The early AI use cases are familiar: drafting investment commentary, summarising research, preparing first-pass reports, interpreting documents and supporting knowledge retrieval.

These use cases are valuable. They create familiarity, improve productivity and help organisations build confidence. But they do not yet amount to operating model transformation.

The real complexity begins when AI moves closer to core investment and operational processes - mandate monitoring, reconciliation exception triage, performance validation, pricing oversight, regulatory reporting preparation, fund accounting quality assurance and data integrity reviews. These are workflows where accuracy, accountability and control are non-negotiable.

That is where AI stops being a helpful assistant and starts becoming part of the way the business works.

Financial services businesses are often highly process-based, but not always well documented end to end. People understand their part of the chain, but upstream and downstream dependencies are often far less visible. Decision logic may sit with individuals rather than within formal documentation. Control ownership may be assumed rather than clearly defined.

This has always been manageable - up to a point.

AI changes that.

If an organisation wants AI to support or automate part of a process, it has to make that process explicit. Inputs, outputs, business rules, exception pathways, escalation points and ownership all need to be clear. What used to live as tacit knowledge in the business has to be translated into operating logic.

That is why AI adoption is not primarily a tooling problem. It is a process architecture problem.

There is a tendency to think about AI adoption as a technology challenge. In reality, it is just as much a business implementation challenge.

The organisations that will struggle are not necessarily the ones with the weakest models. They are the ones with the least process clarity, the most fragmented ownership, and the greatest resistance to changing how work gets done.

That resistance is not new. Every major transformation program runs into it. Teams protect spreadsheets. People hold onto manual controls. Business units resist standardisation. Functions optimise for their silo rather than the full process. When the proposed change involves AI, that hesitation is amplified by understandable concerns around trust, control and role impact.

That is why the transition matters so much.

AI changes work. It moves people from doing the task to overseeing the task.

For many organisations, that is the real transformation. Not prompt engineering. Not model selection. But redesigning roles, clarifying accountability, and helping teams understand how to supervise, challenge and improve AI-supported outcomes rather than simply execute the process themselves.

As AI investment grows, the question is no longer whether there is value in the technology. The question is whether that value can be translated into measurable operational outcomes.

That requires discipline in three areas.

First, firms need to structure the opportunity properly. Not every process is suitable. Not every workflow needs AI. Some activities are better solved through traditional automation. Others may benefit from AI-assisted support rather than higher-autonomy execution. The starting point is clear use case definition, business case assessment and practical prioritisation based on business impact, operational feasibility and control requirements.

Second, firms need to support adoption in the business. That means documenting SOPs, clarifying decision logic, mapping controls, defining oversight models and guiding teams through the shift from execution to supervision. Without that work, AI remains a pilot or an adjacent tool rather than an embedded capability.

Third, firms need to build assurance in from the start. AI cannot scale without confidence. That confidence is built through testing, validation, traceability, exception management and clear governance. If the business cannot explain how the output was reached, who reviewed it, what controls apply and how exceptions are handled, adoption will remain slow and fragile.

In other words, AI return on investment is not determined at procurement. It is determined in implementation.

This matters even more in Australia’s current regulatory environment.

CPS 230 is not an AI regulation. But it is highly relevant to AI adoption. Once AI becomes part of a critical workflow, it becomes part of the operational risk environment. It raises questions around accountability, controls, resilience, testing, monitoring and third-party dependency.

That creates a clear divergence.

The firms that build AI into structured operating models, with documented processes, clear control points, audit trails, human oversight and repeatable validation, will scale faster and with greater confidence.

The firms that move forward in an unstructured way may still progress quickly at first. But they are more likely to spend the following years remediating control gaps, reworking workflows and rebuilding trust in outcomes that were never properly embedded.