Commonwealth Bank reversed 45 AI layoffs after call volumes rose
In June 2025, the Commonwealth Bank of Australia put a voice bot on its inbound customer service line. One month later, it cut 45 support roles. It was the first Australian bank to publicly attribute layoffs to AI. By August 20, every one of those redundancies had been reversed, and the bank issued a written apology.
The bot was fine. The measurement wasn’t.
The metric that drove the decision
The figure the industry runs on is called containment: the share of calls the bot handles from start to finish without transferring to a human. A 60 percent containment rate on 1,000 calls means 600 callers never reach a person. On paper, the support team just got 60 percent smaller.
That number is what appeared on the quarterly slide. It’s what the vendor reports, and it’s a real reflection of what the bot did. The problem is that it only measures the bot. It doesn’t measure the work left behind.
What containment misses: the residue
The bot closed the easy calls: balance checks, card resets, payment queries averaging around two minutes each. What remained for human agents was the opposite, every call the bot couldn’t resolve. Support teams have a name for this: the residue.
Residue calls are slow. Before the bot, difficult calls were spread across a shift, balanced by quick ones. Once the easy calls were gone, every call a human agent took was a hard one. Those 400 remaining calls can consume more staff hours than the original 1,000 did.
There’s a compounding effect too. When the bot fails a caller, it doesn’t lose the contact, it transfers it. Some callers don’t wait in queue. They hang up and call back in an hour. One failed interaction becomes two or three.
What happened at the bank
According to the Finance Sector Union (which represents the affected workers and is an interested party), after the bot went live, call volumes climbed. Staff were offered overtime to cope. Team leaders were pulled off their own duties and put back on the phones. For roughly six weeks, the bot’s containment score looked strong while the floor was under pressure.
One data point was on the executive dashboard. The other wasn’t. There is no standard report called residue.
The union requested call volume data from the bank repeatedly. The bank declined to provide it. The union lodged a dispute with the Fair Work Commission, Australia’s workplace tribunal. A hearing was scheduled for August 25. Five days before it, on August 20, the bank reversed all 45 redundancies.
In its written statement, the bank said it had not adequately considered all relevant business considerations and that it should have been more thorough in assessing the roles required. Management acknowledged it hadn’t accounted for call volumes continuing to rise over several months. The bank kept the voice bot. It is still running.
For context: the Commonwealth Bank posted a record cash profit of A$10.25 billion in the same financial year. Forty-five roles was never a rounding error to the people in them.
The one number to ask for before any AI automation decision
If you’re evaluating an AI automation rollout that touches staffing decisions, containment isn’t the number that matters most.
Ask what happened to average handling time on the calls that still reach a person. If that figure went up, your team is doing harder work than before, and the headline metric will never surface it. That’s a four-minute question. It’s the one nobody asked for six weeks at Australia’s biggest bank.
Two smaller examples worth noting
The Commonwealth Bank case isn’t isolated. In April 2025, Cursor users started getting logged out when switching computers. Support emails told them the company had a one-device-per-subscription policy. There was no such policy. A front-line support bot had invented it, described it as a core security feature, and sent it to users. The co-founder publicly corrected it.
In 2024, Jake Moffatt asked Air Canada’s chatbot about bereavement fares before booking a flight to his grandmother’s funeral. The bot told him he could book at full price and claim the discount within 90 days. That wasn’t the policy. He applied for the refund and Air Canada refused, arguing in front of a Canadian tribunal that its chatbot was a separate legal entity responsible for its own statements. The tribunal rejected that argument. The ruling held that it makes no difference whether information comes from a static page or a chatbot: the company is responsible either way. Moffatt got his money back (case: Moffatt v. Air Canada, 2024 BCCRT 149).
The voice bot isn’t the problem in any of these stories. The problem is what gets measured, what gets trusted, and who gets authority to speak for the company.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
