Bloomberg reported July 28 that AI is doing what two years of "AI-adjacent" layoffs mostly weren't: actually replacing customer service jobs, at scale, with receipts to prove it. Microsoft's customer service and support headcount (employees and contractors combined) has dropped from roughly 50,000 to roughly 40,000. Judson Althoff, who runs Microsoft's sales and service operations, put a number on the savings: about $750 million a year, plus a plain example of what AI now closes without a human. "If something happened with little Johnny's Xbox in the middle of the night, we can now solve that with AI," he said. Uber cut about 10% of its customer service workforce. Hyatt is routing reservation changes and receipt requests to AI instead of agents, and its SVP of AI and data analytics, Pat Nestor, wasn't cagey about why: "Cutting costs is clearly a driver of an initiative like this."
That's a real, current, verifiable story: a useful corrective after two years of companies blaming AI for cuts that turned out to be management-layer reshuffles. But read past the headline number, because Bloomberg's own list of proof-point companies includes one that already ran this exact play and had to publicly reverse it.
Commonwealth Bank of Australia is on that list, credited with folding AI into chat support and banking real savings. What doesn't make it into this week's coverage: a year earlier, CBA announced it was cutting 45 customer service roles because a new AI "voice-bot" had reduced call volume enough that the jobs were redundant. Australia's Finance Sector Union checked that claim against what its own members were living through and called it an outright lie — call volumes weren't dropping, they were rising, and CBA was quietly covering the gap with staff overtime and team leaders pulled onto the phones. After the union pushed back, CBA apologized, admitted its assessment "did not adequately consider all relevant business considerations," and rehired the 45 people, giving them the choice to return to their roles, redeploy elsewhere in the bank, or leave.
Same company, same category of claim, about a year apart.
One time the number was fiction that management believed because a vendor sold it to them. This week's version may well be real. The point isn't that AI can't cut a call center's headcount; Microsoft's $750 million says it can. The point is that "AI reduced our call volume enough to cut heads" is a claim, not a fact, until someone outside the vendor's pitch deck checks it against what's actually happening on the floor — and CBA is proof that even a company this size will cut first and verify later if nobody makes it do the reverse.
Contrast that with 1-800Accountant, the small-business tax firm also cutting its outsourced support spend, by half, next year. It has automated the simple tasks (checking the status of a return, that kind of thing) and routes the genuinely complex remainder to a person. Chief strategy officer Ryan Teeples was blunt about where he thinks the pain lands: "I wouldn't want to be an investor in India or the Philippines right now, because that's where we are seeing the biggest cost savings from AI." But he isn't just banking the savings — he's putting them into hiring more accountants, largely in the US. Verify the automation actually works, then reinvest in a smaller, more senior, domestic bench for whatever judgment calls are left over. That's CBA's math, run in the right order.
If you're the CTO or VP of Ops looking at a vendor deck promising a 40% cut to your support desk, the diligence question isn't whether the AI performs in the demo. It's: show me the call-volume data from the last 90 days, not the projection. And budget the reinvestment hire (the senior, domestic, escalation-capable person your remaining volume will actually need) on the same slide as the cut, not twelve months later after a union files a grievance and forces the math to run backward. CBA already ran that experiment in public. You don't have to repeat it.
VC5 Consulting places the senior, domestic support and escalation hires that survive the AI cut — the ones companies like 1-800Accountant are quietly reinvesting savings into, instead of finding out the hard way that they cut too deep. If you're staffing that bench before your vendor's projection gets tested in production, let's talk.