Klarna's AI Agent and the Customer-Service Job: What the Numbers Show

What Klarna announced in February 2024

On February 27, 2024, Klarna, which describes itself as "a global digital bank and flexible payments provider", said the AI assistant it had built with OpenAI had been live globally for one month and had held 2.3 million conversations, "two-thirds of Klarna's customer service chats".⁸ The assistant is a chatbot in the Klarna app, "powered by OpenAI", and the release said it handled "refunds, returns, payment-related issues, cancellations, disputes, and invoice inaccuracies".⁸

The headline claims followed: the assistant was "doing the equivalent work of 700 full-time agents", was "on par with human agents in regard to customer satisfaction score", produced "a 25% drop in repeat inquiries", let customers resolve errands "in less than 2 mins compared to 11 mins previously", worked in 23 markets and more than 35 languages, and was "estimated to drive a $40 million USD in profit improvement to Klarna in 2024".⁸ The release gave no method for any of them. It also said that "customers can still choose to interact with live agents if they'd prefer."⁸

What the IPO filing added: methods

Klarna's registration statement for its US share listing, in the amended version filed on September 8, 2025, repeated the claims with their bases attached.⁶ The assistant "handled 69% of customer service chats in the last twelve months ended June 30, 2025, according to our service chat log data", and had held 25.3 million conversations since launch.⁶ The 700-agent figure became "over 700 full-time agents", "estimated based on the average monthly reduction in chat and telephone conversations handled by full-time agents in 2024 following the launch of our AI assistant".⁶ The repeat-inquiry figure was dated: inquiries "dropped by 25% between December 2023 and January 2024".⁶ Satisfaction "on par with human agents" was attributed to "service chat log data and consumer satisfaction surveys", which elsewhere the filing calls "internal consumer satisfaction surveys".⁶

Two figures shifted between the documents. The launch release compared AI resolutions of under two minutes with "11 mins previously"; the filing compares two minutes with "the 12-minute average for human agents in 2024".⁸⁶ Those are different baselines (before launch, and human agents during 2024), and Klarna doesn't reconcile them. And the release's forecast of "$40 million USD in profit improvement" for 2024 appears in the filing as "approximately $39 million in cost savings" for 2024: a result rather than a forecast, measured as cost savings rather than profit, and still without a stated method.⁸⁶

What "700 agents" measures, and what it doesn't

The agent figure is a work-equivalence estimate: how many full-time agents' worth of chat and phone conversations stopped reaching people after launch, worked out from the fall in agent-handled volume. It is not a count of jobs cut, and the filings don't present it as one.⁶

Klarna's own headcount did fall over the same years. The F-1/A gives approximately 5,527 full-time employees at the end of 2022, 4,352 at the end of 2023, 3,422 at the end of 2024 and 3,098 at June 30, 2025; the annual report for 2025 gives approximately 2,831 at December 31, 2025.⁶² Klarna says the reduction "resulted from our strategic decision to reduce our overall headcount and drive operational efficiency by leveraging AI in our business", and that the workforce shrank "thanks to AI-driven efficiency gains and normal course employee attrition", that is, mainly by not replacing people who left.⁶ It adds: "We expect the number of employees to continue to decrease in future periods."²

Two cautions apply when setting the chart above beside the agent figure. The headcounts cover the whole company, and the filings quoted here give no separate customer-service headcount. And some of Klarna's service work is bought in: its customer service and operations expenses "primarily consist of personnel costs for customer support functions and outsourced assistance".⁶ So 700 agent-equivalents need not correspond to 700 Klarna employees.

The 2025 figures

Klarna's annual report for 2025, filed on February 26, 2026, moved every series up. The assistant "handled 80% of customer service chats in the year ended December 31, 2025", had held 31 million conversations since launch, did "the work equivalent of over 850 full-time agents" (the same method, applied to 2025) and "in 2025 delivered approximately $59 million in cost savings".² The report says there has been "no drop in consumer satisfaction levels since its introduction", again "according to internal consumer satisfaction surveys".²

