On a Wednesday in June, over lunch at Supernormal in Melbourne, we sat down with senior leaders responsible for scaling generative AI across enterprise financial services. The group was candid. Practical. And refreshingly open about what’s working, what’s not, and what’s next.
Co-hosted by DiUS and AWS, the conversation surfaced a familiar truth: most organisations aren’t stuck because of technology. They’re stuck because of the complexity of change. Nearly every organisation is investing in AI, yet only 1% say they’ve reached maturity at scale (McKinsey).
Here are the key themes from the roundtable, and what we recommend if you’re navigating the same terrain.
The messy middle of generative AI adoption in financial services
Some financial services organisations haven’t started. Others are experimenting with copilots, chatbots and sandboxes. And a few are already in production with targeted tools or internal platforms.
But even the most mature teams face the same challenge: turning isolated wins into scalable capability. Building something that can be embedded into governance, delivery and business-as-usual.
And it’s not just us saying this. From our own engagements with Australian financial services clients, and what we heard around the table, there’s a clear pattern:
-
- Internal generative AI platforms are emerging, but scaling them is hard. It requires orchestration, standardisation, and trust.
- Developer copilots show promise, but adoption often stalls. There’s no rollout model, no observability, and devs don’t trust the output.
- In regulated sectors like banking and insurance, compliance readiness and fragmented data access stop generative AI initiatives from moving past PoC mode.
DiUS recommends: Build AI capability into delivery, not on top of it
This is where momentum stalls, not because of a lack of innovation, but because of uncertainty about how to industrialise it. DiUS advocates for AI enablement that spans the full delivery lifecycle. Real scale comes from building alongside your governance, data and engineering practices, not after them.
You don’t need a 50-page strategy deck. You need clear priorities, shared ways of working, and permission to start small and move fast.
From hype to value: making the case for generative AI investment
Most teams still aren’t sure what success with generative AI looks like. It’s easier to frame it as a capability play, a way to help teams move smarter, faster and with more focus, than to articulate specific business returns. But without clear value metrics, scaling becomes a harder sell. And when infrastructure and orchestration costs start to mount, so does the scrutiny.
For financial services, we know most of the value from AI lies in a handful of domains: customer operations, software engineering, risk and compliance, and corporate support. But the real question is: which ones matter most in your environment?
From our work across banking, insurance and adjacent sectors, we’ve found that the most successful programs don’t chase every new idea. They double down on high-leverage, cross-domain use cases, like internal search and summarisation, knowledge assistants, and developer productivity. These deliver repeatable value when tied to shared platforms, patterns and measurable outcomes. The key is choosing what matters most in your context and scaling with consistency.
DiUS recommends: Anchor your efforts in real, measurable improvement
It assesses whether an idea is worth pursuing based on three criteria:
- Desirability: Do people want it? Does it reduce cognitive load, speed up answers, or remove friction for users?
- Viability: Will it deliver meaningful business value, like improving customer satisfaction or reducing operational risk
- Feasibility: Can it actually be built and scaled with the data, platforms and people you already have?
Start where the opportunity is clear and the friction is high. Ask: what are you trying to improve? Speed, quality, consistency, cost? And how will you know when you’re getting there?
Progress comes from proving value early. For example:
- Quantitative impact: Teams using a generative search tool reduced time-to-insight by 40%.
- Qualitative value: Frontline teams now feel confident navigating complex policy documents in under 3 minutes.
And be mindful of infrastructure. The cost of scaling isn’t just in model usage, it’s in orchestration, security, compliance and data integration. That’s where we often see budgets blow out and timelines slip.
Outcome-focused delivery doesn’t just prove ROI, it builds alignment and momentum. Don’t chase interesting use cases. Anchor your efforts in real, measurable improvement.
What’s really holding back generative AI in finance?
Model switching. Latency. Evaluation. Privacy. Security. These aren’t signs of resistance, they’re signs of caution in a space moving faster than most systems, teams, and frameworks can keep up with.
Even technically mature organisations are pausing, not out of fear, but due to the sheer pace of change. The risk of making the wrong call, on tooling, architecture, or governance, is real. And the biggest barrier we continue to see? Poor data foundations. Even the most compelling idea will stall if the data is messy, disconnected or unreliable.
DiUS recommends: Design for change and build in flexibility early
Progress doesn’t come from waiting for the market to settle. It comes from designing for change, iterating fast, with a clear path forward. We help clients build flexibility into solutions. That means:
- Architecting for flexibility across models and vendors
- Embedding observability from day one
- Creating governance that evolves with the use case
And while imperfect systems and messy data don’t mean you can’t get started with generative AI, they do heavily impact the results. And yet, 78% of businesses say they are unprepared for generative AI due to poor data foundations. We’ve seen:
- Poor data quality and unstructured content stop agents from reasoning well in assistants for FAQs, claims and policy navigation.
- Duplicate or mislabelled product data break logic and erode trust in generative AI-powered discovery tools.
So in addition to building for flexibility, we help extract relevant data from varied formats, design scalable pipelines, and ensure the underlying data is usable, observable and governed. So your AI can actually reason, not just respond.
We’re also watching early momentum around agentic AI, where systems act with greater autonomy to achieve business goals. But this isn’t just a step up in model capability. It’s a reconfiguration of how work happens.
For business and technology leaders, the shift to agentic AI demands more than experimentation. It requires reusable architecture, orchestration patterns, and trust frameworks that scale, safely. As we explore in The road to agentic AI, the challenge isn’t just about what agents can do, it’s about what your organisation is ready to let go of, delegate, and evolve. Just as importantly, it’s about what your underlying systems can support — and the skills your teams need to build, govern and adapt as the technology matures
Change management is the hidden multiplier
The hardest part of scaling generative AI isn’t the tech, it’s the shift in how people work. Yet many leaders still underestimate what meaningful change looks like.
We’re seeing organisations raise valid concerns about the cultural impact of generative AI, but few have a clear change management strategy. Training is limited. Onboarding is shallow. New tools are introduced with little context or support. And when adoption lags, the technology takes the blame, not the rollout.
DiUS recommends: Treat generative AI as a capability, not a tool
Scaling generative AI means rewiring how teams work and decisions get made. You can’t bolt AI on to a broken process and expect magic.
We encourage clients to treat generative AI as a long-term capability, not a standalone project. That means:
- Building shared platforms and services that promote reuse and reduce risk
- Establishing internal communities of practice, coaching and peer support
- Giving teams space and permission to experiment, reflect and improve
- Delivering uplift alongside rollout, so capability grows as adoption deepens
In our experience, the organisations making real progress are the ones that design for adoption, not just delivery.
Final word for financial services scaling generative AI?
Generative AI is moving fast. But scaling it takes clarity, flexibility and focus. If you’re in the messy middle, you’re not alone. We’re helping Australia’s banks, insurers and fintechs unlock value from their data and legacy systems, while building what’s next. Let’s talk.
