AI consultancy for small and mid-sized business
We work with small and mid-sized businesses to find the work worth automating, apply AI at the points it genuinely helps, and leave you with something your own team runs and understands. No jargon, no year-long programme, no department you have to hire first.
Most conversations start with one job that eats a day a week.
Form submitted
sarah.keane@…
Nora read it and drafted a refund
€42.50 · damaged planters
Held
Held for you — Nora drafted a €42.50 refund and needs your OK.
Send the reply to Sarah Keane
pending your OK
This is the part most vendors skip. The same technology, pointed at the same business, produces wildly different outcomes — and the gap almost never comes down to which model you picked.
Used well — 10×
Used badly — a tenth
The difference is placement — which decisions you hand over, which you keep, and what happens at the boundary. That is the whole of our job.
How we work
Four steps, usually a few weeks rather than a few quarters. You can stop after any of them and still be better off than when you started.
01
We sit with the people who actually do it and watch a real case travel through the business — every handoff, every spreadsheet, every “just check with Marie first”.
02
And, just as usefully, where it doesn't. Plenty of what slows a business down is a missing integration or an unclear rule, and no amount of AI fixes either.
03
Something small, in production, doing real work within weeks — with a person approving anything that touches money or a customer while trust is being earned.
04
Your team can read it, change it and explain it. We're a phone call, not a dependency. If we've done this right you need us less each quarter.
The thinking
Five things we've found to be true across every engagement. None of them are about technology, which is rather the point.
One
AI pays for itself on volume, not novelty. The impressive demo is the one-off; the return is in the thing that happens forty times a week and nobody enjoys. If a task happens twice a year, automating it is a hobby.
Two
Almost every business process is about 90% routine and 10% judgement, tangled together. Pull them apart. The routine should be deterministic — same input, same result, every time, no cleverness. The judgement is where AI belongs, and it's a much smaller surface than people expect.
Three
Money leaving the business, a message going to a customer, anything with a legal edge. Not forever, and not on everything — but until the pattern has proven itself. A human gate is what makes it safe to move quickly everywhere else.
Four
If you can't find out why something was decided, you can't improve it, defend it to a customer, or spot it drifting. “The system said so” was never a good answer when it was a spreadsheet, and it isn't one now.
Five
Hours given back, error rate, how long a customer waits. Pick the number before you build, and be willing to conclude it didn't work. Most AI disappointment is really the absence of a baseline.
Where this is going
In a few years, having AI in the business will be about as distinguishing as having email. The gap will open up somewhere less obvious: between the companies whose work is written down clearly enough to hand over, and the companies where it still lives in six people's heads.
That's a much older problem than AI, and it's the one worth solving first. Every business we've worked with has found that the act of describing a process properly — what triggers it, what the rules are, who decides what — was worth doing on its own merits, before a single model was involved.
We think software itself keeps drifting in the same direction: away from buying a rigid product and bending your business around it, toward describing how your business actually works and having the software follow. That suits smaller companies far better than large ones. You have less legacy to unpick and a much shorter path between noticing something and changing it.
What gets scarce in that world isn't labour. It's judgement — knowing which call matters, which risk is worth taking, what a fair outcome looks like for a particular customer on a particular day. That stays yours. Everything around it is increasingly negotiable.
Where the work ends up living
The work we build together runs on our own platform, crate. You don't need to care about that on day one — but it's why what we hand over stays readable, traceable, and yours to change.
Inbox
order query arrives
Orders
looked up order 4821
Accounts
credit note drafted
Customer
reply waiting on you
What's running keeps running while you edit. Changes go live when you publish them, and anything already waiting on you stays where it is.
Every job keeps its own record — what came in, what was decided, who said yes and when. So “why did it do that?” has an answer months later.
If something goes down mid-job it picks up where it stopped. If something goes wrong you can read it back step by step rather than guess.
One job is enough to start. We'll come back with what we'd look at first, what we'd leave alone, and whether we think there's anything here worth doing at all — including when the answer is no.