Commonwealth Bank of Australia has built a new AI-powered contact center platform with Microsoft that now handles more than two million customer conversations every month. The system, launched in November 2024 after less than two years of co-engineering, uses a central AI orchestration agent to route voice and messaging queries to the right responder-whether that's a conversational AI, a retrieval-augmented generation (RAG) bot, or a human specialist. The partnership shows how a major bank can move beyond standalone chatbots to a unified omni-channel architecture that separates intelligence from individual channels.
Martin Lindsay, Executive General Manager of Customer Service Direct, began the project in early 2024 with the goal of replacing multiple legacy systems with a single platform. "We wanted to work with Microsoft to shape their products and deliver a platform aligned to our future strategy," Lindsay said. "Together, we've combined CommBank's customer scale and domain expertise with Microsoft's AI and cloud capabilities to build something genuinely new."
Separating intelligence from channels
The core design principle, described by engineering leader Shashank Verma, is "separating intelligence from channels." Instead of embedding AI logic inside each touchpoint, the team built a central orchestration agent using Copilot Studio and Microsoft Foundry. The agent interprets a customer's intent and dynamically decides where to send it. A simple balance inquiry might go to a purpose-built conversational AI. A request involving public bank information uses RAG to pull current content. A fraud dispute follows a deterministic path with strict guardrails. If the system detects vulnerability or sensitivity, it hands off to a human specialist inside Dynamics 365, transferring the full context and providing the specialist with AI-generated summaries and suggested responses.
Rachel Round, who leads self-service customer capabilities, said the handoff logic is deliberate. "Conversational agents can support part of the customer interaction, but we're intentional about recognizing when to bring in a human, especially for interactions where trust and nuance are important. Where a customer's language indicates vulnerability, we expect a human involved to help them with empathy, problem-solving and deeper support."
Co-engineering from Sydney to Seattle
Lindsay tapped Verma to collaborate directly with Microsoft's development teams. The CommBank group spent three weeks at Microsoft headquarters near Seattle validating new AI capabilities against real banking scenarios. After the architecture was proven, Microsoft sent engineers from multiple product groups to Sydney, embedding them on-site with CommBank's team. The joint group navigated the challenge of bringing emerging AI services into a live production environment that was simultaneously supporting 50,000 phone calls a day.
"We started exploring these solutions when they were in their product infancy, so there were a lot of unknowns," Round said. "But Microsoft's AI ambition matched our AI ambition, and it's been a bit of a voyage of discovery for all of us." Verma added, "SaaS platforms can't be black boxes when you run millions of customer interactions. Customers have very low tolerances for failure." To manage that risk, the teams established operational-readiness criteria, deployment safeguards, and automated escalation paths directly with Microsoft's engineering groups.
Measurable impact for customers and staff
The platform's results are already visible. In May 2026, approximately 84.6% of self-service messaging interactions were resolved end-to-end within the messaging channel. The team migrated nearly 700 existing chatbot topics into Copilot Studio and launched what was then Australia's first generative AI banking chatbot. For frontline specialists, conversations are automatically summarized so they can get up to speed quickly, and an AI assistant surfaces relevant answers and policies during live interactions.
The orchestration layer was designed from the start to support future extensions, including voice bots and multi-agent workflows, with the ambition of scaling conversational banking across the entire enterprise. "Conversational experiences from both chat and voice are going to help solve real customer problems in Australia," Verma said. "That's the next scaling challenge."
Why this matters for customer support leaders
The CommBank project offers a practical blueprint for contact center transformation that goes beyond deploying a single chatbot. The architecture separates decision-making from delivery channels, allowing the same intelligence to serve voice, messaging, and future interfaces without duplicating logic. For supervisors, learning how to integrate AI into call center workflows is increasingly critical. The AI Learning Path for Call Center Supervisors provides a structured approach to these skills. The shift toward AI for Customer Support is accelerating, and the CommBank example shows that co-engineering with a technology partner-rather than treating the platform as a black box-can produce systems that meet both operational reliability and regulatory requirements at scale.
Your membership also unlocks:
