AI Customer Service Lessons From Klarna in 2025

What Happened With Klarna's AI Customer Service?

Klarna announced in February 2024 that its OpenAI-powered assistant had handled 2.3 million conversations, two-thirds of its customer service chats, in its first month live. By May 2025 the company was recruiting human support staff again while saying the assistant still handled two-thirds of enquiries. In November 2025 its CEO said the assistant did the work of more than 853 full-time agents.

The February 2024 release also said customers could still choose to speak to a live agent. So a human route existed at launch. The 2025 announcements added active recruitment and an explicit promise that customers can always reach a person.

Why Did Klarna Bring Human Support Back?

Klarna brought human support back because its CEO, Sebastian Siemiatkowski, said cost had become too dominant a factor in organising customer service, and that this lowered quality. Klarna spokesperson Clare Nordstrom said the company wanted customers to always have the option to speak with a person.

Siemiatkowski made the cost comment to Bloomberg in May 2025, as CX Dive reported. Klarna began with a pilot offering flexible, remote roles to experienced professionals and students, and one stated aim was to replace some of the thousands of outsourced workers it uses.

Kate Leggett, a VP and principal analyst at Forrester, told CX Dive in November 2025 that Klarna over-pivoted to cost containment without weighing the longer-term effect on customer experience. She said customers had complained of generic answers to nuanced questions. These are an analyst's views, not Klarna data.

Did Klarna's AI Customer Service Fail?

Klarna's assistant delivers large efficiency gains by the company's own account, and Klarna has also acknowledged a quality problem, so the evidence supports neither a clean success nor a clean failure.

On efficiency, Klarna reported that the assistant handled two-thirds of enquiries, cut response times by 82% and cut repeat issues by 25%. Its CEO said the assistant had saved $60 million. In its Q3 2025 presentation, Klarna said customer satisfaction with the assistant was on par with a human agent.

On quality, the CEO tied the cost focus to lower quality. Klarna's customer service and operations costs were $50 million in Q3 2025, up from $42 million a year earlier. A Forrester analyst told CX Dive that customers complained of generic answers.

What Should AI Handle In Customer Service, And What Should People Handle?

AI customer service suits simple, repetitive requests such as refunds, cancellations and password resets. People should handle unusual, sensitive or high-stakes cases, and any customer who asks for a person. Forrester analyst Kate Leggett told CX Dive that highly repetitive, fairly simple questions are a strong fit for AI, and that nuanced questions fall to human workers.

The table below is a suggested routing, not a Klarna policy.

Klarna's spokesperson described a similar split: AI for straightforward requests and human specialists for the moments that count.

How Do You Measure AI Customer Service Quality?

Measure AI customer service on resolution, repeat contact, customer feedback and escalation reasons, alongside the cost and speed figures most companies report. Klarna's published figures cover the share of enquiries handled, response time, repeat issues and, in its Q3 2025 presentation, a satisfaction claim. The sources reviewed do not report escalation rates.

Monolit Labs' AI Operations Consultation reviews how a team handles support today, identifies where AI can remove friction and where it cannot, and delivers written findings with a prioritised action plan. Book a free consultation.

How Do You Build An AI Support System Customers Trust?

An AI support system earns customer trust when the service process comes first, the handoff to a person is easy, and quality is measured from day one. Five steps cover it, in this order:

1. Map the requests. Pull a sample of real conversations and sort them by type and difficulty before choosing what to automate.
2. Decide what AI handles. Start with simple, repetitive requests and keep everything else with people.
3. Design the handoff. Set triggers for a person to take over, such as a customer request, repeated failed answers or a sensitive topic. Pass the full conversation so nobody repeats themselves.
4. Set measures before launch. Agree targets for resolution, repeat contact and feedback alongside cost and speed.
5. Review on a schedule. Give a named owner the job of reading escalations and updating the rules.

Any trigger passes the conversation to a person, and an owner reviews each escalation to update the rules.

Monolit Labs builds the Digital Front Desk, which connects website, WhatsApp and email enquiries through capture, qualification, booking or handoff, follow-up and reporting. Where a simpler system is more reliable than AI, Monolit Labs builds the simpler system.

What Can Businesses Learn from Klarna’s AI Case?

Klarna’s reported experience raises a useful question for any company adopting AI: What are you optimizing for?

If the main goal is reducing costs or handling more requests, those numbers can look promising while parts of the customer experience suffer. If the goal is better service, businesses also need to measure whether people get accurate answers, resolve their issue and know how to reach a human when the automated path isn’t working.

A balanced service model can use AI for routine requests and human agents for complex or sensitive ones. But that only works when the handoff is clear. Customers shouldn’t have to repeat their story or get stuck navigating a bot to find help.

Before expanding AI in customer service, ask:

1. Which requests are genuinely routine?
   Look at real customer cases before deciding what to automate.
2. When should a person take over?
   Set clear triggers for complex issues, repeated failed answers or customer requests to speak to someone.
3. Can customers reach a human easily?
   Test the handoff from the customer’s point of view.
4. How will you measure quality?
   Track resolution and customer feedback alongside cost and response time.
5. Who reviews how the system is working?
   Assign people to spot recurring problems and improve the process.

The right measures depend on the business. Faster replies and fewer repeat inquiries can be useful signals, but they don’t tell the whole story on their own.

Find the Right Role for AI in Your Customer Service

Monolit Labs is a Bali digital agency for AI automation, custom software and web design. It starts with how your team handles enquiries today, then builds the workflow around that, using AI where it improves response and a simpler system where that is more reliable.

Every Klarna figure here is company-reported. The sources reviewed contain no independent measure of how often the assistant resolved issues well.

Let’s avoid your AI tools and system being slop and vague? Book a short consultation to review your rollout plan.

About the author: Shahab Hosseini is Director of Products and Ops at Monolit Labs and holds a PhD in Computer Science. He was previously CTO of a Bali-based wholesale marketplace and a front-end lead in Singapore. Connect with Shahab Hosseini on LinkedIn.

About Monolit Labs: Monolit Labs builds the systems that run a business: websites, apps, automations and operations platforms. It is an AI agency with one rule, AI only where it earns its place. Based in Canggu, Bali, it is starting with hospitality groups and works with professional services and growing companies across Indonesia and Southeast Asia.

This article is general information, not professional, legal or financial advice. Figures about Klarna come from its February 2024 press release and CX Dive reports of May and November 2025. They are company-reported, and no independent measure of resolution quality was found in the sources reviewed. Analyst comments are opinions. The routing and measurement tables are suggested frameworks, not Klarna policy. Details may have changed since publication, so check the linked sources for current figures. Monolit Labs is not affiliated with Klarna. Any engagement is scoped after a discovery conversation, and this article does not promise results or features.