Anyone can build an AI demo now
When building is cheap, the expensive mistake is building something that never pays for itself.
ReadAnyone can build an AI demo now. Making it pay is harder than it looks.
That’s the part I do. I locate the places in a business where AI will actually move revenue or cost, and I build what it takes to get there.
Find out if it will payPractitioner and PhD researcher, UTS Business School
What I do
The brief isn’t the problem
Building has become the easy part. A working prototype takes less than an afternoon. What’s expensive now is employing AI in the wrong place100×, paying to keep it running10⁵×, and supporting all the work that makes you money before knowing if any of this will actually pay.
I trace how the work actually moves through the business: where a request sits waiting for someone to approve it, and where the same details get typed into a second system by hand. That’s usually where the value is, and it’s rarely where the brief points.
My background is at the intersection of humans and technology, and my PhD at UTS10⁷× is on exactly this problem. The human side keeps me looking at real work instead of the org chart. The technology side informs what AI can do right now, what it might cost to run and most importantly, where the value is for you.
One engagement
An Australian allied health provider
01 · The brief
The client wanted AI to support its clinicians. The thinking behind this brief was ensuring that clinicians with access to the right tools would enable leaps in productivity.
02 · What we found
It turns out the costly problem sat mostly unobserved in the back office. Invoices regularly landed in a growing pile of aged debt when small details (incorrect billing codes and minor handling errors) invalidated them. These circled the process and were often written off. This was where we located and unlocked a massive opportunity.
03 · Where AI went
The problem wasn’t at one point. It sat at every step that relied on someone doing something perfectly: patient onboarding, scheduling, payment code validation and governance, and invoice management. With that many places to break, their existing systems and processes couldn’t plug the gaps. So I redesigned the end-to-end chain, locating where AI could address existing shortfalls.
04 · The result
$1.5M
a year in aged debt, eliminated.
The lesson
The value was a few steps from where everyone was looking. It usually is.
After the value lands
Getting the value is half the job
Even AI that’s meant to be invisible changes people’s days, and most implementations never plan for this. I help your team manage what comes next.
People use it.
The team it supports helps shape it, so it fits their actual work.
Trust is set at the right level.
We agree what “good enough” means against today’s error rate, not against perfection.
Freed-up time has somewhere to go.
That’s decided before go-live, with the people whose roles change. At the provider, the team that chased invoices is moving to customer service.
The follow-up happens.
It’s booked from day one, with a dollar figure on every week the value sits unused.
Proof
Elsewhere
A fleet operator3×
On track for 3x fleet growth, with AI carrying five areas of operational work.
An investment manager3×
3x portfolio growth, with AI handling reporting and investor documents.
A national allied health provider7/12
Seven of the twelve changes I recommended are now running.
How I work
I call it Value Location
Find it.
Get past the brief to the real problem, and pinpoint where value moves.
At the provider, that meant looking past the clinic to the back office.
Prove it.
Check the value survives real running costs, agree what “good enough” means, and rank the one or two that matter.
The aged debt went ahead of everything in the original brief.
Employ it.
Only what that value needs. An AI agent, an AI-enabled process or an AI-supported tool.
AI went in at each point in the chain where a small detail could break.
Land it.
Make sure it’s used, and that freed capacity goes somewhere.
The invoice team is moving to customer service.
Value Location takes two weeks. What comes after is drawn from what it finds.
Most of what I learn doing this ends up in writing.Writing
Notes from the work
The brief and the problem are usually different things
A brief points to where the pain is felt, which isn’t always where it starts.
ReadThe research
Where AI value leaks, and how to keep more of it
My PhD at UTS Business School asks the question every business with an AI budget ends up asking. Once AI goes in, where does the value go, and how do you keep as much of it as possible?
Part of the answer is that AI doesn’t behave the way people expect. It’s probabilistic. It’s right most of the time rather than every time, and it changes as it’s used. Companies still manage it like any other technology, with a fixed end state, a rollout and a plan to measure against. That plan assumes a predictable system, so value leaks out in the gap between what was planned and how the AI actually behaves.
The research and the client work feed each other. Engagements show me where the value goes missing, and the research sharpens how I find it and hold on to it.
At this scale
Get the occasional essay
Longer pieces from the research, when there’s something worth saying.
The question
Will it actually pay?
Every level above comes back to this one question.
There are two ways to get it answered, depending on the size of what you’re weighing up.