S-02 / 06

00 AI automation

AI automation that runs your busywork while you run the business.

You know which work eats your team's week. We build the AI workflows that take it on: the triage, the filing, the chasing and the reporting, done the same way every time, with a person approving anything that matters and a log of every step. We build them, host them and keep them running. They run on a trigger over your internal work, with no customer on the other end. Anything that talks to customers is an AI agent.

Book a discovery call See how we build
First workflow
Live in two to four weeks
Approval gate
On from day one
Quote
Fixed, after a discovery call
Ownership
Yours, with the handover pack
Fig. 01 · One run Watch, understand, act, log
One run of an AI workflow A loop of four moves. Watch an inbox, form or folder; understand what arrived and how urgent it is; a person approves before anything that matters goes out; act by filing, updating or routing; log every step; then the next item. WATCH inbox, form, folder UNDERSTAND type and urgency PERSON APPROVES ACT file, update, route LOG every step, kept NEXT ITEM
One run of an AI workflow A loop of four moves. Watch an inbox, form or folder; understand what arrived and how urgent it is; a person approves before anything that matters goes out; act by filing, updating or routing; log every step; then the next item. WATCH inbox, form, folder UNDERSTAND type and urgency PERSON APPROVES ACT file, update, route LOG every step, kept NEXT ITEM
One run of a workflow. Something arrives, the system reads it, a person approves anything that reaches a customer, moves money or carries risk, then it acts and logs every step before the next item.

01 What we mean

Watch a channel, understand what arrived, do the next step. Log everything.

AI automation is not a mystery. Something arrives on a channel worth watching: an inbox, a web form, a phone line, a shared folder. An AI step reads it and works out what it is. The system then does the next obvious thing: it files a document, updates a record or notifies a person, and writes down what happened so a human can check the trail later.

That is the whole shape of it, and it is what most people mean by workflow automation. The intelligence sits in the understanding step, where older rules-based automation used to break the moment an email was phrased differently or a form was filled in the wrong order. An AI reading step copes with the mess that real work throws at it.

What it is not: a stand-in for human judgement, or a robot clicking around your screen pretending to be a person. Work that answers customers lives one step along, on our AI agents page, and answering questions from your own documents lives on our AI knowledge base page.

02 What we automate first

The work that repeats and follows a pattern. The work that steals whole afternoons.

  • On a new email or web form

    Inbox triage

    Every message read, sorted by type and urgency, and the useful details pulled out. The right item reaches the right member of staff with the context already attached. Nothing is sent back to the sender.

    Unclear items held for a human

  • When a document or invoice arrives

    Document filing with an audit log

    Invoices, certificates and PDFs read, the key fields extracted, filed in the right place and recorded in a log that shows exactly what happened and when.

    Exceptions held for a human

  • Today: retyped three times

    Intake and data entry

    The record is captured once and written everywhere it needs to go, so nobody types the same details into three systems.

    Mismatches flagged, never overwritten

  • Overnight, on a schedule

    CRM clean-up in batches

    A scheduled run through the CRM: contacts matched and tidied, duplicates flagged, and a list of deals gone quiet sent to whoever owns them. No customer is contacted.

    No silent overwrites

  • Weekly, on a schedule

    Weekly digest reporting

    The numbers and events that matter compiled into one short digest, so the team starts Monday knowing where things stand without anyone building a spreadsheet.

    Delivered before the week starts

  • As a renewal or deadline nears

    Compliance and approval chasing

    Certificates, licences, sign-offs and recurring obligations tracked, with a reminder drafted and the right people prompted well before the date, not after it has passed.

    Notice drafted, held for sign-off

Every one of these keeps a human checkpoint. Anything that reaches a customer, moves money or carries risk waits for a person to approve it. The system does the reading and the routing. People keep the decisions that ought to stay with people. When the job is replying to customers in your inbox or on WhatsApp, that is work for an AI agent.

03 Whole processes, end to end

Sometimes the job is a whole process, not one workflow.

When you want a whole business process automated and it needs redrawing before anything is built, that is the work on our AI consultancy page; when it is already clear, we build it here.

04 How we build

First workflow live in weeks. Not quarters.

  1. 01 · Listen

    We watch the actual work

    A thirty-minute call, then time with the people who do the work, seeing where information gets copied between apps and where the same decision gets made over and over. The best first workflow is almost always already there, in plain sight.

  2. 02 · Map

    We pick one and draw its checkpoints

    One workflow, chosen because it repeats and has a visible cost. We agree the trigger, the steps and exactly where a human approves before anything leaves the building, and we put a rough minutes-per-week figure on it so the saving can be checked later.

  3. 03 · Build

    We ship something real, early

    The first workflow goes live in weeks, running on real data with the approval gate switched on from day one. You watch it work before we widen the scope.

  4. 04 · Run

    We keep it honest

    Monitoring on the runs, alerts to the person who can actually fix a problem, and a monthly review of failures, missed cases and hours handed back, set against the figure we started from.

05 What we check

Six checks before a workflow touches real work.

  1. C-01

    A person approves what matters

    Anything that reaches a customer, moves money or carries risk waits for a person. The gate is on from day one, not added later.

