Klarna’s customer service assistant is one of the most cited AI deployments in business, and one of the most instructive because its story has two chapters. In February 2024 the payments company reported that an OpenAI-powered assistant had taken over two-thirds of its customer chats in a month. In May 2025 Klarna said it would again hire humans and guarantee customers a path to one. This case study sets out what was claimed, what changed and what any team planning a support bot can take from it.

The setup: what Klarna launched

According to Klarna’s February 27, 2024 press release, the assistant was built with OpenAI and had been live globally for one month. It ran inside the Klarna app, around the clock, in 23 markets and more than 35 languages. Its tasks included refunds, returns, cancellations, payment questions, disputes and invoice errors, plus real-time information on balances and payment schedules. Customers could still ask for a live agent.

What the first month showed

Klarna published these results, all of which are company-reported:

  • 2.3 million conversations, equal to two-thirds of customer service chats
  • work equal to roughly 700 full-time agents
  • customer satisfaction on par with human agents
  • a 25% drop in repeat inquiries, which Klarna linked to better accuracy
  • resolution in under two minutes, compared with 11 minutes before
  • an estimated $40 million profit improvement for 2024

Two cautions apply. First, “equivalent to 700 agents” is a workload conversion, not a count of layoffs or an audited saving. Second, these were first-month figures from the company that built the system. They show the assistant handled volume quickly; they do not show how customers felt about hard cases.

The reversal: what changed in 2025

Customer Experience Dive reported on May 9, 2025 that Klarna was recruiting people for customer service again. A spokesperson said the company wanted customers always to have the option of speaking with a human, and CEO Sebastian Siemiatkowski told Bloomberg that cost had been too dominant a factor, which led to lower quality. Klarna described a pilot of remote, flexible human roles, with a goal of replacing thousands of outsourced workers.

A telephone call centre with staff at workstations
Human-staffed contact centres remain part of Klarna's model; this photo is illustrative and does not show Klarna. Photo: Petiatil / Wikimedia Commons, CC BY-SA 3.0

The assistant did not disappear. Klarna said it still handled two-thirds of inquiries, with an 82% improvement in response times and the 25% drop in repeat issues. The change was one of balance: AI for routine queries, people for sensitive or complex ones. A Verint survey cited in the report found more than two-thirds of customers had suffered a bad chatbot experience, and a Five9 survey found 86% valued empathy and human connection over speed.

How the case compares with typical deployments

Many teams copy the first chapter and ignore the second. Klarna’s numbers are strongest where questions are repetitive and answers are clear, such as order status or a payment date. They weaken when a customer is upset, a case is unusual or a policy needs judgment. A measure like average resolution time rewards fast closure of simple chats and can hide failures elsewhere. For rollout tactics, see our pre-launch audit for AI support and our overview of how AI is transforming customer service.

Lessons for teams planning a support bot

  1. Measure more than speed. Track repeat contacts, escalation rates, complaints and satisfaction by issue type, not only handling time.
  2. Keep a visible human option. Make the route to a person easy at any point, as Klarna now stresses.
  3. Do not set savings as the main goal. Klarna’s own CEO linked a cost-first approach to weaker quality.
  4. Segment by risk. Let AI take low-risk questions and send disputes, fraud and vulnerable customers to staff.
  5. Plan the staffing curve. Cutting teams before quality is proven makes it expensive to rebuild skills later.
  6. Read the legal side. Disclosure and liability rules differ by region; our rollout checklist is a start.

What to watch next

  • Whether Klarna publishes updated satisfaction and cost data under its mixed model.
  • How other consumer-finance firms describe their human-handoff policies.
  • Regulatory moves on the right to reach a human in automated service.
  • Independent studies comparing chatbot and agent outcomes on complex cases.

Key takeaways

  • Klarna reported its assistant handled 2.3 million chats, two-thirds of volume, in month one.
  • The 700-agent figure is an equivalence claim by the company, not an independent audit.
  • In 2025 Klarna said it would again offer a human option and recruit service staff.
  • The AI still handled about two-thirds of inquiries after the shift, per the company.
  • Hybrid designs, not full replacement, are the lesson.

Frequently asked questions

Did Klarna stop using its AI assistant?

No. According to Klarna’s 2025 statements, the assistant still handles about two-thirds of customer inquiries, but the company is adding human agents and a guaranteed option to reach one.

Was the 700 agents claim verified?

It is Klarna’s own estimate of work equivalence from its February 2024 release, and we have found no independent audit of it.

What should a company copy from Klarna?

Copy the segmentation: automate routine queries, measure quality beyond speed and keep humans easy to reach for sensitive cases.