Enterprise fintech has spent a decade trapped in an ugly operational reality: user growth demanded a linear increase in support payroll. Scaling transaction volume inevitably clogged multi-departmental escalation pipelines, ballooning headcount costs. Klarna, managing an active user base of over 85 million accounts, broke that linear curve not by deploying another scripted chatbot, but by rebuilding its Tier-1 operational plumbing on top of an agentic graph framework.
Agentic Workflows in Production
Klarna built its AI assistant on LangGraph, integrating LangSmith for live runtime observability. Unlike standard conversational bots that merely spit FAQ text, this architecture handles stateful, multi-step actions: running customer payments, calculating refunds, and managing disputes across 2.5 million conversations handled to date. The business impact is concrete: the system absorbed the workload equivalent of 700 full-time support agents.
Running autonomous workflows at this volume introduces severe compute liabilities. Klarna mitigates token burn by dynamically routing requests and scoping prompt contexts strictly to explicit transactional branches. This keeps latency down and inference bills predictable across high-frequency interactions.
"LangChain has been a great partner in helping us realize our vision for an AI-powered assistant, scaling support and delivering superior customer experiences across the globe."
As Klarna CEO Sebastian Siemiatkowski noted during the rollout, graph-based routing allows the organization to execute autonomous, deterministic logic without human intervention in the critical path.
Operational Performance and Evaluation
Transitioning from basic conversational wrappers to deterministic agent graphs requires rigorous evaluation to prevent transactional regressions. Klarna implemented a test-driven development workflow inside LangSmith, stepping through agent trajectories to catch edge-case hallucinations before pushing updates to production. The engineering team coupled LLM-as-a-judge evaluations with continuous prompt tuning to refine intent recognition across live payment flows.
Over nine months in production, the architecture slashed average query resolution time by 80%. For enterprise leadership, Klarna's deployment provides a clear benchmark: the economic payoff of agentic AI lies in modeling multi-departmental escalation paths into deterministic graphs, replacing manual ticket routing with verified autonomous execution.