In June 2025 the Commonwealth Bank switched on a voice bot on its enquiries line, and a month later it told 45 call-centre people their jobs were going because the easy calls were no longer reaching them. The people answering the phones, and their union, said the opposite was happening: calls were rising, overtime was being offered and team leaders were being pulled back onto the handsets, which is why the dispute went to the Fair Work Commission. On 21 August 2025 the bank reversed itself, saying the first assessment had not weighed every relevant business consideration and that the roles were not redundant, then gave each of the 45 a choice to stay, move inside the bank or leave with a payout. Julia Angrisano's line that workers want a tech-savvy bank but expect to be part of the change, not replaced by it, sits beside the later figures: by May 2026 the chatbot was closing almost nine in ten conversations with no person involved, even as the bank said it had added more than 140 people to its Australian contact centres, while Bloomberg reported hundreds of contractor chat roles gone in Johannesburg. Separating what the dashboard claimed from what the phones were actually doing keeps the argument honest, and the closing point lands cleanly: they counted what the bot took and did not count what came back to people. Work notes from an Australian town where long shifts still depend on human judgment are collected on my profile.
The line that matters is the one from the phones. The dashboard said calls were falling. The people answering them said calls were going up, and managers were offering overtime to cover it. The bank trusted the number, not the workers, and 45 people lost their jobs for a month. The bot took the easy calls and left the hard ones to people. Check what an agent really does in use, not what its score says. https://aideliverybrief.substack.com/p/singapores-new-test-kit-says-a-right
I think the happy ending here is a bit too tidy. The bank said sorry and gave the 45 their jobs back, but a year later the chatbot is closing almost nine in ten chats and contractor roles are gone. That looks less like a lesson learned and more like the same cut just moved where fewer people notice it. The dashboard was wrong, sure, but the bigger miss is treating a public apology as proof the pressure went away when the numbers say it just found a quieter target.
man the dashboard vs the actual phones thing got me, like the numbers can look so clean and still be totally wrong and then real people are doing overtime while the bot is supposedly handling everything. and I keep thinking about that with my own stuff cause I'm recommending agentic workflows to clients all morning and then I catch myself like wait would I actually roll this out in my own businesses like this, no human checking the messy calls. the bot taking the easy ones and leaving the hard ones for the humans is so real too, that never shows up in the demo and then the team just gets slammed. anyway love that he sells automation and still called this one wrong, that's the kind of thing that makes me trust the take more. and yeah ask the people on the phones first, always
The union saying calls were going up while the dashboard said down is the whole cautionary tale. The people closest to the work were right and nobody asked them.
In June 2025 the Commonwealth Bank switched on a voice bot on its enquiries line, and a month later it told 45 call-centre people their jobs were going because the easy calls were no longer reaching them. The people answering the phones, and their union, said the opposite was happening: calls were rising, overtime was being offered and team leaders were being pulled back onto the handsets, which is why the dispute went to the Fair Work Commission. On 21 August 2025 the bank reversed itself, saying the first assessment had not weighed every relevant business consideration and that the roles were not redundant, then gave each of the 45 a choice to stay, move inside the bank or leave with a payout. Julia Angrisano's line that workers want a tech-savvy bank but expect to be part of the change, not replaced by it, sits beside the later figures: by May 2026 the chatbot was closing almost nine in ten conversations with no person involved, even as the bank said it had added more than 140 people to its Australian contact centres, while Bloomberg reported hundreds of contractor chat roles gone in Johannesburg. Separating what the dashboard claimed from what the phones were actually doing keeps the argument honest, and the closing point lands cleanly: they counted what the bot took and did not count what came back to people. Work notes from an Australian town where long shifts still depend on human judgment are collected on my profile.
All my articles: https://lukeybkarratha.substack.com/archive
The line that matters is the one from the phones. The dashboard said calls were falling. The people answering them said calls were going up, and managers were offering overtime to cover it. The bank trusted the number, not the workers, and 45 people lost their jobs for a month. The bot took the easy calls and left the hard ones to people. Check what an agent really does in use, not what its score says. https://aideliverybrief.substack.com/p/singapores-new-test-kit-says-a-right
I think the happy ending here is a bit too tidy. The bank said sorry and gave the 45 their jobs back, but a year later the chatbot is closing almost nine in ten chats and contractor roles are gone. That looks less like a lesson learned and more like the same cut just moved where fewer people notice it. The dashboard was wrong, sure, but the bigger miss is treating a public apology as proof the pressure went away when the numbers say it just found a quieter target.
man the dashboard vs the actual phones thing got me, like the numbers can look so clean and still be totally wrong and then real people are doing overtime while the bot is supposedly handling everything. and I keep thinking about that with my own stuff cause I'm recommending agentic workflows to clients all morning and then I catch myself like wait would I actually roll this out in my own businesses like this, no human checking the messy calls. the bot taking the easy ones and leaving the hard ones for the humans is so real too, that never shows up in the demo and then the team just gets slammed. anyway love that he sells automation and still called this one wrong, that's the kind of thing that makes me trust the take more. and yeah ask the people on the phones first, always
The union saying calls were going up while the dashboard said down is the whole cautionary tale. The people closest to the work were right and nobody asked them.