The Bank That Fired 45 Humans for an AI Bot—Then Hired Them Back What you can learn from their mistakes… Commonwealth Bank of Australia tried to replace 45 call-center staff with an AI voice bot. The bank said the bot was handling thousands of routine calls a week and cut positions to match. Then the cracks showed. Call volumes actually went up. Customers were frustrated, frontline staff were swamped, and even team leaders had to jump back on the phones. Under union pressure and public backlash, CBA reversed course. They admitted the savings weren’t genuine, apologized, and offered jobs back. What started as a cost-cutting move turned into a reputational black eye. What they should have done instead: Start small. Let the bot handle low-stakes, high-volume tasks—password resets, balance checks, card activation. Segment calls. Route sensitive cases—bereavement, fraud disputes, mortgage hardship—straight to humans. Pilot, don’t plunge. Test in a limited branch or call group first. Measure performance and customer feedback, not just “calls deflected.” Involve frontline staff. They know where callers get frustrated. Ignoring their input guarantees blind spots. Scorecard it. Track: average handling time, customer satisfaction, first-call resolution. If those dip, pause. Iterate before scaling. Fix bugs and edge cases in the sandbox before putting headlines—and reputations—on the line. AI will replace a lot of routine work. But the rollout can be the difference between a win and a very public U-turn. How are you testing AI tools at your company before betting the brand on them?