AI automation and integration
Automate the repetitive work. Then let AI see the business.
Two halves of the same job. Handing the work nobody should be doing by hand to software, and connecting AI to your own systems, so asking a question of the business gets you an answer from your data.
- Price
- Fixed price against a written scope. A single automation is one of the smallest jobs we do.
- Timeframe
- Days, not months. A set across a department gets scoped first
- You get
- Repetitive work handed to software, and AI connected to your own data
- Best for
- A repetitive job eating a morning a week
Every business has work that has to happen and that nobody wants to do. Somebody opens a PDF and types what's in it into a spreadsheet. Somebody writes the same email with three details changed. Somebody checks one system against another and finds they disagree. That's the automation half, and it's what most people arrive asking about.
The second half is newer. AI is now good at answering questions, but only about things it can see, and by default it can't see your business at all. Connect it properly to your own systems and someone can ask which jobs are overdue, which certificates lapse next month, or what a client has spent this year, and get a real answer instead of a confident guess.
What this usually looks like now.
If two or three of these land, you're in the right place.
Somebody spends a morning a week copying figures out of PDFs into a spreadsheet.
Every supplier invoice gets read by a person and typed into the accounts by the same person.
The same enquiry gets the same reply written from scratch, forty times a month.
You've tried an AI tool and it was useless, because it knows nothing about your business.
The data is all there and getting a straight answer out of it still means asking someone.
Somebody's whole Friday is chasing people for things they said they'd send.
What we build.
Four kinds of work, plus where it lives once it's built. Most projects need one or two of the four rather than all of them.
Streamlined data processing
The big one, and where the hours are. Invoices, delivery notes, certificates, timesheets and forms read and put into the system that needs them, matched against what's already there, with anything uncertain flagged rather than guessed. It's the least glamorous item on this list and it's usually the one that pays for the project on its own.
Logic built into the system
Decisions that follow rules get made by the software: what gets routed where, what's flagged, what's chased, what's approved automatically and what isn't. The rules are yours and they're written down. Where a decision needs judgement rather than a rule, it goes to a person: that boundary is something we agree with you rather than something we decide.
Agents that do the repetitive jobs
The work that has to happen and that nobody wants to do. Drafting the reply that gets written forty times a month, chasing the thing somebody said they'd send, summarising a long document and filing it where it belongs. Working inside your systems rather than in a separate tool somebody has to remember to open.
Connected to your own data
The half most people can't buy properly, and what makes the other three worth having. A secure connection between an AI assistant — Claude, or whichever model suits the job — and your actual records, so someone can ask which jobs are overdue or which certificates lapse next month and get an answer from your data rather than a confident invention. Read-only where it should be, able to act where you've decided it should.
Some of this goes straight into software you already run. The rest sits on a workflow platform, usually n8n, which we host, monitor and maintain, so nobody in your business quietly becomes the person responsible for keeping it running. That's usually the part that gets skipped, and it's why automations built in-house tend to stop working about eight months later.
Where it goes.
One job through a contractor's business, from the enquiry landing to the invoice being paid. Four steps handed to software, two that stay with a person, and the two are the ones worth keeping.
Nothing here replaces a process. It takes named steps out of one that already exists, which is why we map it first, and why the parts involving a customer or a judgement are the parts we'd argue against automating at all.
How it's scoped and what it costs.
We start by watching one process end to end and working out what it costs you in hours. If the answer is small, we'll say so.
A single automation is one of the smallest pieces of work we do: days rather than months, and quoted as a job against a written scope. What takes longer is a set of them across a department, or anything that has to reach into several systems at once, and those get scoped properly before we quote.
Fixed price against a written scope. Where there's a per-use cost for the underlying tooling you're told what it is, and it's billed at what it costs us rather than marked up.
Not sure what's right for you?
Take the 5-minute assessment. 12 quick questions, and we'll tell you honestly whether you need a partnership, a one-off build, or nothing from us at all.
Sound like the right size of job? Half an hour, no charge, and you'll know.
Book a 30-minute callWe build the process before we bring in the AI. Automating a mess gives you a faster mess, and most of what makes these projects worth doing happens before anything is automated at all, in working out what should happen, who decides what, and where the judgement sits. AI is very good at running a process. It is no good whatsoever at working out what the process should be.
Anyone can wire up an automation. Here's the difference.
- The process comes before the tool.
- We map how the work actually moves before automating any of it. That's where the savings turn out to be, and it's why we sometimes recommend a smaller automation than you asked for, or none, because the real problem was two steps upstream.
- A person stays where judgement lives.
- Anything touching money, safety or a legal record keeps a human approval step. We'll tell you which parts of your process should never be fully automated and why, and that boundary is agreed with you rather than decided by us.
- We run our own business on this.
- Our invoicing, clients and projects are connected exactly this way, built for ourselves before we sold it to anyone. It's how we know where the awkward parts are, and there are considerably more people talking about this than shipping it.
- We'll tell you when it isn't worth doing.
- Automating a job that takes twenty minutes a month is a way to spend money, not save it. We work out what the process costs you in hours first, and if the number is small we say so before you've spent anything.
What people should be doing instead.
This is the part of an automation conversation that usually goes unsaid. Nobody is at their best punching the same figures into a system forty times a week, and no business gets better at anything because a person spent Friday chasing paperwork. That work has to happen. It just doesn't have to be done by a person.
- Talking to customers
- The conversation that wins the job, saves the relationship, or catches the problem before it becomes a complaint. Nothing has ever automated it and nothing is close.
- Looking after each other
- The pastoral half of running a business: noticing someone is struggling, training the new starter properly, having the difficult conversation early. It's the first thing squeezed out when everyone is busy with admin.
- Winning work
- Business development is what most owners say they'd do more of if they had the time. The time is usually sitting inside the jobs on this page.
- The decisions that need a person
- Judgement, exceptions, and the calls where being right matters more than being fast. That's what people are for, and it's what gets crowded out by everything else.
We don't sell headcount savings and we won't put them in a proposal: that's a rule here rather than a preference. What we sell is the same team with their week back. Being clear about that early is also the practical answer: a system people are frightened of is a system nobody uses properly, and the project fails on adoption long before it fails on engineering.
Where this can go next.
One caveat: automation works on a process that's been mapped and records that have a shape. If your data lives in six places and disagrees with itself, automating on top of it multiplies the mess rather than removing it. So the sequence we'd usually recommend is a short business analysis first, so the process is understood before anything is automated on top of it, and where the records themselves are the problem, a system of your own to hold them properly. Neither is a reason to wait: plenty of clients start here, and we'll tell you honestly if the order is wrong.
Both are bigger jobs than this one, and both are worth knowing about before you spend anything. What a business analysis involves · Custom operating systems
The questions this usually raises.
Sound like what you need?
Half an hour, no charge. Tell us what the job is and we'll tell you what we'd do, what it'd cost, and whether it's worth doing at all.