Approach
I don't start by building. I start by working out what should never be built.
Most AI engagements begin with a tool and work backwards to a justification. Mine begins with your work, sorted by what it costs you to be wrong.
The triage
Three buckets, sorted by consequence
Annoying is not the same as expensive. I sort on what it costs you when the answer is confidently wrong.
01
Contained, low consequence
Your team owns it.
Work where a mistake is cheap and stays where it happened. Drafting, summarising, first-pass research, the reformatting that quietly eats an afternoon.
You don't need me for this. Paying a consultant to hold it forever is how AI budgets disappear. You need your own people fluent enough to do it themselves, and to spot when a tool is confidently wrong.
So this bucket is where I start: I train your team on it first, and then I get out of the way.
02
High consequence
I own it.
Work where being wrong is expensive, hard to reverse, or lands in front of a customer or a regulator.
Non-deterministic tools don't belong here without proof. I build it, I test it against cases where being wrong actually costs you, and I show you the evidence before anyone depends on the output.
I maintain it as your processes and the models change. You can take ownership whenever you decide you want it.
03
Not worth automating
Left alone, on purpose.
Work where the automation would cost more than the problem, break something that currently works, or replace judgement that is the actual value of the job.
This bucket is why the other two can be trusted. Every hour spent automating something in here produces nothing, and it's where most wasted AI spend goes.
I'm paid the same either way, which is the point.
In practice
How an engagement runs
First
The triage
I sit with the people doing the work, not just the people describing it. You get the three buckets in writing, including the one I'm not going to charge you to build.
Then
Your team takes bucket one
A full day of training, pitched so it lands with executives and analysts alike. They leave with transferable principles and real output from your own tools.
Then
I take bucket two
I build it, test it against the cases where being wrong costs you, and show you the evidence before anyone relies on it. Then I maintain it.
Whenever you want
You take it back
Everything I build is yours to own. There's no lock-in and no proprietary layer you can't see inside.
Why hire this out at all
Gen AI, not Gen Z
When a business feels behind on AI, the instinct is to hire for it, usually someone younger who supposedly just gets it.
Being comfortable with technology is not the same as being fluent with AI. I have spent years deliberately immersed in this one thing: building with it, teaching it, and watching what happens when 2,000 people are handed it at once.
You don't need to hire the gap away, you need to bring the people you already have.
Photo of Oliver
Drop a real photo at /public/oliver.jpg and set founder.photo.enabled to true.
The part everyone skips
AI isn't a technology problem, it's a human one.
Frightened people don't experiment, and they won't tell you which parts of their job quietly eat their week, so you automate the visible trivia and land back where you started.
That's why the training comes first, and why it has to reach every tier. Adoption that only lands with executives produces mandates nobody follows. Adoption that only lands with engineers produces tools nobody sanctions.
It has a measured price, too. People who send AI-generated work their colleagues have to unpick get rated as less capable and less trustworthy.
Who you would be working with
Oliver, and only Oliver
I lead AI adoption at a major Australian retail and business bank and a world leader in infrastructure asset management, across more than 2,000 people.
A governance model for non-deterministic tools that survives financial regulation. If it holds there, your environment is far more forgiving.
What that means for you
Everything on this site is one person's work. That caps how many clients I can take. It's the trade: you get the person who did the thing, not the person who sold it.
Still not sure which bucket your work falls into?
There's no deck and no obligation. You leave with my read on where AI is worth your money, including "you don't need this yet" if that's the truth.