Klarna's AI Assistant Handles Two-Thirds of Customer Service Chats in Month One — Then Course-Corrects Back to Humans
Klarna's OpenAI-powered assistant handled 2.3 million conversations in its first month, doing the equivalent work of 700 full-time agents, cutting resolution time from 11 minutes to under 2 minutes, and projecting $40M in 2024 profit improvement — but by May 2025 the company reversed course and began rehiring human agents for complex cases.
Background
Klarna serves 150 million consumers across 23 markets handling billions of payments annually. Customer service at this scale — across languages, time zones, and a wide range of inquiry types — is a core operational challenge. The AI assistant was designed to handle the long tail of routine inquiries (refunds, returns, payment schedules, disputes) that previously required human time. The partnership with OpenAI was Klarna's flagship AI initiative in 2024.
What Was Implemented
- OpenAI-powered AI assistant deployed globally in the Klarna app, February 2024
- Scope: refunds, returns, payment-related issues, cancellations, disputes, invoice inaccuracies
- Personal financial assistant features: outstanding balance updates, payment schedule reminders, spending limit explanations
- Multilingual support: 35+ languages across 23 markets, available 24/7
- Human agent option retained: customers could still request live agents
Results
First-month confirmed results (February 2024 press release): - 2.3 million conversations handled - Two-thirds (approximately 67%) of all Klarna customer service chats - Equivalent work of 700 full-time agents - Customer satisfaction on par with human agents - 25% reduction in repeat inquiries (more accurate resolution) - Resolution time: under 2 minutes (from 11 minutes previously) - Projected $40 million profit improvement in 2024 Course correction (May 2025): Klarna began rehiring human agents after CEO acknowledged AI produced "lower quality" for complex cases. Full AI automation was walked back; human availability guaranteed on request.
Lessons
- AI assistants at scale excel for high-volume, well-defined inquiry types; complex disputes, fraud, and hardship cases require human judgment and carry compliance risk in regulated industries
- The $40M projected savings was primarily avoided hiring cost (700 FTE equivalent), not headcount reduction — an important distinction for how AI economics are communicated
- Building in a guaranteed human-access option from the start is both good practice and customer trust management
- Early metrics (one-month snapshot) may overstate long-term performance; longitudinal tracking and quality monitoring are essential before declaring full deployment success
- Course correction is normal and does not negate the value AI added for routine inquiry handling; the hybrid model that emerged may be more durable than full automation