Klarna's AI Assistant and the "Work of 700 Agents" — Replacing Customer Service, Then Bringing People Back

On 27 February 2024 Klarna, the Swedish buy-now-pay-later lender and digital bank, reported the first month of its AI assistant built with OpenAI technology. The assistant, Klarna said, had handled 2.3 million conversations, two-thirds of its customer service chats, and was doing the equivalent work of 700 full-time agents. The "700 agents" became one of the most quoted examples of generative AI replacing human work.

The story did not end there. Klarna's headcount fell steeply afterwards, but in May 2025 its chief executive, Sebastian Siemiatkowski, admitted that cost had been weighted too heavily and quality had suffered, and said the company would invest more in human support. Klarna is the best-known case of AI replacing customer service, and also a case that showed the limits.

The February 2024 announcement

The figures Klarna reported, all of them the company's own:

The release added that customers could still choose to deal with human agents. In the same release Siemiatkowski called the result good for customers, employees and investors, but said it also "underscores the profound impact on society that AI will have", and urged society and politicians to consider the change carefully.

Who was affected

Markets reacted first. On 28 February, the day after the announcement, shares in Teleperformance, the French call-centre company, fell as much as 29% in Paris trading, the steepest drop since November 2022, Bloomberg reported, because Klarna's statement rekindled concern that AI would hurt the call-centre business. That an outsourcer was sold off showed the market expected the effects to reach not only the announcing company's staff but the workers of firms that handle customer service under contract.

Inside Klarna, headcount shrank. According to the registration statement Klarna filed with the US Securities and Exchange Commission in August 2025, its full-time employees numbered:

The filing attributed the fall to a strategic decision to reduce overall headcount and drive operational efficiency by leveraging AI, and said it expected the number of employees to keep falling. Reports said Klarna froze hiring in December 2023 and shrank mainly through natural attrition. Its third-quarter 2025 results release said revenue per employee had tripled since 2022.

How the 700 was measured

The SEC filing gives more detail. The assistant handled 69% of customer service chats in the twelve months to 30 June 2025, doing the work equivalent of over 700 full-time agents. That figure was estimated from the average monthly reduction in chat and telephone conversations handled by full-time agents in 2024 after the assistant launched. Cost savings in 2024 came to about $39 million, close to the $40 million estimated at launch. AI-handled chats were resolved in two minutes on average, against 12 minutes for human agents in 2024.

The 700, in other words, was not 700 people dismissed. It was the fall in human-handled conversations converted into a headcount, and it was not independently verified.

The 2025 correction

In May 2025 Siemiatkowski told Bloomberg the approach was being corrected. As reported by Computing, he said: "Cost, unfortunately, seems to have been a too predominant evaluation factor when organising this. What you end up having is lower quality." He added: "Really investing in the quality of the human support is the way of the future for us." Klarna was reported to be recruiting fully remote customer service staff, targeting students, people in rural areas and dedicated Klarna users.

This was not a retreat from AI. Klarna said it was still reducing staff through natural attrition and was hiring the new agents on a freelance basis for its outsourcing division. Customers are routed to the AI assistant first and escalated to a person when needed. The question became not "AI or humans" but where to keep people once AI sits in front.

What research suggests

Customer support was also one of the first jobs where generative AI's effect was measured. Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied the staggered introduction of a generative-AI conversational assistant to 5,179 customer support agents at one company. Issues resolved per hour rose 14% on average and 34% for novice and low-skilled workers, with minimal effect on experienced and highly skilled agents. AI used to support people rather than replace them also raised productivity.

Klarna's case showed that handing more to AI lowers costs, but that at some point the balance with quality has to be faced. Research using real data on young workers' employment is collected on the "canaries in the coal mine" page, and the wider sequence of events is on the AI and society timeline.