Klarna's AI: 700 Agents' Work in 30 Days, Then a Twist

In February 2024, Klarna deployed an AI-powered customer service assistant built on OpenAI’s technology across all its markets simultaneously. The company serves 85 million active users across 23 countries. The assistant went live in 35 languages, with zero phased rollout. Day one, full deployment.

The results from the first month were published in an official press release on March 4, 2024, a document Klarna co-authored with OpenAI and made available on its international press page.

Month one, verified results

• 2.3 million conversations handled by AI, two thirds of all customer service contacts [Klarna official press release, March 2024]
• Equivalent workload of 700 full-time agents
• Average conversation time: 11 minutes → 2 minutes (−82%)
• Repeat inquiries for the same issue: −25%
• Customer satisfaction score: on par with human agents
• Projected annual savings: $40 million

The “700 agents” figure is a workload-equivalence calculation, made explicit in the official release. Klarna did not fire 700 people on launch day, most headcount reduction across 2023–2024 came through hiring freezes and natural attrition.

The original idea: full deployment, zero tiers

What made Klarna’s approach distinctive was the decision to deploy AI as the primary contact layer for all customer types, across all markets, from day one. The conventional approach is a tiered rollout: start with one market, one language, low-complexity cases. Klarna compressed all of that into a single launch.

The company had built its AI assistant on OpenAI technology and tested it internally for months. When it launched externally, the scale was immediate: 23 markets, 35 languages, two thirds of all contacts routed to AI before a human was ever involved.

The correction, and why it matters more than the launch

By Q3 2025, Klarna reported the AI assistant doing the equivalent work of 853 agents and generating approximately $60 million in annual savings, up from $40 million at launch.

And yet in May 2025, CEO Sebastian Siemiatkowski publicly acknowledged that Klarna had “over-rotated on AI” and was reintroducing human agents for premium and complex customer cases. His framing: “investing in the quality of human support is a wise way forward” for higher-value tiers.

This was a scope correction, not a reversal. Klarna was redefining the boundary, not retreating from AI. The high-volume, standardized tier remained fully automated. The premium and complex tier added human expertise back in.

By September 2025, Klarna went public at a $15 billion valuation, with shares opening at $52 on the first trading day.

What you can take from the full story

The launch story is well known. The correction story is more useful.

Klarna’s experience shows that AI customer service at scale works, measurably, financially, and in customer satisfaction terms. It also shows that the right configuration of AI and human involvement is something you discover through deployment, and that the willingness to adjust that configuration is a sign of operational maturity, not failure.

The question for any organization considering AI in customer-facing roles is the same one Klarna answered through 18 months of operation: where does AI create value, and where does the presence of a human create more of it?