Few companies have been cited as often in the AI-and-jobs debate as Klarna, the Swedish fintech. In February 2024, the company said its AI assistant was handling 75% of customer service chats — about 2.3 million conversations in its first month — and that it was doing the equivalent work of 700 full-time agents. CEO Sebastian Siemiatkowski talked openly about shrinking the company's workforce from roughly 5,000 to 2,000 through natural attrition, largely on the strength of that automation. It became a go-to example for how fast AI could hollow out support roles.
By 2025, Klarna's bot was still handling a large share of chats — around two-thirds, by the company's own account — and resolving them faster than humans: about two minutes on average, versus eleven minutes for a human agent. But speed wasn't the whole picture. In a Bloomberg interview, Siemiatkowski acknowledged that the AI-first approach had produced "lower quality" service in practice, and said the company was now actively hiring human customer service agents again. He added that Klarna would make sure customers could always reach a human if they wanted one — a notable walk-back from the original framing of AI as a near-total replacement.
The pattern that emerges from Klarna's experience is a familiar one in automation generally: AI handles high-volume, repetitive, well-defined queries extremely well — password resets, order status, simple refund rules — and struggles with the smaller share of cases that require judgment, empathy, or handling an upset customer who needs to feel heard rather than processed quickly. Resolving a chat in two minutes isn't a win if the resolution is wrong, or if the customer leaves frustrated. Klarna's reversal suggests that even a company with the engineering resources and incentive to make AI-first support work ran into that ceiling.
The lesson isn't that customer service jobs are safe from AI — clearly, a large share of ticket volume genuinely has moved to automation, and that's not reversing. The lesson is that the roles left behind skew toward exactly the things AI is currently bad at: de-escalation, judgment calls outside a script, and cases where getting it right matters more than getting it fast. If you're in a support career track, the more defensible position over the next few years is specializing in complex case handling, quality oversight of AI-generated responses, or escalation management — not competing with the bot on volume of simple tickets, which is a race you can't win and increasingly don't need to.
Klarna's experience is a useful caution against treating AI substitution targets as a simple headcount math problem. "AI can do 70% of the volume" and "AI can replace 70% of the team" are not the same claim, and the gap between them is where companies that move fast on AI-driven restructuring are most likely to get burned.
More on Klarna's reversal via Entrepreneur's report on the CEO's comments.