In late July 2026, Bloomberg reported something support teams had been bracing for since ChatGPT first shipped: AI is now cutting real customer service jobs, at real scale, at real companies. Commonwealth Bank of Australia, Microsoft, Uber, and Hyatt Hotels are among the names now using automated chat and voice systems to do work that used to require a person on the other end of the line — and Salesforce's own 2026 research helps explain why: AI agent adoption in customer service nearly doubled in a single year.
But buried in the same news cycle is a quieter, more useful story: some of the same companies making these cuts have also had to walk them back, hire people right back, or shift headcount rather than eliminate it — because the AI didn't actually reduce the volume of calls it was supposed to replace. If you're the one deciding how much of your support team to automate this year, that's the part worth reading closely.
The layoffs are real — here's what's actually happening
The scale of the shift by mid-2026 is no longer speculative. Microsoft has trimmed its customer service workforce — a mix of contractors and full-time staff — from about 50,000 to roughly 40,000 in recent years as it leans on its own AI tools internally, according to Bloomberg. Judson Althoff, who runs Microsoft's sales and service organization, said earlier this year that AI is now saving the company around $750 million annually in customer service costs. Uber has cut about 10% of the jobs in its customer service operation as part of a push to route more support requests through an in-app AI chatbot, Bloomberg also reported. Hyatt is using a third-party AI vendor, Sierra, to automate simpler requests like reservation changes and receipts, and Brink's Home Security says AI cut its inbound call volume by roughly two-thirds, letting it shrink its call center staff from about 800 to 400 — mostly through attrition and internal transfers rather than layoffs.
Commonwealth Bank of Australia (CBA) is the most-cited example, and the most complicated one. Bloomberg reported that CBA has shed hundreds of positions from its chat support line, largely contractor roles at a call center outside Johannesburg run by the outsourcing firm Nutun, saving the bank tens of millions of dollars a year. At the same time, a CBA spokesperson said the bank has added more than 140 roles at its own Australia-based call centers over the past six months — a reminder that "AI is cutting jobs" and "AI is shrinking the total support workforce" aren't always the same claim.
The pressure isn't limited to in-house teams. Business process outsourcers like Teleperformance, Concentrix, and TTEC — the companies that staff a huge share of the world's tier-one support desks — have seen their share prices drop sharply as clients bring more of that volume in-house or automate it away. Forrester analyst Kate Leggett estimated earlier this year that close to half of all customer service roles could be affected by AI by 2030. None of this is a distant forecast anymore; it's a live restructuring.
The data behind the shift
Salesforce's State of Service: AI Agents Edition report, based on a double-anonymous survey of 3,075 service professionals worldwide conducted in March and April 2026, put a number on what's driving the layoffs: agentic AI adoption in customer service organizations rose from 39% to 66% in a single year — a 1.7x increase. Seventy percent of the organizations that deployed AI agents said they saw measurable value within 60 days, and 89% said their organization would benefit from expanding AI use further.
The more interesting finding, though, is what actually improved. Salesforce asked respondents which KPI moved the most after deploying AI agents, and the top answer wasn't cost per contact or average handle time — it was customer satisfaction. That matters, because it suggests the companies getting real value from AI support aren't just the ones cutting the most heads; they're the ones whose customers are having noticeably better experiences. The same report found that 97% of service leaders with AI say it's changing how they plan their workforce — and 72% of the people actually running service operations (versus 59% of leaders one level up) say data readiness, not the AI model itself, is the biggest blocker to doing this well.
When "replace first, ask questions later" backfires
A year earlier, Commonwealth Bank offered a preview of exactly what goes wrong when a company skips that data-readiness step. In July 2025, CBA cut 45 customer service roles after introducing an AI-powered voice-bot, telling staff the technology would reduce call volume. By August, the bank had reversed the decision entirely, apologizing to the affected employees and calling the redundancies an "error." According to the Finance Sector Union, call volumes didn't fall after the bot launched — they rose, with managers pulling team leaders back onto the phones to keep up. CBA itself admitted it "did not adequately consider all relevant business considerations" before making the cuts.
The lesson isn't that AI support doesn't work — Salesforce's survey of thousands of other teams, and CBA's own decision to keep investing in the technology afterward, say otherwise. It's that a bot which can't actually resolve the full range of real customer requests doesn't lower contact volume; it just adds a frustrating extra step before a human gets involved anyway, and by then you may have already cut the humans who were supposed to catch that overflow. That trap shows up especially fast in markets where the AI's language coverage doesn't match how customers actually communicate — a bot fluent in Modern Standard Arabic but not the Gulf or Levantine dialects people actually type and speak will get escalated, not resolved, and the "savings" evaporate the same way CBA's did.
