Westpac Banking Corporation, Australia’s third-largest bank by assets, has begun systematically tracking artificial intelligence token consumption across its workforce and is actively routing lower-complexity tasks to cheaper models in an effort to control AI costs that have grown significantly since the bank’s initial generative AI deployments. Dan Jermyn, Westpac’s new chief AI officer – who joined from larger rival Commonwealth Bank of Australia earlier in 2026 – disclosed the program publicly this week, describing a governance approach centered on ensuring staff understand which AI model is appropriate for which task rather than defaulting to the most powerful and expensive available option. The bank wants staff to make sensible choices about the models and tools they use to get the best return, Jermyn said, and is tracking token consumption including in its software development teams, which have access to the most advanced and therefore most expensive model tiers. NEWSCENTRAL reads this as an important institutional confirmation that the enterprise AI cost problem that erupted across the technology sector in the first half of 2026 has now reached the conservative, compliance-sensitive environment of major Australian financial services – a sector not previously known for rapid AI adoption at the scale required to generate meaningful token governance concerns.
Westpac’s token governance initiative sits inside a larger transformation program called UNITE, which is designed to consolidate the bank’s sprawling technology stack – a legacy of multiple acquisitions including the 2008 merger with St. George Bank – from approximately 180 discrete applications to fewer than 60. The technology complexity that UNITE addresses is itself a partial explanation for why AI governance is both more important and more difficult at Westpac than at banks that built their infrastructure more recently. A bank with 180 applications running partially redundant processes has more integration points where AI costs can accumulate invisibly than one whose systems are centralized. Jermyn has noted that the UNITE consolidation will ultimately improve how the company makes use of AI by providing cleaner data infrastructure and simpler integration pathways for AI tools.
More than 15,000 Westpac staff – under half of its total employee base – are currently using generative AI to support their work, according to the bank’s most recent annual report. That adoption level, while already substantial, is expected to grow significantly as the bank embeds AI capabilities directly into customer-facing mobile and online banking applications by the first quarter of its 2027 fiscal year. An internal coding experiment produced a 46% productivity gain for software engineers using AI assistance compared with a control group performing the same tasks conventionally, with no measured reduction in code quality. A separate pilot with AI agents and Accenture’s agentic infrastructure team completed a software migration task in one hour that would have taken a human engineer approximately six days. Nathan Clark, Enterprise IT and Systems Architecture Analyst at NEWSCENTRAL, notes that the coexistence of demonstrated productivity gains and emerging token cost pressure at the same institution illustrates the maturation inflection that enterprise AI is reaching: adoption has succeeded well enough to generate meaningful cost exposure, which is a different problem than the adoption challenges that dominated enterprise AI discussions eighteen months ago.
The competitive context is one in which Commonwealth Bank, Australia’s largest bank and Jermyn’s former employer, is widely considered to have established the most advanced AI deployment among Australian financial institutions. Jermyn’s move from CBA to Westpac is itself a signal: banks are competing for AI leadership talent as directly as they compete for capital, and the appointment of a chief AI officer with direct experience at the market leader is a deliberate attempt to compress the gap. Westpac’s consumer division data shows digital cost-to-serve at 65% lower than branch-equivalent interactions, with three in four simple sales now executed digitally – figures that quantify the financial case for continued AI investment even as token costs create pressure to optimize the economic model of that investment.
The token governance challenge that Westpac is now formalizing will be recognizable to technology companies that confronted the same problem earlier in 2026: the transition from treating AI as a capability to be adopted to treating it as an operational cost to be managed requires a governance infrastructure – model routing rules, consumption monitoring, usage attribution to business units, and ROI frameworks – that most organizations did not build during the adoption phase. Building that infrastructure after adoption is already widespread is harder than building it alongside deployment, because it requires changing established workflows and user behaviors rather than setting expectations from the start.
The comparison with the technology sector is instructive but imperfect. Technology companies confronted the AI cost problem in a context where token governance failure was an embarrassing earnings miss; for Westpac, operating under APRA oversight with strict operational risk management requirements, the governance gap carries additional regulatory exposure if AI cost controls are found inadequate in a supervisory review. That difference in consequence – reputational versus regulatory – is why NEWSCENTRAL expects Australian financial services institutions to implement AI cost governance frameworks more formally and more quickly than their technology sector counterparts did.
Jermyn’s framing of the challenge – competitive advantage comes not from access to any one model but from how AI is used efficiently across the bank – is the mature institutional perspective on AI deployment that NEWS CENTRAL expects to hear from more financial services leaders over the remainder of 2026. The technology sector learned this lesson under fiscal pressure in the first half of the year; the financial services sector is learning it under the combined pressure of adoption maturity and cost visibility that comes with scale. The banks that build effective token governance frameworks now will hold an operational efficiency advantage as the next generation of more powerful and more expensive AI capabilities arrives.