Klarna Pulls Staff From Engineering, Legal, and Marketing Onto Support Calls as Its AI Strategy Rebalances
The most cited AI support deployment in tech has become its most cited cautionary tale, and the real lesson is about where the human-AI boundary belongs.
If this caught your attention, that’s not accidental.
The best editorial systems don’t happen by accident. Outlever builds them.
The story arc is complete, and every CX leader should study it. Klarna's AI assistant, launched with OpenAI in February 2024, handled 2.3 million conversations in its first month and cut average resolution time from 11 minutes to under two. The company claimed the system did the work of 700 agents, froze support hiring, and became the reference deployment for AI-run customer service.
Then came the correction. By May 2025 the CEO was publicly conceding that quality had suffered and that the company would hire human agents again. By late 2025, reporting showed Klarna redeploying employees from marketing, engineering, legal, operations, and analytics onto customer service, rebuilding capacity it had eliminated a year earlier.
What actually broke
The instinct is to read this as an AI failure, and the record does not support that reading. The system genuinely absorbed enormous tier-one volume at speed and savings the company estimated around 60 million dollars. What broke was the boundary. Klarna treated automation as a one-time headcount cut instead of a design decision about which conversations belong to software and which belong to people. Complex cases degraded, institutional knowledge walked out the door, and customers with real problems met a system optimized for deflection.
The distinction matters because the deflection model and the resolution model look identical in a board deck and behave nothing alike in production. Deflection counts conversations that never reached a human. Resolution counts problems that actually ended. A support AI built to deflect will happily generate the first number while the second collapses.
The rebalancing is industry-wide
Klarna is the loudest example, not the only one. In an Orgvue survey cited in coverage of the reversal, 55 percent of leaders who cut jobs citing AI now say it was the wrong call, and a Gartner study of 350 firms this spring found the hardest headcount cutters showed no better returns than companies that held steady.
The model that is emerging from the wreckage is neither the old call center nor the all-bot fantasy. Autonomous systems own the volume tier end to end, and they own it well, with real answers rather than ticket triage. Humans move up the stack to the conversations where empathy, judgment, and relationship carry the outcome. Companies that draw that line deliberately get the economics and the experience. Companies that draw it with a layoff announcement get to run the experiment Klarna already ran, at their own expense.
Read what the room is reading.
New pieces for growth leaders, delivered as they publish. Unsubscribe whenever.