Klarna's reported spending moves more slowly than the chat share. Customer service and operations expenses fell $37 million, or 15%, in 2024, which Klarna attributed to "a decrease in customer service costs".⁶ In 2025 they rose $4 million, or 2%, with "volumes up 32% year-over-year and transactions up 25% year-over-year".² At company level, Klarna said in November 2025 that "Since 2022, revenue per employee has tripled", reported $1.24 million of revenue per employee for 2025, "up 3.6 times since 2022 as headcount declined 49%", and "nearly $1.4 million" in the first quarter of 2026.⁹¹⁰¹¹ Its second-quarter 2026 release says operating expenses rose 16%, "supported by AI-enabled productivity gains and continued cost discipline", without restating the assistant's chat share or agent equivalence.¹²

Humans stay on the menu, by design

Klarna has kept a human route open since launch. Its filings describe a "dual-track approach of combining broad and continuing implementation of scalable AI in our customer service with high-quality human support", offered because "certain consumers may nevertheless prefer to interact with human representatives".⁶²

Speaking at SXSW London in June 2025, in answer to headlines that Klarna was looking to hire human workers again, chief executive Sebastian Siemiatkowski said: "We think offering human customer service is always going to be a VIP thing." He added: "We can use AI to automatically take away boring jobs, things that are manual work, but we are also going to promise our customers to have a human connection."¹³ He put the company's workforce at around 3,000, down from 5,500 two years earlier.¹³

The same tasks at other operators

Klarna's assistant covers part of the job O*NET describes. The scorecard above sets the seven core tasks O*NET 31.0 lists for customer service representatives against operator-confirmed deployments.¹ Insurer Lemonade's annual report for 2025 says its CX.AI bot platform is "built to understand and resolve customer requests without human intervention" and that "over half of Lemonade's customer inquiries are handled this way"; those requests range "from coverage questions to making changes to their policy, such as adding a spouse, updating coverage amounts, changing payment methods". Its onboarding bot, AI Maya, "handles everything from collecting information and personalizing coverage to creating quotes and facilitating payments securely".⁵ Its claims bot, AI Jim, took the first notice of loss (the customer's initial report of a claim) "without human intervention" 96% of the time, "roughly 55% of our claims were automated" as of December 31, 2025, and AI Jim "triages and assigns claims he is not authorized to settle" to human experts.⁵

Public agencies report similar systems. In the 2025 federal AI use case inventory, the Social Security Administration lists a deployed bot on its 800 number, noting that "Calls answered solely by the bot reduce the wait for other callers needing agent service", and the Federal Trade Commission lists a deployed tool that helps its consumer service representatives summarize calls.³

The one core task with no operator-confirmed case is checking afterwards that a fix took. Microsoft documents a Case Management Agent for Dynamics 365 that "Sends follow-up emails and resolves cases based on predefined rules", but that is a product description with no named operator, so it grades as a vendor claim.⁴

For people who stay in the job, the strongest independent evidence comes from a peer-reviewed field study of 5,172 customer-support agents given a generative AI assistant (AI that writes text, here suggested replies). "Access to AI assistance increases worker productivity, as measured by issues resolved per hour, by 15% on average," it found. "Less experienced and lower-skilled workers" improved in both speed and quality, while "the most experienced and highest-skilled workers see small gains in speed and small declines in quality."⁷

A different route to a smaller workforce

Payments company Block took another path. Its fourth-quarter 2025 shareholder letter, filed with the SEC on February 26, 2026, announced: "We're reducing Block by nearly half, from over 10,000 people to just under 6,000," on the reasoning that "Intelligence tools have changed what it means to build and run a company."¹⁴ That was one company-wide reduction, announced on one day, and the letter doesn't say how many of the roles were in customer support.¹⁴ Klarna's filings describe a decline spread over three years through efficiency gains and attrition.⁶²

What the numbers don't show

Every Klarna figure on this page is company-stated. The satisfaction comparisons rest on Klarna's internal surveys, and we found no independent audit of the chat share, the satisfaction scores or the agent-equivalence estimate.² The figures also say nothing about the occupation as a whole. For the US, O*NET, citing Bureau of Labor Statistics data, lists 2,666,000 customer service representatives employed in 2025, a projected change over 2025–2035 of "Decline (-1% or lower)", and 289,500 projected job openings for the same period. That projection is the Bureau's, and the O*NET page does not attribute it to AI.¹