  2. C-02

    Every step is logged

    Each run records what arrived, what the system decided and what it did, so anyone can check the trail later.

  3. C-03

    Unsure or malformed means held, not guessed

    Fields that are empty, renamed or typed in a hurry are checked before the workflow writes anywhere. When the reading step is not confident, the item waits for a human instead of being filed or sent anyway. No silent overwrites.

  4. C-04

    Failures reach someone who can fix them

    Retries, alerts and a place for failed records to wait on every run, a clear rule for duplicates and replays, and a named person who owns the fix. A workflow that fails quietly is worse than no workflow at all.

  5. C-05

    Credentials have an owner and personal data stays minimal

    A named service account, a documented permission scope and a renewal plan for every connection. We map what personal data the workflow touches, decide what is logged, and set retention and access rules. Higher-risk uses get a DPIA before go-live.

  6. C-06

    It works with your tools and someone else can run it

    The workflow reads from and writes to your CRM, inbox and document store. If a tool cannot support a reliable connection, we say so before any budget is committed. A new supplier, an extra field or a different approver is usually a settings change, not a rebuild, and every build comes with a runbook and a change record.

06 The alternatives

AI automation, and the three alternatives.

  1. 01 · Us

    AI automation

    Built on understanding: it reads documents, classifies cases and decides between paths, with a person on every exception. When a layout changes, it adapts instead of snapping. Trade-off: it needs mapping and judgement upfront.

  2. 02

    RPA suites

    Software robots that click through screens on a fixed script. Quick to demo, but a supplier redesign or a new field quietly breaks them. Trade-off: licences, plus constant repair as the screens beneath them move.

  3. 03

    Hire more admin

    It works, and sometimes it is the right call. But the cost grows every time the workload does. Trade-off: linear cost, and good people bored by the dull half of the job.

  4. 04

    Leave it

    Fair enough for a process that runs a few times a month. The bill lands when the same manual step repeats hundreds of times. Trade-off: an invisible cost, paid in hours and in mistakes caught late.

Screen-clicking robots age badly because they copy what a person does without understanding why. We prefer automation that understands the step, so it survives the next redesign. The full argument is in RPA versus AI-powered automation for UK business.

07 Already automating?

DIY works until the business depends on it.

Plenty of teams build their first automations themselves, in n8n, Zapier or Make, and that is a sensible way to start. We take over the ones that have started to matter. We audit them, fix what is fragile, add the error handling and write the runbook, or rebuild them on our own managed servers.

  • Risk

    Error handling with an owner

    If a failed run can lose money, delay a client or create duplicate records, the workflow needs retries, alerts and a named person to fix it.

  • Logic

    Branching people can debug

    A workflow with many branches can look fine on the screen and still be impossible for staff to reason about later.

  • Scale

    A real database and workers

    Larger setups need a proper database, a queue and worker processes, so one big job does not block everything else.

  • Hosting

    Self-hosting decisions

    Backups, credentials, domains, certificates, upgrades and restore plans matter once automation is touching client data.

  • Access

    Credentials with an owner

    Production work should not depend on one staff member's personal login or a token pasted into a note. Named service accounts, scoped credentials and a renewal plan.

  • Visibility

    Silent failures

    The worst failure is the one nobody sees. We design status checks so a missing run is noticed by us, not discovered by a client.

DIY is fine for a personal notification or a one-step handoff. Bring us in when the automation carries business risk, private data, several branches or enough volume that checking it by hand becomes its own job. We compare the tools honestly: n8n for production logic and self-hosting, Zapier for quick handoffs between apps, Make for visual business logic, custom code when the workflow is the product. The trade-offs in money are in five costs that change the choice.

08 Our own proof

We run our own product on this.

Our content engine is a live workflow we operate every week: research, drafting, checks, images and publishing, each stage passing what it learned to the next and every run logged.

09 Handover

Handover means someone else can operate it.

The bar is simple. A staff member should be able to see the last run, understand the next action and know who owns the fix without asking the original builder.

We also make the commercial trade-off clear. Some workflows should stay small and manual because the volume is not there yet. Others should be automated because the same checks happen every day, the same data is copied between systems, and the same people are interrupted to confirm what a well-designed workflow could record by itself. That is where automation pays.

10 How we quote

Three shapes of work. Each quoted to your process.

  • 01

    First workflow

    Two to four weeks

    • For when the process is already clear: the workflow itself, built and live.
    • One workflow, live.
    • Real data and a visible result.
    • Human approval gate built in.
    • Fixed quote before we start.
  • 02

    Production build

    Six to ten weeks

    • Several linked workflows, or a whole process rebuilt (eight to twelve weeks).
    • Monitoring, retries and alerts.
    • Documentation and staff handover.
    • Quoted fixed after discovery.
  • 03

    Managed and measured

    Rolling monthly

    • We watch the runs, not you.
    • Monthly review of failures and hours saved.
    • Fixes and dependency changes.
    • One named point of contact.

A number on a website is priced to the website, not to your process. We quote each build as a fixed figure after a discovery call, once we know how many systems we are connecting and how much judgement the workflow carries. Model usage, hosting and third-party platform fees are itemised separately and shown before anything goes live.

