Beyond experimentation: What it takes to scale AI safely and effectively

In August, TAL partnered with the Fund Executives Association Limited (FEAL) to deliver an AI-focused luncheon series across Melbourne, Sydney and Brisbane. The sessions explored how the discussion has shifted from  AI experimentation to scaling it safely, with governance, leadership and human judgement now central to success.

Speakers:

• Cameron Pelling, Chief Risk Officer at TAL
• Amber Auld, Head of Insurance, Capital Markets & Banking at Microsoft
• Rakesh Kalyankar, Head of Data Services at AustralianSuper (Melbourne)
• Sadeer Maghak, Chief Technology Officer at NGS Super (Sydney)
• Randike Gajanayake, Chief Technology Officer at Brighter Super (Brisbane)

TAL Moderators:

• Jenny Nguyen, General Manager, Growth & Digital Experience (Sydney and Melbourne)
• Saj Arachchillage, General Manager, Platform Delivery & Engineering (Brisbane)

Last year, TAL, Microsoft and FEAL explored how AI could be integrated into everyday ways of working. At the time, the focus was on experimentation: where AI could create value, how employees could begin applying it, and what responsible adoption might look like. One year on, the conversation has moved as AI becomes more closely connected to organisational strategy. Employees are already using these tools, which are developing quickly, and the number of possible use cases continues to grow.

But that speed creates a new leadership challenge. Governance isn't a handbrake here, done well it gives businesses the confidence to scale safely, direct effort to the right priorities and deliver value that lasts.

From experimentation to execution

Many employees are already applying AI support to everyday workflows. In many cases, they’re moving faster than the businesses around them. The most effective users aren’t outsourcing their thinking to AI: they’re using it as a starting point, then applying critical thinking, quality control and professional judgement to the output. As roles evolve, these human capabilities will become more important, not less.

“As AI agents become part of business workflows, you need to manage the identity of an agent in the same way as you manage the identity of a human being,” Amber said.

“At the same time, organisations need to help manage anxiety about job loss and help their people build AI capability, because it's far worse for somebody to lose their job and not have AI skills at all.”

To move from scattered testing to meaningful adoption, leaders need clear direction on where AI should be used, how it supports the business strategy and the value it’s expected to create.

Leadership needs to focus AI on the right opportunities

With new tools, models and possible applications emerging quickly, the challenge for leaders is no longer how to encourage experimentation. It’s how to concentrate effort on the work that matters most.

Innovation may start with employees testing new ideas, but transformation can’t rely on grassroots activity alone. Scaling AI requires a transparent leadership position, alignment between the board and executive team, and visibility of the priority use cases being funded across the business.

AI shouldn't become the strategy itself: its role is to solve real business problems, not to justify its own use. As organisations move from individual use cases to broader workflows, the possibilities become almost infinite, and it’s the responsibility of leadership to decide what proceeds, what gets paused or reshaped, and what gets set aside.

Governance is what turns pilots into production

Moving from a controlled pilot into production introduces risks that can’t be managed once and then left alone. One risk is automation bias, the natural tendency to place greater trust in automated output once it becomes part of everyday work. Another is model drift, where an AI system’s behaviour changes over time as its data, context or interactions evolve.

“Governance needs to follow the data: you need to know who has access to it, which models are using it and where it is stored because those details become much more important once AI is operating in everyday business workflows,” Cameron said.

These risks show why AI governance can't be treated as set and forget. As organisations introduce more tools and agents, they need strong auditability, clear accountability and ongoing evaluation, applied more heavily where the risk is higher, to confirm systems continue to behave as intended.

That need becomes even more important as AI agents begin interacting with other agents or making decisions at speed. In these environments, monitoring and evaluation will need to be continuous, with governance operating in or near real time rather than through periodic review.

This is where strong oversight becomes an enabler of responsible innovation. It provides the structure to test, learn, strengthen controls and move forward with confidence, rather than relying on enthusiasm or speed alone.

Boards need ongoing engagement, not one-off briefings

Boards have a strong appetite to understand how AI can support business strategy, but they’re also being asked to carry growing oversight. That means directors need more than occasional updates. They need practical examples, regular education and clear explanations of where the AI is being used, where it’s not being used and where uncertainty remains.

Management has an important role to play in bringing boards on that journey. This includes being clear about what is known, what’s still being tested, what risks are being managed and what decisions may need to come back for further consideration.

A shared language makes governance practical

Almost every part of an organisation is pushing for acceleration. Business teams want to address opportunities and pain points, employees want access to new tools, and leaders want to realise value. Technology, information security and risk teams are often the groups asking whether implementation will be secure, responsible and sustainable.

To make good decisions, businesses need people who can translate between these perspectives. Operational teams understand the problem and desired outcome. Technology, information security and risk teams understand the systems, data, controls and operating environment needed to deliver it responsibly.

That shared language also helps teams avoid chasing the next ‘shiny’ AI opportunity without understanding whether it’s genuinely needed, how it will be sustained and what protections need to sit around it. It allows governance to act as a navigator, helping the business move from pilot to scaled adoption with greater confidence.

If you'd like to revisit where this conversation started, last year's session is covered in  Embedding AI into the workplace.