Data foundations driving investment decisions and operational excellence
Data infrastructure was the throughline at AM Tech Day APAC 2026. Andrew Byron (AustralianSuper), Maddison Liles (Flag Asset Management) and Tom McHugh (FINBOURNE) sat down to work through what solid foundations actually look like in practice, and why they matter before any AI agenda can take hold.
The pressure to do more with less
Tom McHugh opened by setting the industry context. Fee compression, rising product complexity, globalisation requirements and growing internalisation demands among superannuation funds have made the old adage of “doing more with less” more acute than ever.
Andrew Byron, at AustralianSuper managing roughly $420 billion for 3.3 million members, echoed this. With a growing cohort entering the decumulation phase and an increasingly global footprint, the fund has had to rethink its entire operating model. “Data unification and simplification is vitally important,” he said. “Getting the data foundations right so that we can build on them, extend and scale is absolutely essential.” Crucially, AustralianSuper began building its data fabric before AI became a dominant conversation, and Andrew was clear that the work is never finished. “You never hit the perfect operating model. You’re always iterating towards it.”
For Maddison Liles at Flag Asset Management, the challenge was fragmentation. With data spread across multiple systems, Flag’s priority was not to build a data lake but to connect existing systems through an integration layer, a pragmatic approach that delivered speed to market and has since distributed data visibility across the organisation, moving from a situation where only one or two people could access core data to one where five or six can query and interpret it in real time.
What separates successful transformations
Asked what distinguishes firms that succeed at data transformation, Tom gave an answer that surprised some in the room. The most valuable people today are not necessarily those with the deepest subject matter expertise, but those who know where legacy knowledge is buried. “To get to the future, you have to know where you’re starting from. Those people who know where you’re starting from are the ones who will be best able to judge what the systems produce as you move into a world driven by data and AI.”
From foundations to operations
Asked how data has improved AustralianSuper’s ability to execute rebalancing strategies and respond to market opportunities, Andrew was direct. “I see rebalancing as a data execution problem,” he said. The quality and timeliness of data flowing into the rebalancing function determines whether the activity is reactive and costly or thoughtful and optimised, netting trades, minimising cost, done in a rational way with a focused outcome rather than as an afterthought.
Maddison gave a vivid illustration of why data accuracy matters at the operational level. Incorrect start-of-day position data flowing into a risk system creates compounding errors that only surface at end of day. Flag’s response has been to shift checks and balances from the back end of the process to the front end. “That’s a game changer because it’s all about time lag and information.”
Getting ready for AI
The panel agreed that AI readiness is, fundamentally, a data question. Tom identified two core requirements most firms are still struggling to meet: making data available as tools accessible to AI models so they can reason over portfolio positions, and ensuring that data access is properly permissioned and audited. “If I make a decision based on data I wasn’t supposed to see, I’m in just as much trouble as not being able to explain why I made the decision.” The number of platforms capable of doing both correctly, he noted, remains vanishingly small.
On practical AI applications, the panellists touched on several areas. Workflow automation was Maddison’s priority. At Flag, AI could cut credit research time on mortgage-backed securities from four to five hours down to thirty minutes, and free up operational staff from the kind of repetitive email and attachment processing that consumes much of their day. He also gave a striking example of AI mapping obligations across approved policies against a controls database, completing in half an hour what would otherwise have taken weeks. Research augmentation was a key theme for AustralianSuper, helping investment teams evaluate opportunities faster without removing the expert from the loop. And Tom highlighted what he called the “last line of defence” use case: feeding data quality and reconciliation outputs to an AI model to identify what the rules engine may have missed, a low-barrier, high-return entry point gaining rapid traction among clients.
The closing message
All three panellists returned to the same theme in their closing advice: get the foundations right first, favour shorter delivery cycles over multi-year programmes, and start with an integration layer and avoid the data lake trap altogether.
While much of the industry debate around AI focuses on displacement, Andrew offered a counterpoint: “We see things more as an opportunity than a problem. The opportunity to do our job better, share information better, and get better results.”
