How Klarna’s AI Assistant Resolves 80% of Queries in Under 2 Minutes

Klarna sits at the intersection of commerce and consumer finance, operating one of the world’s most widely used buy-now-pay-later platforms. With over 85 million active users and 2.5 million daily transactions, the company faces a customer support challenge that scales with its business: payments, refunds, and escalations require fast, accurate responses across multiple languages and time zones. Traditional support scaling meant adding headcount, which introduced costs and quality inconsistencies.

The pressure point was escalation management. As Klarna’s user base grew, multi-departmental escalations became harder to route and resolve. Customers arriving with complex payment disputes or refund issues often waited while queries moved through internal chains before reaching the right team. The friction was measurable, and the costs of unresolved escalations compounded across millions of interactions.

Klarna built its AI assistant on LangGraph, using a controllable agent architecture to route requests and handle different task types across a structured multi-agent system. LangSmith provided the observability layer: the team could see exactly how the AI behaved step by step, run experiments against multiple prompts and model configurations, and measure outcomes with custom evaluation metrics. A key collaboration came through prompt optimization — Klarna’s insights directly shaped LangSmith’s meta-prompting feature, which allows users to suggest specific improvements and measure their impact on response quality.

Over nine months, the results shifted the support operation at scale. Average customer query resolution time dropped by 80%. Approximately 70% of repetitive support tasks were automated, freeing human agents to focus on complex, high-value interactions requiring judgment and empathy. With 2.5 million conversations logged, the AI assistant delivers the equivalent output of 700 full-time employees, fundamentally changing the economics of Klarna’s support function.

The broader implication for fintech and consumer platforms is that AI can meet the dual demand of scale and quality in customer support. Klarna’s model — pairing LangGraph’s structured agent architecture with LangSmith’s rigorous evaluation and iteration cycle — demonstrates that automating support at this scale requires not just a capable model, but a disciplined, testable approach to deployment and refinement.