Do banks accurately assess whether AI can replace customer service jobs?

In July 2025 the Commonwealth Bank of Australia announced it would cut 45 customer service roles, saying the work would be covered by an AI-powered "voice bot" it had been testing under the name Hey CommBank. Information Age (ACS) reports that weeks of pressure from the Finance Sector Union (FSU) led CBA to reverse the decision in August, with a spokesperson admitting "CBA's initial assessment that the 45 roles in our Customer Service Direct business were not required did not adequately consider all relevant business considerations and this error meant the roles were not redundant." The FSU said call volumes to the bank's customer service line were increasing, not falling, and that management had been covering the gap with overtime and by pulling team leaders onto calls — the opposite of what a successful AI substitution would look like.

CBA's own account frames the reversal as a failure of internal process rather than of the technology itself: the bank says it is "reviewing our internal processes to improve our approach going forward," implying the error was in how thoroughly the staffing need was assessed before the AI tool was judged capable of covering it. The report places CBA alongside two other named cases of the same sequence — Klarna, which rehired human customer-service staff after replacing them with AI chatbots, and IBM, which rehired staff after replacing much of its HR division with AI — treating "cut jobs for AI, then rehire" as a repeating pattern rather than an isolated misstep, and quotes the FSU calling the CBA case "the tip of the iceberg" of financial institutions using technology to cut jobs.

This sits in tension with evidence elsewhere in the library that AI assistance tends to produce measurable productivity gains for the support staff who keep their jobs rather than eliminate the need for them: Does AI assistance help less experienced workers most? found an assistant raising throughput for a large support workforce, not replacing it. The CBA case is a sharper, earlier-stage version of the gap the Danish data describe at national scale: Does AI chatbot adoption change worker pay and hours? found employers reorganizing tasks around AI well before any effect on headcount or pay appeared, consistent with CBA jumping to a headcount decision before the task-level reality (rising call volume) was accounted for. It is also a visible, case-level instance of Do firms substitute labor for AI at different rates?, which found substitution away from labor happens at firm-specific rates rather than uniformly — CBA substituted faster than its internal AI capability actually supported.

The excerpt is a single news report relying on the company's own admission and a union's account of call volumes; it does not give the voice bot's actual performance data, how CBA measured the roles as "not required" in the first place, or whether the 45 roles would have been restored without six weeks of organized union pressure. It cannot establish how common this reversal pattern is beyond the three named companies, only that a trade press outlet treats it as recurring. The implication it supports is narrow but concrete: at least some AI-driven layoff decisions are being made ahead of operational evidence that the AI can absorb the work, and are reversible under worker pressure once that evidence — rising call volume, overtime — surfaces.