matter more than which model or platform you pick.
That is the recruitment version. The same pattern applies to quotes, enquiry replies, reports, onboarding paperwork and anything else that arrives in volume and follows a shape.
What an AI automation agency does
The plumbing is workflow automation: forms, CRM, inbox, spreadsheets and finance tools passing data to each other without anyone copying and pasting. The AI sits inside those workflows, doing the steps that used to need a person to read something: classifying an enquiry, extracting fields from a document, summarising a call, drafting a reply.
Business process automation is the older name for the same job. What has changed is that AI can now handle the messy, text-heavy steps that rules-based automation couldn’t, which is why far more of a small team’s week is worth automating.
| Process | What the automation does | Where a person stays in |
|---|---|---|
| Lead capture and routing | Reads each enquiry, scores it, adds it to the CRM and sends it to the right person within minutes | Sales reviews the scoring rules monthly |
| Enquiry replies | Drafts a reply from your own past answers and product information | A person approves before anything is sent |
| Reporting | Pulls ad, CRM and finance figures into one weekly report with the changes flagged | The owner reads it; nobody builds it |
| Documents and admin | Extracts fields from invoices, forms and CVs into the system they belong in | Low-confidence items go to a review queue |
| CRM hygiene | Finds duplicates, stale deals and missing fields, and logs calls and emails automatically | Merges and deletions wait for approval |
| Recruitment workflows | Watches hiring signals, drafts adverts and outreach, and tracks applications to source | Recruiters approve every message and every shortlist |
Typical examples from UK service businesses. The right list for you comes from mapping your own week, not from ours.
What to automate first, and what to leave alone
The first workflow should be boring, frequent and easy to measure. We score every candidate process on 3 things: how often it happens, how long it takes each time, and what a mistake costs. High frequency, real time and a low cost of error goes first.
- Automate first: anything done the same way more than 20 times a week, anything that copies data between 2 systems, and any report somebody rebuilds by hand every Monday.
- Automate with an approval step: anything that speaks to a customer or candidate in your name, spends money, or changes records you rely on.
- Leave alone for now: a process that is broken or different every time (fix it first, or you automate the mess), low-volume work that needs judgement, and decisions about people that would be made solely by a machine.
How a build runs
Every stage is fixed scope, and you can stop after any of them.
- Map the process. We write down how the work happens today, step by step, with the people who do it. This alone usually removes a step or 2.
- Measure the baseline. Time per task, volume per week and error rate, so the result can be judged against something real.
- Build in your accounts. n8n, Make or Zapier for the workflow, Claude or OpenAI for the AI steps, all running under logins you hold. You control the data and the running costs.
- Put the approval step where it matters. Anything that sends, publishes, spends or deletes waits for a person until you decide otherwise.
- Test on real data. We run it alongside the manual process first, with error alerts so nothing fails silently.
- Hand over. Written documentation your team can follow, a named owner on your side, and a review at 30 days against the baseline.
An automation nobody owns is a liability with a timer on it. Every workflow we build has a named owner, an error alert and a written procedure before it goes live.
AI automation and UK GDPR
Most automation touches personal data, so data protection is part of the design, not a sign-off at the end. Workflows run in your accounts, only the minimum data each step needs is sent to an AI model, and sensitive data stays on systems you control.
Decisions made solely by automated means that have a legal or similarly significant effect on someone, such as rejecting a job applicant or a credit application, carry specific safeguards. The Data (Use and Access) Act 2025 reformed these rules in new Articles 22A to 22D of the UK GDPR, in force from 5 February 2026. People must be told about such decisions, be able to ask for meaningful human review, and be able to contest them. Our default is simpler: a person makes those decisions, and the automation prepares the information. This is design guidance, not legal advice.
Why a marketing agency builds automation
Our roots are in marketing and recruitment marketing: getting the right people to enquire, apply and buy. Almost every problem we fix there ends in the same place, a process that depends on someone remembering to do something. So we build the machine as well as the campaign, and we judge it the way we judge a campaign: did it move the number?
Our own operations run the same way. Every month The SetpointHQ Index scores 100 UK employer brands on how AI assistants describe them, through an unattended pipeline that queries 4 AI models, versions its method and publishes the result. We build client automations to the same standard: logged, monitored and written down.
Agency, software, or doing it yourself?
| Factor | Do it yourself | Off-the-shelf AI software | AI automation agency |
|---|---|---|---|
| Who designs it | Whoever has time | The vendor, for an average customer | Built around your process |
| Fits your tools | If you can connect them | Only its own integrations | Connects what you already use |
| Who owns it | You, if it is documented | The vendor | You, documented at handover |
| When it breaks | You find out late | Vendor support queue | Error alerts and a named owner |
| Best when | One simple, low-risk workflow | A standard process many firms share | Several connected steps, customer-facing output or personal data |
If your automation need is recruitment content rather than a custom build, Studio is the self-serve route: add a job and it generates an 11-asset playbook across 6 stages, plus 3 Desk Radar tools, for £149 per seat per month, excluding VAT.
Common questions
What does an AI automation agency do?
It designs, builds and maintains workflows that take repetitive work off people, with AI handling the steps that involve reading or writing, such as triaging enquiries, extracting data from documents or drafting replies. A good agency also maps the process first, measures a baseline and hands the finished work over documented.
How much does AI automation cost?
It depends on how many steps and systems are involved, so we quote a fixed price after mapping the process rather than before. Running costs are separate and usually small: a workflow platform subscription and AI usage, both billed to your own accounts so you can see them.
Which business processes can be automated?
Anything frequent, rules-based or text-heavy: lead capture and routing, enquiry replies, weekly reporting, invoice and form processing, CRM updates and recruitment admin. Processes that change every time, or decisions with a significant effect on people, should keep a person in charge.
Which tools do you use?
Usually n8n, Make or Zapier for the workflow and Claude or OpenAI for the AI steps, chosen to fit your existing systems and budget. Everything runs in accounts you hold, so nothing is locked to us.
Is AI automation compliant with UK GDPR?
It can be, if it is designed that way. Send AI models only the data each step needs, keep workflows in your own accounts, and keep a person in charge of decisions with a legal or similarly significant effect. Since 5 February 2026, Articles 22A to 22D of the UK GDPR set the safeguards for solely automated decisions.
Do you only work with recruitment businesses?
No. Recruitment marketing is where we started, and many clients hire at volume, but the same automations serve any UK service business with enquiries, a CRM and a team short on time. We are based in Suffolk and work UK-wide.
Tell us the task that eats your week.
Describe how it happens today and roughly how long it takes. We’ll tell you whether it is worth automating, what the first workflow should be, and where a person needs to stay in the loop.
Sources: ICO, “What are the ADM safeguards?” (updated for the Data (Use and Access) Act 2025, draft guidance); Articles 22A to 22D commencement 5 February 2026, SI 2026/82. Checked 15 October 2026.