A Bank Just Drew a Hard Line Between What Its AI Can Say & Decide To Do

The Rollout

Commonwealth Bank of Australia has begun testing CommBank Companion, an AI-powered conversational assistant built directly into its mobile app, backed by roughly $1 billion in security investment. Unlike most vendor-led AI launches, this one was designed and built entirely in-house by CommBank’s own engineering and data science teams, and it is being tested first with employees and select business banking customers before any wider rollout to the bank’s more than 9 million app users.

Why This Matters for BFSI and Collections

What makes this deployment distinctive is not the technology; it is the explicit boundary CommBank drew around it. The bank stated plainly that Companion does not provide regulated financial advice or credit assistance on regulated products, and does not make decisions on behalf of customers. As the announcement put it, customers remain in control of decisions and actions at all times, with Companion designed to support and inform, not replace, customer judgment or professional advice. For collections and lending operations, this is a concrete example of a regulated institution defining in public exactly where its AI’s authority ends and human or customer judgment begins, rather than leaving that line implicit.

What the Numbers Do Not Say Out Loud

The $1 billion security figure sounds like it answers the trust question, but it mostly buys infrastructure: fraud monitoring, real-time safeguards, and identity protection. It does not by itself answer the harder design question CommBank is actually wrestling with, which is how a conversational AI should behave when a customer’s request edges toward something that looks like advice. Executive Mike Vacy-Lyle’s comment about the assistant eventually helping business owners “sense-check financial decisions before they act” is telling: even a bank with this level of security investment is deliberately keeping the AI in a summarizing and flagging role rather than a recommending one. That restraint, not the security spend, is the harder and more consequential decision.

The Practical Read

Sam Altman’s comment at the same summit, that leading organizations succeed by allowing controlled adoption rather than rigidly pre-setting every policy, matches CommBank’s own staged approach: test with employees, then select customers, then expand gradually. For BFSI and collections operations building or evaluating voice AI, the lesson is to match CommBank’s discipline about scope. A collections voicebot should be just as explicit about what it will never decide unilaterally (payment plan approvals, hardship determinations, dispute outcomes) as CommBank has been about credit advice. Governance done well is a boundary stated clearly enough that everyone knows exactly where it sits.

[Read the full report]

Related Post

91% of Service Leaders Face Executive Pressure to Deploy AI

A major Gartner survey of 321 customer service and support leaders, conducted in October 2025, has delivered a striking reality check for corporate leadership: 91% of service and support leaders are facing intense executive pressure to implement AI in 2026. While the baseline urgency to automate is clear, the downstream impact on frontline personnel reveals […]

A Tribunal Ruled an Airline’s Chatbot Was the Airline

The Trust Gap New data paints a sobering picture of where AI-driven customer service actually stands. Forrester found US CX quality fell again to a record low of 68.3 in 2025, while the CFPB reported roughly 6.6 million consumer financial complaints that same year, more than double the prior year. Separately, Salesforce research found that […]

P&C Insurers Are Tripling AI Spend, But Only 38% Are Seeing Value at Scale

The Gap Between Spending and Results BCG’s AI Radar shows property and casualty insurers will triple AI spending as a share of revenue in 2026, yet only 38% are generating value at scale from AI in core workflows like underwriting and claims. BCG frames this as a strategic gap rather than a technical one: AI […]