11 Where this goes deep

Automation rarely works alone. Here is what usually sits next to it.

Each has its own page, built the same way: what it is, how we build it, what we check and how we quote.

12 Case

Invoice triage, in production. Four hours a week handed back.

A shared inbox at a professional services firm, with invoices, contracts and noise all arriving together. Before building anything we agreed which system held the true record of each fact, which kept the bookkeeping software out of scope. The workflow reads, classifies, files and reports, and sends only what it is unsure about to a person.

CaseProfessional services firmUK · anonymised

Finance operations

How we replaced four hours a week of manual invoice triage with an AI workflow that captures, classifies, files and reports, without disturbing how the finance team works.

Finance operations case study

  • 4h

    Hours of invoice triage back each week

    Invoice capture and routing, hands-off.

  • 96%

    Classification accuracy

    Validated against three months of manual review.

  • 0

    Lost invoices since launch

    Every email accounted for, every attachment filed.

Names redacted out of respect for our clients.

13 Questions

AI automation, answered plainly.

The same answers we would give on a call, written down.

Q-01 What is AI automation, and how is it different from normal automation?

Normal automation follows fixed rules: if this exact thing happens, do that exact thing. It breaks the moment reality is phrased differently. AI automation adds a reading step. The system understands a messy email, a scanned invoice or an oddly worded form, decides what it is, then runs the right next action and logs it. Same plumbing, far fewer broken rules.

Q-02 Is AI automation worth it for a small business?

Often yes, but not always, and we will tell you when it is not. The test is whether there is a repeated task, done the same rough way each week, that eats real hours and follows a pattern a person could describe. The quickest returns come where there is a lot of structured back-office work: professional services, property, healthcare admin, finance operations and groups running the same task across several sites. If the volume is tiny, we will say so and save you the spend.

Q-03 How much does AI automation cost, and how long does it take?

It depends on how many systems we connect, how much judgement the workflow needs and how much of the process wants redrawing first, so we do not publish a rate card. A thirty-minute discovery call gets you a fixed quote for your actual process, with model usage and hosting itemised separately. A focused first workflow is usually live in two to four weeks. A larger build with several linked workflows tends to take six to ten weeks, and a whole process rebuilt eight to twelve. Our guide to AI workflow automation costs breaks down what moves the number.

Q-04 Do you replace our existing tools?

No. We build around the systems you already run: your CRM, your inbox, your document store, your phones. The automation reads from and writes to those tools rather than asking your team to learn a new platform. If a tool genuinely cannot support a reliable connection, we say so early, before any budget is committed.

Q-05 Who owns the system after delivery?

You do. The workflows, the accounts, the credentials and the documentation are yours. We can host and monitor the system for you on managed servers in the EU, but that is a service you choose, not a lock-in. If you ever want to bring it in-house or move it elsewhere, the handover pack is written so another competent team can pick it up.

Q-06 Is it GDPR-compliant for UK data?

It can be, and that starts at design, not as an afterthought. We map what personal data the workflow touches, keep it to the minimum, decide what is logged and set retention and access rules. Anything we host for you runs on managed servers in the EU, under UK GDPR. For higher-risk uses we recommend a DPIA before go-live, and the ICO guidance on AI and data protection is the standard we design against.

Q-07 Does it replace staff?

Mostly it replaces tasks, not people. The work we automate is the repetitive reading, sorting, filing and chasing that few people enjoy and that keeps them from the judgement work only they can do. In small teams that usually means the same people getting their week back. We have not been asked to help a client cut headcount, and it is not what we sell.

Q-08 What is the difference between RPA and AI-powered automation?

Traditional RPA clicks through screens the way a person would, following a fixed script. Change a button, a form or a supplier layout and the script breaks. AI automation works from understanding instead of screen positions: it can read a document, classify an email, or decide which of three paths a case should take, then pass anything unusual to a person. We wrote a longer comparison if you want the detail.

Q-09 Can you take over automations we built ourselves?

Yes. We audit what is there, fix the fragile parts, add error handling, alerts and credential ownership, and write the runbook. Then you choose: we leave it running where it is with a named owner on your side, or we move it onto our managed servers and run it for you. Either way you keep the accounts and the documentation.

Q-10 Can you move our workflows from other automation tools?

Yes. We rebuild the logic rather than copying it blindly, because a move is a good moment to remove duplicated steps, add logging and make the workflow easier to operate. Before we start, we tell you which parts are worth keeping as they are and which are better redrawn.

Q-11 How do you stop silent failures?

We design explicit error routes, retries, alerts, a place for failed records to wait, and regular status checks. If a workflow fails at 2am, someone should know what failed and which records were affected. A missing run should be noticed by us, not discovered by your client.

14 Enquiry

Tell us the task you would hand over first.

Thirty minutes on a call and we will tell you honestly whether automation pays for itself in your case.

We use your details to reply to this enquiry. Nothing else. See our privacy notice.

Let's see if we're the right fit.

Pick a time that suits you. Thirty minutes, no preparation needed.

HAVING TROUBLE? Open the calendar in a new tab