AI Automation
The work nobody should be doing by hand.
Orders retyped from WhatsApp into a spreadsheet. Invoices read one at a time. The same report rebuilt every Monday. When your staff are the wire between two systems, that's the part we replace.
Scoped per task · Usually live in 2–6 weeks · Runs on your own accounts
Which one is for you
Start here. A lot of this doesn't need us.
Plenty of automation is an off-the-shelf tool and an afternoon. We'd rather tell you that than sell you a project.
A ready-made tool is enough if
- —The two apps you want to connect already talk to each other
- —The rule is simple — when this happens, do that
- —Nothing needs judging, only moving
- —You're happy for the logic to live in someone else's account
Custom automation earns its cost if
- —The work needs a decision, not just a transfer
- —Your process has exceptions a template can't hold
- —It touches a system with no proper integration
- —It has to keep working when a supplier changes their format
- —The data can't leave your own accounts
Four places this pays for itself.
Most jobs are one of these, or a mix. None is more advanced than the others — they're the same waste in different shapes.
The copy-paste job
Something arrives in one place and gets typed into another. WhatsApp orders into a sheet. Invoices into accounting. Form entries into a customer list. Nobody is deciding anything — a person is just carrying data between two systems that never learned to talk.
You'll know it's this if someone's day starts with “let me update the spreadsheet.”
The document pile
Quotes, receipts, delivery orders, IC copies. They arrive as photos and PDFs, and someone opens each one to pull out the same four fields. This is the work these tools are genuinely good at, and it is usually the fastest thing to prove on your own backlog.
You'll know it's this if there's a folder nobody volunteers to go through.
The same answer, fifty times a day
Opening hours. What is in stock. Where you deliver. How much it costs. Your staff answer the same handful of questions until the enquiries that actually matter get buried under them. Handling the repeats — and knowing when to pass a real one to a human — gives those hours back.
You'll know it's this if your replies get slower exactly when business gets better.
The report that eats Monday
Numbers pulled out of three places, pasted into a template, formatted, sent. Every week, by hand, usually by the person who can least afford the morning.
You'll know it's this if the report is always late and never wrong.
How long? A single document or copy-paste task is often two to three weeks. Something touching several systems, or needing a review step, is longer. You will have a timeline with your quote, not after it.
Sometimes the answer is no.
Not everything is worth automating. A task done twice a month is cheaper by hand. A process that changes every quarter will break faster than we can maintain it. And anything where a wrong answer costs you a customer needs a person checking the output — which sometimes means the automation saves less than it costs to run.
We will tell you when that's the case. It's a shorter conversation than the alternative, and it means you can trust the next thing we say.
What moves the number.
Every quote comes from the same few things. Knowing them makes the first conversation much faster.
- How messy the input is — A clean export is an afternoon. Photographs of handwritten forms is not.
- Whether it has to decide — Moving data is cheap. Judging it — approve, reject, flag for a human — is where the real work is.
- What it has to connect to — A system with a proper integration is straightforward. One without is a project of its own.
- What happens when it's wrong — Low stakes means it can simply retry. High stakes means review steps, logs, and a way to undo.
- Who watches it afterwards — Automation is not furniture. Something upstream will change eventually, and someone has to notice.
Describe one real task end to end and you'll get a figure — not a range that quietly doubles later.
Process
How it actually goes.
Watch the work
We sit with whoever does the task today and follow it end to end — including the exceptions they've stopped mentioning because they're used to them. Most of the value gets found here.
From you: An hour with the person who actually does it.
Prove the hard part
We build the riskiest piece first — usually the reading or the deciding — and run it against your real backlog, so you can see how accurate it is before committing to the rest.
From you: Real examples, including the awkward ones.
Build it
The full flow, the error handling, and the point where a human steps in. Depends heavily on how many systems it has to touch.
From you: Access to the accounts it will run on.
Run it alongside you
It runs next to the manual process rather than instead of it, until the two agree. Then we switch over.
From you: Someone glancing at the output daily.
Handover
Documentation, the logs, and a way to see what it did and why. We stay reachable when something upstream changes.
From you: An hour or two for training.
Automation is judged on its exceptions, not its happy path. Step 2 exists so you find out early whether this is worth continuing — it is the stage we would least like you to skip.
Always included.
None of this is an add-on.
FAQ
Common questions.
Usually it replaces the part of their job they like least. In most of what we build, the same people handle more volume — not fewer people handling the same. If reducing headcount is genuinely the goal, say so early and we'll be straight with you about whether the numbers actually support it.
Onto your own accounts wherever the task allows. If something genuinely needs an outside service, we'll tell you exactly what leaves and what it's used for, and you decide before we build anything.
That depends entirely on the task, which is why we prove the hard part against your real backlog before you commit to the rest. You get a number measured on your own documents, not a claim measured on someone else’s.
You get an alert, the work falls back to the manual process, and nothing is silently lost. That is designed in from the start — an automation that fails quietly is worse than no automation at all.
No. This is separate work, and plenty of it sits behind businesses that barely have a website. If you need both, we can sequence them.
Because the honest answer to “what does automation cost” is “what are you automating.” Two tasks that sound identical can differ tenfold depending on how clean the input is and whether anything has to be judged. Describe one real job end to end and you will have a firm figure, usually within a day or two.
Tell us what your team keeps doing by hand.
Describe one task from start to finish, including the exceptions. We'll tell you whether it's worth automating — including when it isn't.