Not advice about AI, and not another subscription. The systems themselves, running inside your business, in an account with your name on it.
We work out where AI genuinely helps in your business, we build the systems that do that work, and we set them up so the business owns them rather than whoever built them.
Most of this industry stops after the first one. You get a strategy, a workshop, a list of sixty ranked opportunities, and then it’s your job to make something happen. A plan needs somebody inside with the time and the mandate to carry it, which is quite a lot to ask of a business that’s just been handed a document.
So we build the thing, and we leave it running.
A chat window gives you a draft that you then paste somewhere. The real work sits one level down, where AI can reach your files, your systems and your routines, and actually finish things off.
Nearly every business we meet is stopped at the first one.
That’s the whole idea, and it’s the difference between asking something for advice and asking it to go and do the job. A chat assistant writes you a draft proposal. An agent reads the brief, pulls the numbers out of your systems, writes the proposal in your template, and puts it in front of a person to approve.
Nothing irreversible happens without a human saying yes. That part’s designed in from the start rather than bolted on when somebody gets nervous.
The most useful AI tool available to a small business right now doesn’t have a friendly interface. It runs in a black window with a blinking cursor, which is enough to put most people off before they start.
Getting going means installs, accounts and API keys, and every guide written for it assumes you already know what those words mean. That’s the real barrier, and it’s got nothing to do with how capable you are.
No coding experience required. It’s called Claude Code, and you won’t write any. We do that bit with you, on a call, in about an hour.
Pick something your team does every week that everybody quietly resents. We build that one thing, it runs, and you get to watch it work.
You won’t know what you actually want until you’ve seen one working, which is why we don’t start with a strategy. Three half-finished systems teach you nothing at all. One finished one changes what you ask for next.
Setting AI up for one person takes an afternoon. Setting it up for a business means an organisation account, users invited properly, a shared workspace and managed version control, so the work becomes a company asset with a history behind it.
Done that way nothing lives on one laptop, and nothing walks out the door when somebody resigns. Skip it and you’ve quietly handed your operations to whoever set them up, and you won’t find that out until they hand in their notice.
So we do it in a particular order. You create the organisation account, with us on the call while you do it. Not as a test, but because an account we set up and hand over later isn’t really an account you own. Your account, your billing, your keys, from the very first minute. Then at the end we delete our own access while you watch. Nothing breaks, because nothing was ever ours.
Before any of that there’s a conversation. An hour, virtual or in person, and the point of it is to work out together whether there’s anything here worth doing at all.
Repositories, shared workspace, version control, permissions. The company owns the account, so the work is an asset with a history.
Built for the jobs you’ve actually named rather than a generic toolkit. The list grows as people see what’s possible.
How the work gets done, written down as we build it. The documentation’s a by-product rather than a project you run afterwards.
The walk away book. If it’s any good, it’s the thing that makes us unnecessary.
Conventions, naming, structure. Completely invisible until a new person arrives, and then it’s everything.
Hands on, on your real work, person by person. Adoption is the deliverable, not the software.
Written guidance at every level, so a new starter picks it up and adds to it instead of you paying somebody to explain it all over again.
What the business knows, gathered once, somewhere the AI can read it. Every skill built afterwards is better because it exists.
Connected properly, so the systems have somewhere real to read from and write to.
AI wired into the tools you’re already paying for, rather than another window to copy and paste between. This is where most people give up.
Built for the business and for each person in it, so what comes out sounds like you and not like everybody else’s AI.
Every skill versioned, every change dated and attributed. It keeps itself, because that’s how the system was built.
Everybody gets set up individually, on their own real work, trained with their hands on the keyboard rather than watching somebody demonstrate. We check back in at thirty days.
We also build it with your people rather than for them. It’s slower, and it’s the entire reason the thing still works six months later. A system built for you is one you ring somebody about. A system you helped build is one you can change yourself.
Because the work is version controlled from day one, the audit trail falls out of how the business operates rather than being a document somebody writes the week before an assessment. Every change to every skill is dated and attributed, approvals sit on anything that matters, and your data stays yours.
If a board, an insurer or a tender has started asking what AI you’re running and who’s accountable for it, the answer’s already sitting there, because the register keeps itself.
We work fluently across the frameworks that matter: the Essential Eight, ISO/IEC 42001, the AICD AI governance framework, Australia’s AI Ethics Principles, the EU AI Act and NIST. Which one carries the weight in your case comes down to your sector and to who’s doing the asking.
The AI Business Lab is an Australian firm founded by Sarah-Jane Kavanagh in 2023, working with clients across Australia, New Zealand, Europe and the United States, in government and in private industry. Everything we recommend is grounded in systems we’ve designed, shipped and run inside heavily regulated sectors, where an answer that merely sounds right isn’t good enough and somebody’s name has to sit against every decision.
An AI answer platform for National Construction Code compliance. Ask a code question, get a grounded answer, built for the practitioners who carry the liability.datumdesk.ai
A sovereign retrieval platform that produces regulatory compliant training courses in 24 hours, powering a national CPD academy.cpdondemand.com.au
Turns business procedures into professional multilingual video training, delivered in under a week and tracked for compliance.talkinghead.solutions
If you run an organisation of fifty to five hundred people, the first question usually isn’t which routine to build. It’s where AI is already touching the business, what it’s worth, what it exposes you to, and what order to move in.
That’s the audit. Twelve areas scored across four dimensions, a roadmap ranked by value and risk, and a plan you can run with us or entirely without us. For boards and executives who are personally on the hook there’s also the governance work: the board briefing, the AI council setup, and implementation through to certification readiness.
An hour, virtual or in person, no sales script anywhere near it. You bring the problem, we work out together where AI genuinely helps and where it doesn’t, and you’ll leave knowing your next move whether or not that involves us. What would you want to talk about first?