AI BriefingHow Klarna's AI Assistant Redefined Large-Scale Customer Support for 85 Million Active Users
Key point
Klarna leveraged LangGraph and LangSmith to build a high-performance AI agent that replaces the work of 700 employees.
Details
Fintech leader Klarna introduced an AI assistant built on LangGraph and LangSmith to handle 85 million active users and 2.5 million daily transactions. Going beyond a simple chatbot, this system operates as an Agent that handles payments, refunds, and escalations, processing a workload equivalent to that of 700 full-time employees and maximizing operational efficiency.
To achieve precise performance, Klarna adopted the following technical approaches.
- Controllable agent architecture: Using LangGraph to route requests and separate tasks, reducing latency and increasing reliability.
- Context-aware intelligence: Applying customized prompts per scenario to reduce token costs and provide context-appropriate responses.
- Test-driven development: Using LangSmith to monitor agent behavior step by step, and continuously verifying and improving performance through LLM evaluation.
As a result of the rollout, over the past 9 months Klarna reduced customer inquiry resolution time by 80% and achieved automation of about 70% of repetitive support tasks. It also improved its ability to identify root causes behind rejections, significantly lowering the rate of customer escalations.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
LangGraph Agent Server Vulnerability Allows Approval Bypass via State Mutation