Skip to content
← All AI workflows

Klarna: draft multilingual payment replies with human support

Klarna's app assistant handles payment questions, return and refund inquiries, disputes and invoice inaccuracies. The pattern combines multilingual support, verified account facts and a route to a live agent. Klarna's production prompt, tools and application code are not public.

From the source

Public article page from Klarna
Klarna · Captured 2026-10-14Read the original

How it works

  • The first-month release reports 2.3 million conversations and two-thirds of customer-service chats. Resolution was reported below two minutes versus 11 previously. Sampling and calculation details are not supplied.
  • The 700-agent workload equivalence is not a confirmed job reduction. The $40 million profit improvement was a projection, not achieved savings.
  • The described support scope includes payment questions, returns, refunds, disputes and invoice issues. Prompt, tools and application code are undisclosed. Steps in the original record are an editorial reconstruction, not Klarna’s original workflow.

Results

Klarna's February 2024 first-month announcement reports 2.3 million conversations, two-thirds of customer-service chats and errand resolution below two minutes versus 11 minutes previously. It also reports fewer repeat inquiries but omits sampling and calculation details. These historical launch claims do not verify sustained 2026 performance.

Workflow guide

  1. Limit an isolated reproduction pilot to approved payment-status and invoice-explanation topics in defined languages. Provide an obvious route to a live agent.

  2. Retrieve only authenticated balances, payment schedules and policy facts through approved read-only access.

  3. Draft a cited response in the customer's language. Do not invent balances or financial entitlements.

  4. Obtain reviewer approval before any message is sent. Keep refunds, disputes and invoice changes as drafts for separate authorized handling.

  5. Transfer unresolved or sensitive cases to a live agent with checked facts and a conversation summary.

  6. Compare resolution time, correctness, seven-day repeat contacts and completed human-support requests across the fixed pilot sample. Disclose eligible-ticket counts.

Sources and downloads

See this workflow in order-to-cash →Discuss this workflow →

Book a call

A free 30 minute call. Tell us where you are with AI, what your IT allows, and what management needs to see. We'll tell you honestly what we would do first.