AI customer service at scale, done right
The problem
Klarna wanted to cut the cost and wait times of customer support across a very large, multilingual customer base.
The AI approach
An OpenAI-powered assistant handled end-to-end support conversations — answering questions, looking up accounts, processing refunds — across many languages.
Evidence it works
Klarna reported the assistant handled 2.3M chats in its first month (~75% of volume) in 35+ languages, doing work it equated to ~700 agents. In May 2025 the CEO refined the approach, conceding a cost-first push had reduced quality and beginning to rebalance toward human agents for complex cases — citing satisfaction on complex queries and edge-case accuracy in a regulated context. Reporting also clarified the “700 agents” framing referred to avoided hiring, not layoffs.
What “good” looks like
AI resolves the high-volume simple tail well; complex disputes, fraud and hardship cases route reliably to humans; and customers can always reach a person.
Feasibility & cost shape
The technology is mature for simple queries; the hard, decisive work is defining what AI must not handle autonomously.
The most useful scoping lesson in the library, and closest to our own philosophy. We’d deliberately under-scope the first deployment, instrument quality, and design the human hand-off before launch.
Based on publicly reported information about the Klarna (with OpenAI) work.
This is an industry example included for illustration. It is not a Leia Intelligence project, and no client of ours is implied. Figures are as publicly reported by the original parties.
Sources: Bloomberg · Fortune · CNBC · Klarna statements (2024 launch · May 2025 update)