AI in customer service: why Klarna walked it back and what that teaches
Klarna — a digital bank and global payments giant serving more than 120 million active users in 26 countries — became one of the most emblematic cases of applying artificial intelligence to customer service. With more than 3.8 million daily transactions and partnerships with major brands like Uber, Nike, H&M and Airbnb, the impact of its operational decisions echoes across the market.
Still, Klarna didn’t abandon AI in customer service. After stating that its assistant did the work equivalent to 700 people, the company began prioritizing the quality of human support and working on two tracks: AI and people. The lesson is one of design: AI serves part of the cases well, and the rest needs a person.
Timeline of the case
| When | What happened | Who said it |
|---|---|---|
| Feb/2024 | The AI assistant, built with OpenAI, reportedly handled 2.3 million conversations in 35 languages in the first four weeks, serving two thirds of chats and the equivalent of 700 full-time agents | Klarna (statement) |
| Feb/2024 | Tickets resolved in under 2 minutes, versus 11 minutes before, and 25% fewer repeat queries. Customer satisfaction on par with human agents | Klarna (statement) |
| May/2025 | CEO Sebastian Siemiatkowski tells Bloomberg the focus is now the quality of human support. Klarna describes a pilot with two tracks, scalable AI and quality human service | Bloomberg, via tech.co |
| Jun/2025 | Gartner predicts that half of the companies that planned to cut human service heavily will give up by 2027 | Gartner |
Almost all the numbers in the case come from Klarna itself. The reports we consulted treat them as company claims, without independent verification.
What AI did well
By Klarna’s numbers, volume and speed. 2.3 million conversations in four weeks, in 35 languages, resolved in under 2 minutes, is work a human team can’t do in the same timeframe. For repetitive, well-defined questions, AI delivers.
Where the math didn’t add up
The tech.co report, from May 2025, cites accounts of customer dissatisfaction and low-quality support as context for the decision to bring humans back. It doesn’t provide data on that. The CEO spoke of prioritizing quality, not of AI’s failure.
In our reading, the case shows a common risk: measuring service by volume and time and leaving quality out. A fast answer that doesn’t resolve counts as success on the dashboard and as a problem in the customer’s mind.
The hybrid model
Klarna called the new phase a pilot with two tracks: AI that scales and quality humans. It’s what Gartner recommends in its June 2025 statement: digital first, but not digital only. We explain how to design that in hybrid service: human + AI done right.
5 precautions before automating
- Define what AI resolves on its own and what passes to a person, by case type.
- Measure quality, not just volume: customer score, recontact and first-contact resolution. The guide is in how to measure an AI service agent.
- Let the customer choose to talk to a person. The 2026 Globo survey shows that 79% want that option (see 74% accept AI in customer service).
- Keep a human team trained for the cases AI passes along, instead of cutting it before testing.
- Start with one channel and review the math after a few weeks. WhatsApp is Brazilians’ favorite channel in the 2024 CX Brain survey, and the precautions to automate it deserve their own read.
To design that test with NorthStar, start on the sales and service with AI page.
Frequently asked
Did Klarna stop using AI in customer service?
No. According to the tech.co report (May/2025), the company moved to a two-track model, scalable AI and quality human support. The CEO said the change is one of priority.
How many people did Klarna's AI replace?
The company stated, in February 2024, that the assistant did the work equivalent to 700 full-time agents. It’s a Klarna claim, without independent verification. And it didn’t simply mean replacing people: the next move was to prioritize the human team for the cases AI doesn’t resolve.
Does this mean AI in customer service doesn't work?
It works for part of the cases. What the case shows is the risk of betting only on AI. Gartner predicts that, by 2027, half of the companies that planned to heavily reduce human service will give up on those plans.
How to avoid the same mistake in my company?
Define what AI resolves, measure quality and not just volume, and offer the exit to a person. Run the test on one channel before scaling.
Sources
- Tech.co. Klarna reverses AI overhaul, 20/05/2025. tech.co
- Tech.eu. Power of AI is “happening right now” says Klarna boss, 28/02/2024. tech.eu
- Gartner. 50% of Organizations Will Abandon Plans to Reduce Customer Service Workforce Due to AI, 10/06/2025. gartner.com
- Globo Gente / E&I Inteligência Setorial, Aug/2026. gente.globo.com
MARKETING · SALES · FINANCE · LOGISTICS
Intelligence for the whole team.
Scale for the business.
All your data in one layer: every team decides better and the company grows on numbers.
SCALE WITH DATA →