The board approved the AI budget eighteen months ago. The CFO is asking what it returned. Nobody has an answer they would put in writing.
THE VALUE PROBLEM
> INDEPENDENT AI ADVISOR · SYDNEY
Your people are faster. Your business is not.
Most organisations can show their staff use AI. Few can show it in the accounts. The time people save gets lost in handoffs, approvals and decisions nobody redesigned.
I work with CIOs and executive teams in banks, insurers, super funds, critical infrastructure operators and other regulated businesses. I help them redesign that work and decide who answers for what AI does.
Send me one AI initiative the board is questioning. I will tell you in writing where the value is going and whether it can be fixed.
> tony@tonycollins.co
/ THE PATTERN
This may be your organisation.
Three functions bought three copilots. None connects to the others. Risk discovered two only after they were in use. Nobody can say which decisions they influence, what information they use or who remains accountable for the result.
THE GOVERNANCE PROBLEM
The pilot worked brilliantly for twelve people. Extending it to twelve hundred produced activity. The P&L did not move. Individuals are measurably faster. The wider process is not. The gain disappears somewhere between the task and the P&L.
THE DESIGN PROBLEM
All three have one cause. The work around the technology was never redesigned.
/ HOW I HELP
You cannot delegate a decision you never defined.
Start with one live initiative. Find the structural constraint. Redesign the work around it. Then decide whether the wider programme deserves to continue, change or stop.
The Structural Read
SHORT DIAGNOSTIC · WRITTEN READBACK · FIXED FEE
The first conversation has no sales agenda. Bring one AI initiative that is under scrutiny, stalled at scale or producing activity without a defensible return. I test five things.
1.
What business outcome is the initiative meant to change?
2.
Has the workflow been redesigned or merely accelerated in places?
3.
Can the system access reliable data and the context required to use it?
4.
What may AI support, decide or execute, and who remains accountable?
5.
How will the system and the organisation learn from use?
The result is a written readback to the accountable leaders. It identifies where value is leaking, what is structurally in the way and whether there is something worth fixing.
If the problem is technology, data, adoption or the use case itself, I will say so. If the organisation does not need me, I will say that too.
BEST SUITED TO
CIOs and transformation leaders preparing to commit significant investment, or already carrying an AI portfolio whose enterprise return is becoming difficult to defend.
The AI Operating Design Sprint
FIXED SCOPE · FIXED FEE · TYPICALLY FOUR WEEKS FOR ONE PRIORITY WORKFLOW
The sprint follows a live workflow from individual task to enterprise outcome. It finds where the gains disappear, usually into handoffs, approvals and decisions nobody owns.
Then it redesigns the work. Where should people step in? Which decisions can the system make on its own, and when must it stop?
THE WORK PRODUCES
A value-leak map. Where task-level gains are being lost and why.
An operating design. The future workflow, human and AI roles, handoffs, learning loops, decision rights and controls required to carry the value at scale.
An implementation sequence. What to change next, in what order, with data and delivery dependencies stated rather than buried.
Named people inside the organisation learn to maintain the operating design, decision inventory and delegation rules. When the sprint ends, your people run it.
Where the scope crosses several functions or decision domains, the engagement may run longer. The scope, fee and stopping point are agreed before work begins.
BEST SUITED TO
Organisations with AI initiatives in production but no operating model redesigned around how the technology changes work, authority and accountability.
The Independent Advisory Seat
MONTHLY · ONGOING · YOUR AGENDA
An independent voice for the AI commitments that carry material risk. An honest read before the decision is made.
The question is simple: is the organisation designed to carry what it is about to commit to?
Most advisory relationships begin after a sprint and run alongside a programme already in motion. I test proposals, challenge assumptions and help leaders distinguish a technology constraint from a data, workflow, governance or operating-model constraint.
BEST SUITED TO
Business and technology leaders that want senior AI oversight independent of vendors, platforms and systems integrators.
/ WHAT CHANGES
You know which investments deserve to continue
Separate initiatives producing enterprise value from activity that merely looks like progress. Put more behind what works. Change or stop what does not.
The workflow carries the gain
Redesign the handoffs, roles, information flows and incentives around the technology so that task-level productivity becomes an enterprise result.
Authority becomes explicit before autonomy expands
Decide what AI may do on its own, what it must refer up, and who answers for the result.
Your organisation can continue without depending on me
Internal owners maintain the operating design, improve it through evidence and extend the method into new workflows as systems and risks change.
/ WHY INDEPENDENT
In most AI programmes the builder checks its own work at three points. I don't sell technology or delivery, so I can check it for you.
When the case is sold.
The builder prices the value of its own work. I set the baseline and test the claim.
When it scales.
The builder proposes to run what it built, and the client is locked in. I design how it runs: who owns each agent, what it may decide, how change is controlled and how value is reported.
When results are reported.
The builder marks its own delivery. I measure against the baseline set at the start.
I am paid for the right answer. The size of the programme does not change my fee.
/ THINKING
Field notes on making AI work at enterprise scale.
2026-07
The dissipation problem: why faster people don't make faster companies
Task-level gains dissolve in handoffs, review loops and approval chains. Where to look for the leak, and what to measure instead of adoption.
>
2026-06
Decision rights are the missing layer of your AI operating model
Before autonomy expands, authority must become explicit: what AI may support, decide or escalate, and who owns the result.
>
SUBSCRIBE FOR NEW NOTES
> follow on linkedin/ BACKGROUND
I have spent twenty years designing how large organisations run technology, across Asia Pacific. Thirteen of those were at Cisco, where I led operating model and value consulting work for clients including Telstra, CBA, BHP, Woolworths, Etihad, Rio Tinto and the Department of Defence.
MBA (TECHNOLOGY), AGSM UNSW · MSC COACHING PSYCHOLOGY, UNIVERSITY OF SYDNEY
/ FAQ
Should AI be run centrally or by business units?
Centralise decisions that are high-risk, regulated or cut across business units, such as credit, pricing, claims and hardship. Let business units run use cases that sit inside them. Leave low-risk productivity tools largely alone. The test is what a wrong decision would cost, and who would answer for it.
Who is accountable when an AI system makes a decision?
Most organisations cannot answer this. Accountability must be designed: state what AI may support, decide or escalate, and who owns the result. That design is the missing layer of most AI operating models.
What is a Structural Read?
A short fixed-fee diagnostic. You bring one AI initiative under scrutiny and I test what it was meant to change, how the work was redesigned and who holds the decision rights. The result is a written readback to the accountable leaders.
What is an AI operating model?
The design of how work flows between people and AI systems. It covers workflows, decision rights, learning loops and accountability. Without it, AI improves tasks while the enterprise result stands still.
Can a director rely on an AI agent?
Section 189 of the Corporations Act covers reliance on employees, advisers, other officers and board committees. An AI agent is none of those, so accountability stays with a named person. Every agent needs an owner who set its limits and answers when it gets something wrong. This is a governance view. Take legal advice on your own position.
How do you work with clients?
Three formats. The Structural Read is a short fixed-fee written diagnostic. The AI Operating Design Sprint redesigns one priority workflow at a fixed fee, and the Independent Advisory Seat provides ongoing senior oversight independent of vendors.
/ CONTACT
I have a specific view on why enterprise AI stalls. It may explain your situation. It may not. If it does not, I will say so.