Commonwealth Bank is preparing to take an artificial intelligence system it co-built with Microsoft well beyond the retail bank, after the technology quietly became the traffic controller for more than two million customer conversations a month.
The system is an AI orchestration agent. Rather than being another chatbot bolted onto a single channel, it sits in the middle of the bank’s contact operations, interprets what a customer is actually asking for, and routes each interaction to the most appropriate destination, whether that is a specific AI capability or a human specialist. The bank and Microsoft spent two years developing it.
The plans were detailed by iTnews, drawing on a public case study published by Microsoft. CommBank now intends to extend the platform across the entire bank, adding voice bots, multi-agentic workflows and, eventually, enterprise-wide conversational banking.
Separating intelligence from channels
The architecture is the interesting part. According to Microsoft’s case study, the platform is built on Copilot Studio, Dynamics 365 and Microsoft Foundry, the vendor’s “AI app and agent factory”. CommBank engineering leader Shashank Verma described the guiding principle as separating intelligence from channels, meaning the logic that decides how to handle a request lives in one orchestration layer rather than being duplicated inside every app, phone line and messaging window.
That design has a practical payoff. The bank modernised and migrated close to 700 existing chatbot topics into Copilot Studio, and in November 2024 launched what Microsoft describes as Australia’s first generative AI banking chatbot. The orchestration agent handles more than two million conversations a month across voice and messaging, and supports around 50,000 phone calls a day.
The headline number is resolution. In May 2026, about 84.6 per cent of self-service messaging interactions were resolved end to end in the messaging channel, without being handed off. Rachel Round, a general manager in the bank’s customer experience group, told the case study the system produced a step-change in how effectively enquiries are being resolved through digital and messaging channels. Round also stressed the human boundary, saying the bank is intentional about recognising when to bring a person in.
Martin Lindsay, executive general manager for Customer Service Direct, framed the relationship as shaping Microsoft’s products around the bank’s roadmap rather than simply buying a finished tool.
Why it matters for Australia
This is the largest bank in the country moving past the chatbot era into agentic orchestration at genuine production scale, and it doubles as a template. Copilot Studio and Foundry are Microsoft’s pitch to every large Australian organisation weighing how to deploy agents in high-volume, high-stakes operations. A working reference at CommBank’s scale gives boards in insurance, telecommunications, utilities and government a concrete answer to the question of whether this stack survives contact with millions of real customers.
It also lands in a politically charged spot. CommBank’s automation of customer service has already collided with its workforce. In July 2025 the bank moved to cut 45 call-centre roles, saying an AI voice bot had reduced call volumes. As The Register and Information Age reported, the Finance Sector Union took the matter to the Fair Work Commission, arguing call volumes were in fact rising, with staff offered overtime and team leaders pulled onto the phones. In August 2025 the bank reversed the redundancies and acknowledged an error in declaring the roles surplus.
The lesson embedded in that episode maps closely onto the technology CommBank now wants to expand. An orchestration agent that resolves 84.6 per cent of messaging enquiries is not the same as one that removes the need for humans, and the bank’s own experience shows how quickly a productivity claim can unravel when the underlying call data is contested. The 84.6 per cent figure covers self-service messaging resolved in-channel, not the full spread of voice and complex enquiries, and the roadmap explicitly keeps humans in the loop for sensitive conversations, with AI drafting summaries and suggested responses in the background.
An agentic layer, and the workforce question
The expansion also fits a broader CommBank strategy of building what it calls an AI-native bank. The lender has committed a $90 million, three-year Future Workforce Program to reskill staff, and opened a technology hub in Seattle to sit near partners including Amazon Web Services and Anthropic. The Microsoft orchestration agent is one pillar of that agentic layer rather than the whole of it.
The next moves will test how far the orchestration model stretches. Voice is materially harder than text, multi-agent workflows introduce new questions about error handling and accountability when several agents chain together, and enterprise-wide conversational banking would push the system into business banking and internal operations where the tolerance for a wrong answer is lower and the regulatory scrutiny higher.
For the rest of corporate Australia, CommBank has become the case study to watch. If the bank can scale agentic orchestration beyond retail without repeating the missteps of 2025, it will have written the local playbook for deploying Microsoft’s agent stack in regulated, high-volume operations. If it cannot, the gap between the demo and the day job will be visible at the scale of two million conversations a month.
Sources: iTnews, Microsoft Source Asia, The Register, Information Age (ACS), American Banker.
