What would you have done? Klarna replaced 700 agents with AI, then called humans back

In 2024, Klarna said its OpenAI-built assistant handled two-thirds of customer chats, the equivalent of 700 agents. A year later the company was rehiring humans: satisfaction was collapsing on complex cases, hidden by average metrics (handle time, aggregate CSAT) that measured speed, not resolution. Its CEO conceded that cost had become 'a too predominant evaluation factor', and what you end up with is lower quality.

The move was not an AI retreat but a redrawing of the line: AI on repeatable volume, humans on whatever carries the relationship. The lesson holds across sectors: AI rarely fails on the technology, it fails on scope. Before automating a customer process, draw the line between standard volume, handed to AI, and complex or high-value cases kept under human control, and track the latter with its own metric rather than lost in an average.

Source: Bloomberg, Sebastian Siemiatkowski interview, May 2025.