How Klarna’s AI Agent Strategy Backfired But Became A Useful Lesson

Klarna’s AI customer service agents handled millions of conversations, cut resolution times and promised major financial gains, yet the company eventually admitted it had reduced its human workforce too aggressively.

Adobe Stock

Klarna’s AI customer service rollout looked like a textbook success. Its autonomous agent handled 2.3 million conversations in its first month, helped resolve two-thirds of all customer service tickets and cut average resolution times from 11 minutes to just two.

The buy now, pay later giant later estimated that the technology could add $40 million to its annual profit. As AI took on more work, Klarna froze hiring and allowed its customer service workforce to fall from around 5,000 people to 3,500.

Then the warning signs appeared. A little over a year later, CEO Sebastian Siemiatkowski acknowledged that Klarna had cut too far, too quickly and lost valuable human expertise. The company began rebuilding its capacity for human support.

Klarna is not alone. Ford has also acknowledged that it was overly optimistic about how quickly AI could enable workforce reductions, suggesting that other businesses may be making similar miscalculations.

This was not a straightforward AI failure. Klarna’s agents delivered impressive results, but the company underestimated the importance of experienced people when customers faced complex, ambiguous or emotionally charged problems. Its experience offers an important lesson for every business racing to replace human work with AI.

What Do Klarna’s Customer Service Agents Do?

Klarna was an early partner of ChatGPT creator OpenAI and moved quickly to integrate natural language into its customer service. But rather than a simple chatbot, it built the offering around agentic AI.

This means it doesn’t just answer questions and generate information. It can work on complex, multi-stage tasks with minimal human involvement and interact with other systems.

In Klarna’s case, this means securely accessing customer data, monitoring the changing state of a ticket, issuing returns and managing payment plans.

These are exactly the types of tasks that agentic AI is good at: high-volume and extremely repetitive. Decisions follow straightforward logic, and clear guidelines can dictate when human intervention should happen.

The results were great. Driven by the headline figures mentioned above and a 25 percent reduction in repeat requests, Klarna estimated that the deployment added $40 million to its annual profit.

So what went wrong, and why was the company forced to perform a U-turn in 2025 and fill many of the vacancies it had allowed to remain empty?

Human Intervention And Edge Cases

Although the headline statistics around problems solved and customers satisfied were high, Klarna couldn’t ignore the fact that some customers were very vocal about their own lack of satisfaction.

For some more complex customer issues, agents just weren’t up to scratch. Specific details here aren't well documented, but given the strengths and weaknesses of agents, these are likely to be tasks involving ambiguity in communication, emotional responses, or edge-case technical issues.

In other words, problems that typically still need an experienced human operator to get involved. With the company recently having reduced customer service headcount by around 30 percent, fewer of these people were available to jump in, leading to poorer outcomes.

This meant reduced human capacity for understanding nuanced situations, acting with empathy, and applying considered discretion. Satisfaction scores for these issues fell.

Siemiatkowski admitted the company had been too aggressive with cuts and had identified a strategic need to reinvest in well-trained human support workers, capable of focusing on high-value support issues.

The U-turn was widely reported on as an "AI failure," but this overlooks an important fact. By most definitions, Klarna’s customer service AI rollout was a big success. However, by overlooking niche situations and problems where AI is less capable, its results were sub-optimal.

What Can Leaders And Professionals Learn?

The important lesson here is that AI is not a magic bullet. Even when the benefits are clear, we need to think about edge cases and where humans still need to be involved.

By identifying its failure and acting to resolve it, even when it meant walking back previous decisions, Klarna learned to adapt its strategy as its understanding increased.

Agents are built for routine, repetitive tasks where they can be given clear instructions and defined goals. But business in the real world is far too complex, nuanced and subjective for today’s best AIs to reliably deal with.

The reason agents failed in Klarna’s customer service edge cases is the same reason autonomous driving still isn’t commonplace for most of us, despite it being theoretically possible.

It also highlights that humans are a critical element of AI infrastructure, even at the scale at which it's operated by Klarna.

Businesses have to chase efficiency and sometimes this means reducing headcounts. But when it leads to an exodus of skill and talent from the building before there’s total confidence in whatever they’re being replaced with, it’s likely to backfire.

This means thinking carefully about what human skills and experience need to be retained in an organization, while also developing the capabilities it will need in the future.