Klarna Uses LangGraph to Automate 70% of Customer Support

Klarna has successfully scaled its flagship AI Assistant to manage customer support for its 85 million active users, conducting 2.5 million conversations to date. Built using the LangGraph framework and refined with LangSmith, the system handles complex tasks such as payments, refunds, and escalations. According to the company, the AI agent performs work equivalent to 700 full-time customer service representatives, significantly boosting operational efficiency across its platform, which processes 2.5 million daily transactions.

Over a nine-month period, the integration of LangGraph and LangSmith allowed Klarna to automate approximately 70 percent of its repetitive support tasks. This automation has freed human customer service agents to focus on more complex, high-value interactions. Additionally, the fintech company achieved an 80 percent reduction in average customer query resolution times, saving its engineering and analysis teams hours of manual investigation each week. The system also improved root cause identification for transaction rejections, which directly lowered the rate of customer escalations.

For AI practitioners, Klarna's deployment highlights the practical value of controllable agent architectures and test-driven development. By using LangGraph, Klarna routed user requests dynamically to specific agents, which lowered latency and reduced token costs. The company utilized LangSmith to run LLM evaluations and iterate on prompts. This partnership even led to the design of advanced meta-prompting features within LangSmith, allowing developers to optimize prompts by suggesting improvements and directly observing their impact on response quality. This case study proves that multi-agent systems can reliably handle high-volume, multi-departmental escalations in highly regulated industries.