The Death of the Call Center: How Conversational AI is Erasing 3 Million Global Jobs

There used to be a specific sound on the call center floor. Over the soft click of keyboards and the occasional supervisor strolling down the aisle, there was a muted hum of a hundred voices, slightly out of sync. The majority of that is still audible if you enter the same building in Quezon City or Gurgaon today, but it’s quieter and has empty rows that weren’t there two years ago. The headsets remain in place. More and more, the seats aren’t.

Approximately three million Americans work in call centers, and many times as many people do the same work worldwide—in office towers constructed especially for English-language voice work in Cairo, Pune’s business parks, and Manila’s high-rises. It’s not a glamorous job. According to McKinsey, stress and the monotonous repetition of “break/fix” calls cause about half of all customer service representatives to quit within a year. However, it has been one of the most dependable routes to the middle class in the developing world for the past 20 years. The AI hype videos do not include that portion of the story.

The chatbot isn’t actually changing right now. The majority of us have learned to type “agent” five times before rule-based bots give up. These bots have been around for years. The shift is the new generation of voice-capable, large-language-model-driven agents — Salesforce’s AgentForce, the systems Klarna deployed, the kind of voice AI that can hold a five-minute conversation about a missing parcel without sounding like it’s reading from a script. According to Joe Inzerillo, chief digital officer at Salesforce, 94% of consumers choose artificial intelligence when given the option. Although the direction is real, that number warrants some skepticism because it greatly depends on how the option is presented.

Everyone in the industry silently studies Klarna’s experience as a case study. In 2024, the Swedish buy-now-pay-later company saved money by using AI to replace 700 of its 3,000 customer service representatives. More intriguingly, it had to return some of them. Cases of identity theft were too disorganized. The model was not compatible with edge cases. Earlier this year, Klarna hired seven freelancers to handle the kind of complicated complaints AI couldn’t quite manage. That isn’t an AI failure story. It’s a story of an industry figuring out exactly which 80 percent of work it can hand off — Gartner’s number — and which 20 percent still needs a human voice on the line.

The harder question is what happens to the people. Anthropic CEO Dario Amodei, in a 20,000-word essay earlier this year, warned that AI could behave less like a tool and more like a “general labor substitute” — a phrase that landed badly because it described, fairly accurately, what’s happening in customer service. There’s a sense among industry analysts that the call center is the canary. If three million American jobs and many millions more abroad can be reorganised this way over five years, the same logic may walk into back-office finance, paralegal work, basic medical triage. Watching this unfold from outside the boardrooms, it’s hard not to feel a kind of unease about how little public conversation there has been about the workers themselves.

It’s worth being careful with the numbers, though. The dire predictions that the industry as a whole would disappear by the following year have not materialized and most likely won’t. The Evri bot that provided photographic “proof” of a package left at the incorrect door and the DPD chatbot that cursed at consumers are examples of everyday, unfinished automation. If businesses push too hard too quickly, they will lose clients. It’s likely that the collapse will be more gradual and ugly than what the headlines portray. However, it is taking place. Now, the question is not so much whether the headset workforce declines as it is whether those who wear them receive any real warning before it does.