The problem
CBA serves more than 10 million customers, and its contact centre carries much of their everyday demand. Messaging had become a popular channel, with retail customers making more than 50,000 enquiries a day, a five-fold increase on usage five years earlier, the bank said in November 2024 (Commonwealth Bank of Australia). The same release said the messaging platform could not answer about 10 per cent of enquiries, and that the bank expected generative AI to raise the speed, quality and precision of responses.
The bank has about 2,400 contact centre staff, and its testing of a generative AI chatbot called Hey CommBank in late 2024 raised fears that it would replace many of them (Information Age). The business question was the one many service leaders face: how much human capacity the contact centre still needs once AI handles a share of the work.
What they did
CBA built AI into two channels. In messaging, it added machine learning and generative AI to its existing infrastructure and put 11 guardrails around customer-facing generative AI to protect its models from malicious inputs and misleading outputs (Commonwealth Bank of Australia). In voice, it deployed a bot that triages inbound calls (techpartner.news).
In late July 2025 the bank placed customer service roles under review after the bot went live. A spokesperson said at the time that “some roles and work can change” as technology is used, according to Information Age. The bank told staff in its Customer Service Direct business that 45 roles were no longer needed.
The Finance Sector Union (FSU) challenged the cuts. The bank had said its voicebot cut call volumes by 2,000 a week, but it had to concede that calls had risen, with managers using overtime and team leaders taking calls, The Nightly reported, citing the Australian Financial Review (The Nightly).
After a three-week review, CBA reversed course on August 21, 2025. Affected staff could stay in their roles, seek redeployment or leave with their redundancy, and the bank apologised and said it was reviewing its internal processes (techpartner.news). It kept the AI. By October 2026 it described a contact centre transformation that brings self-service and specialist support together across messaging and voice (Commonwealth Bank of Australia).
Results
The bank’s own figures, all self-reported, show the AI channels carrying more of the load. In October 2026 CBA said around 85% of in-app messaging interactions are now resolved through self-service, and that the service’s ability to understand customer requests first time rose from 60% to 88% of interactions (Commonwealth Bank of Australia). It said more than 2 million customer conversations a month are supported across self-service and specialist channels, and its messaging service won an industry award for its generative AI capability.
Earlier, the bank said AI-powered app messaging helped reduce call centre wait times by 40 per cent over the financial year to June 2024, according to its November 2024 release.
The workforce result went the other way. The 45 customer service jobs CBA cut in July 2025 were reinstated a month later, after the bank confirmed to Information Age that its decision was incorrect. The bank’s claim that the voicebot cut call volumes by 2,000 a week did not hold up against rising calls, as reported by The Nightly.
What didn’t go to plan
The redundancy decision was wrong on the bank’s own account. A spokesperson said the initial assessment “did not adequately consider all relevant business considerations and this error meant the roles were not redundant” (Information Age). The bank had cut capacity on the strength of a volume reduction that its frontline data did not bear out.
The reputational cost was public. The FSU said CBA “has been caught out trying to dress up job cuts as innovation”, and called the episode the “tip of the iceberg” for financial institutions using technology to cut jobs. Staff who had been told their roles were redundant waited weeks for the reversal.
The tension has not gone away. In July 2026 the FSU said CBA planned to cut 176 technology and engineering jobs and lodged a formal dispute; the bank denied the cuts directly result from its use of AI and pointed staff to its $90 million Future Workforce Program (Information Age). CBA has not published independent evidence of the AI’s effect on cost, call volumes or headcount.
Lessons
- Validate the volume claim before cutting capacity. CBA cited 2,000 fewer calls a week, yet calls rose and team leaders were pulled onto the phones.
- Separate the technology decision from the workforce decision. The bank reversed the job cuts but kept the bot, because the AI and the redundancy assessment stood on different evidence.
- Measure deflection in the channel where it lands. Around 85% self-service resolution in messaging says little about demand in voice, where the bot sat.
- Bring employee representatives in early. The FSU dispute turned a three-week review into a public admission of error.
- Put guardrails in writing before customer-facing generative AI goes live. CBA set 11 guardrails against malicious inputs and misleading outputs before scaling messaging.
Adapt it in your organization
Is this for you?
- Your contact centre handles high volumes of routine enquiries across messaging and voice.
- You can measure demand by channel each week, before and after an AI change.
- Your workforce agreements require consultation before roles change.
- You have a sponsor who owns both service levels and headcount decisions.
- You can test a bot’s answers against a set of misleading and malicious prompts before launch.
A 90-day version
Pilot: route one category of routine inbound calls through a voice bot while holding staffing flat, and track calls, transfers and handling time by week against a matched queue. Owner: head of customer service. Success measure: a sustained fall in human-handled calls for that category over at least eight weeks, with no rise in repeat contacts.
Metric to track
Human-handled contact volume per week by channel, not bot containment alone. CBA reports around 85% self-service resolution in messaging, while its voice channel saw calls rise; the Klarna case shows a similar gap between AI share and staffing needs.
Failure modes to pre-empt
- Cutting roles on a projected volume reduction before it shows up in real call data.
- Announcing AI-linked job cuts without consulting staff or their union.
- Reporting messaging self-service rates as if they applied to voice.
- Leaving overtime and team-leader call handling out of the capacity numbers.
Questions to take to your leadership team
- What evidence of lower demand would we need before reducing contact centre roles?
- Who signs off a redundancy linked to AI, and what data do they see?
- How would we explain a reversal to staff, customers and investors if we got it wrong?
Executive FAQ
Did Commonwealth Bank replace call centre staff with AI?
It cut 45 customer service jobs after deploying a voice bot in 2025, then reversed the decision in August 2025 and called it an error.
Is Commonwealth Bank still using AI in customer service?
Yes; in October 2026 it said around 85% of in-app messaging interactions are resolved through self-service.
About the curator & AI disclosure
Curated by Luciano Ayres
Luciano Ayres is Director of Artificial Intelligence at ArcTouch, with more than 20 years of experience in software and technology leadership and specialized studies in deep learning. Connect with him on LinkedIn.
AI-generated briefing. This case was produced through an automated AI research and writing workflow curated by Luciano Ayres. Sources are linked for independent verification; summaries and analysis may contain errors and do not replace professional due diligence.