Incident file · Production
Klarna's AI customer-service claims and partial walk-back (2024-2025)
cost-optimized rollout, CEO-admitted quality regression, partial reversal
The system
Klarna's OpenAI-powered customer-service assistant, announced in February 2024, handled customer support chats including billing disputes, fraud, and account-closure questions.
What was observed
Klarna announced the assistant handled 2.3 million chats in its first month, about two-thirds of volume, framed as "the equivalent work of 700 full-time agents," with an estimated $40M/year cost avoidance. Fifteen months later, the CEO described a quality problem.
Stop and think
- "2.3 million chats handled, two-thirds of volume automated" is a real number. What would you want to know alongside it before calling the rollout a success?
- Automation rate and cost avoidance both improve the more a system answers without escalating. What kind of query would that incentive structure quietly make worse?
- If you had to design one paired metric to track alongside automation rate from day one, what would it be?
Answer the questions above before you read on: what failed, why, and what would have stopped it.