The smarter playbook: automate the volume, not the team
The businesses quoted in Bloomberg's reporting that seem to be avoiding CBA's mistake share a pattern: they're automating a defined slice of tier-one demand and moving people to higher-value work, rather than making a single all-or-nothing cutover. A few principles worth borrowing:
Automate tier-one volume first, and prove it before you cut headcount. Order status, appointment changes, FAQs, and simple account questions are the requests every team above is targeting first — not the complex, high-stakes cases that still need a person. Measure actual call and chat volume for at least a full cycle after launch, the way CBA didn't, before touching headcount.
Keep a human-AI co-pilot in the loop, not a hard cutover. Instead of routing customers into a bot that either resolves the issue or dead-ends them, a model where AI drafts responses and handles routine cases while a human agent reviews, intervenes, or takes escalations in real time avoids the CBA scenario almost by design — the AI can't quietly fail a customer into a phone queue nobody's watching.
Cover every channel your customers actually use. A bot that only lives in one chat widget pushes overflow into whichever channel isn't automated — usually the phone line, which is the most expensive one to staff. Support that works consistently across voice, WhatsApp, Instagram, web chat, and in-store kiosks avoids that channel-shifting problem entirely.
Make sure the AI understands your customers' actual language, not a textbook version of it. This is the same failure mode as CBA's, just linguistic instead of procedural: if a meaningful share of contacts come in Gulf or Levantine Arabic dialect rather than Modern Standard Arabic, and the AI only handles the latter, expect the same "volume didn't actually drop" outcome.
Plan for new roles, not just fewer old ones. Salesforce's data on workforce planning suggests the teams doing this well are creating roles around AI oversight, prompt/knowledge-base quality, and escalation handling — not simply subtracting headcount and hoping the AI holds the line unsupervised.
Full replacement vs. human-AI co-pilot: a quick comparison
| Full-replacement automation | Human-AI co-pilot model | |
|---|---|---|
| Typical rollout | Bot handles the full conversation end-to-end; humans only get what the bot can't route at all | AI drafts and resolves routine cases; humans review, intervene, and own escalations in real time |
| Risk if the AI misjudges a request | Customer hits a dead end or repeats themselves — volume doesn't actually drop (CBA, 2025) | Human catches it mid-conversation before the customer feels stuck |
| Headcount impact | Fast, often front-loaded cuts before results are proven | Headcount shifts toward oversight, QA, and complex cases as automation proves itself |
| Best fit | High-volume, narrowly scoped requests with little language/dialect variation | Businesses with mixed channels, mixed languages/dialects, or higher-stakes conversations |
| Customer experience risk | Higher if language, dialect, or edge-case coverage is incomplete | Lower — a person is always one step away in the same conversation |
What this means for Gulf and MENA support teams
The dynamics above apply everywhere, but they bite harder in Gulf and wider MENA markets for a specific reason: a large share of everyday customer contact already happens on WhatsApp, Instagram DMs, and voice calls rather than a single website chat widget, and a meaningful share of it happens in dialect rather than Modern Standard Arabic. A support setup that automates only one of those channels, or only understands textbook Arabic, will reproduce CBA's problem in miniature — the AI looks efficient in a dashboard while customers quietly get routed back to a human anyway, just later and more frustrated. Platforms built specifically to run voice, WhatsApp, Instagram, Messenger, web chat, and in-store kiosk support from one place — with a human agent able to step into any of those conversations mid-flow — are a direct answer to that gap; it's the model Qalyb is built around, and it's the same shape of solution the least-burned companies in Bloomberg's reporting are converging on independently.
The bottom line
AI genuinely is cutting customer service jobs in 2026 — that part of the story is no longer up for debate, and Salesforce's own numbers show why: adoption nearly doubled in a year, and most teams that deploy it see real value fast. But the companies handling this transition well aren't the ones making the fastest cuts; they're the ones automating a proven slice of demand, keeping a human one step away from every conversation, and covering the channels and languages their customers actually use before they touch headcount. CBA's reversal is a year old now, but it's aged into exactly the cautionary tale it looked like at the time: measure the volume before you cut the team, not after.