Swedish fintech Klarna, once one of the most enthusiastic adopters of artificial intelligence in customer support, is reintroducing human agents into its operations — a move that has caught the attention of the customer service industry worldwide. Klarna had previously claimed that its AI assistant was handling tasks equivalent to 700 full-time agents, dramatically cutting […]
Swedish fintech Klarna, once one of the most enthusiastic adopters of artificial intelligence in customer support, is reintroducing human agents into its operations — a move that has caught the attention of the customer service industry worldwide.
Klarna had previously claimed that its AI assistant was handling tasks equivalent to 700 full-time agents, dramatically cutting average resolution times from 11 minutes to just two.
This drive toward automation also helped reduce the company’s global headcount by over 1,000 employees. But the company is now quietly recalibrating, embracing a more hybrid approach that blends automation with human intervention.
Earlier this month, the ‘buy now pay later’ fintech’s CEO, Sebastian Siemiatkowski, admitted that Klarna is, once again, hiring new staff because of disappointment with the performance of the AI tools the company had rolled out.
“Cost, unfortunately, seems to have been too predominant an evaluation factor when organising this. What you end up having is lower quality,” he told Bloomberg in an interview.
In a turnaround from his earlier attitude, he added: “Really investing in the quality of the human support is the way of the future for us.”
Nevertheless, the company insists that it is still following its policy of driving down staff numbers via natural attrition and was only hiring new customer service agents on a freelance basis for the company’s outsourcing division.
According to Siemiatkowski, the aim is to channel customers through the AI agent first, but with the ability to escalate to a real, live human if necessary.
Strategic reset
For Liam Dunne, CEO of customer service analytics firm Klearcom, this moment marks a necessary course correction, not a failure.
“This isn’t a U-turn,” Dunne says. “It’s a strategic reset. Klarna is slowing down to go faster. They’ve learned what works — and what doesn’t — when it comes to real-world customer experience.”
Liam Dunne, CEO, Klearcom
Dunne, whose company analyses millions of calls for some of the world’s biggest brands including Google, Mastercard and Visa, believes that most firms are only at the beginning of their AI journeys.
“We’re still in the early days of AI adoption in customer service. Success doesn’t just depend on input and training data. It’s about output. It’s about delivering conversations that feel natural and solve problems. There’s a real difference between what AI promises and what it performs,” he says.
Dunne draws a comparison with previous industrial shifts — from manual labour to machinery, and from filing cabinets to computers. “Those revolutions didn’t happen overnight. This won’t either. What matters now is being measured, not just being first.”
Klarna’s adjustment also points to a growing recognition that human agents still have a valuable role to play — particularly in high-stakes scenarios such as fraud, compliance, and sensitive issues.
Dunne argues that rather than displacing agents, AI should be seen to redeploy them to more meaningful work.
“It’s about elevating the role of the human, not erasing it,” he says. “Let AI handle the repetitive stuff. But when something’s gone wrong — when a customer is upset or confused — that’s where people need to step in. And they need to be trained, empowered, and trusted.”
Adding to the pressure is the evolving regulatory landscape. While the European Union’s Artificial Intelligence Act (AI Act) does not explicitly grant individuals a general “right to talk to a human”, it incorporates provisions that emphasise human oversight and transparency in AI decision-making processes.
In its latest CX report Gartner predicts that within three years, the EU will mandate “the right to talk to a human” in customer service.
Dunne adds: “Firms are being pushed — rightly — to build in a ‘right to a human.’ And so what we’re seeing is a trend toward smarter integration, not full automation.”
Agentic AI
Looking to the future, Dunne sees even more advanced developments on the horizon, including agentic AI — intelligent systems designed to carry out specific tasks and interact across workflows.
In time, he says, AI agents will begin to communicate directly with other AIs (“similar to how search engines were able to read websites in an earlier evolution”), performing end-to-end tasks across platforms like Salesforce or HubSpot.
“Eventually, my AI will talk to your AI, agreeing steps, validating data, and completing actions behind the scenes – it might be a contract that needs renewing, for instance. But we’re five to eight years away from that being commonplace,” he predicts.
For now, however, the key takeaway is patience. AI’s promise is enormous, but real-world deployment — especially in customer service — requires care, iteration, and a strong understanding of human needs.
“This is a marathon, not a sprint,” Dunne concludes. “The companies that take their time, test, adapt, and listen to customers will be the ones who come out on top.
