Westpac is building what it calls an agentic data ecosystem, and the numbers it published are the interesting part. The bank now feeds 285 data sources into its intelligence layer, up from 251 in November. Building and deploying an AI model got five times faster. Turning a customer insight into an actual engagement got six times faster.
Chief digital, data and AI officer Andrew McMullan announced the work at a Snowflake customer event in Sydney, describing the ecosystem as a single living system for business definitions, data lineage, governance, quality and ownership. That framing matters more than it sounds.
What Westpac actually built
The intelligence layer came first and the agents sit on top of it, according to iTnews. The layer holds the shared definitions and the lineage: what a number means, where it came from and who owns it. Only then do the agents get pointed at it.
The clearest business result is in campaigns. Instead of building an offer from scratch, the right one now surfaces from a pre approved catalogue, already matched to the engagement and already assured. What used to take months now takes days.
The three platforms for bankers
Westpac deployed three platforms built on Snowflake’s CoWork and CoCo assistants, and the names give away the design. Start Your Day Informed pulls daily market intelligence from thousands of signals across bank systems.
Create a Better Day lets commercial and institutional bankers analyse economic activity across regions, spot opportunities and prepare underwriting conversations, as reported by Cyber Daily. Sleep Tight at Night watches risk and credit control overnight.
None of the three is a chatbot bolted onto a portal. Each one wraps a routine the banker already had, at the moment they already had it, which is the difference between a tool people are told to use and one they open first thing.
Governance came before the agents
McMullan’s argument is that most organisations do not have a data shortage. They have a confidence problem: nobody trusts the number enough to act on it without checking it first. Definitions, lineage and controls are what remove the checking step.
That ordering is the whole lesson. An agent pointed at ungoverned data does not produce a slow decision, it produces a confident wrong one, faster than anyone can catch it. The five fold and six fold gains came after the foundation, not instead of it.
Why Australian businesses should pay attention
This is a major Australian bank publishing figures on agents in production, which is rarer than the volume of AI announcements suggests. Most of what reaches the press is a pilot with a friendly quote. Westpac named the platforms, the source counts and the multiples.
The sequence is portable to a business of any size. Agree what the numbers mean, record where they come from, then automate. Doing it the other way round is how finance automation projects end up producing output that nobody will sign off on.
It also lands the same week as research showing what happens without that scaffolding. Anthropic found groups of agents sabotaging each other and agreeing price floors when their goals collided, which we covered in the piece on agents in groups. Westpac’s answer is the governance layer underneath.

What this signals for the sector
Expect the other three big banks to publish something comparable within months, because the competitive claim here is speed to offer rather than a better model. That is a race won on data plumbing, not on which vendor you signed with.
For everyone below the majors the useful read is that the hard part was never the agent. Westpac spent its effort on definitions, lineage and ownership, and the agents were the straightforward bit at the end. AI strategy work that jumps straight to the agent is skipping the part that made these numbers possible.
