Klarna and AI customer service: Chronicle of a change in strategy
The summary for decision-makers
Between 2022 and 2025, Klarna implemented the most radical use of AI in customer service of any European fintech company—and then spectacularly backtracked. This case study does not provide you with a blueprint to copy, but rather a lesson on where the line between efficiency gains and brand damage lies.
The passage of time: From AI triumph to reversal
May 2022 – Cost pressure, still without AI narrative
Klarna lays off around 700 employees (10% of its workforce). Officially: market conditions, profitability. Customer service, operations, and marketing are particularly affected. AI is not yet communicated as the cause at this point.
2023 – Infrastructure development and hiring freeze
Klarna enters into a partnership with OpenAI and develops generative AI use cases for support, translations, content, and reporting. At the same time, there is a hiring freeze outside of engineering. The strategy: take advantage of natural turnover, do not refill positions, but instead automate.
Spring 2024 – The PR narrative: "AI replaces 700 full-time employees"
Klarna communicates aggressively: The AI assistant handles two-thirds of all customer chats, around 2.3 million conversations in just a few months. Processing time per case drops from 11 to less than 2 minutes. Repeat inquiries decline by 25%. CEO Sebastian Siemiatkowski says AI can "in principle take over all jobs" – including his own.
August 2024 – Headcount reduction becomes program
The workforce shrinks from around 5,000 to 3,800. The plan: to reduce it to around 2,000 in the long term. Reason: AI-driven efficiency gains, especially in support.
May 2025 – The turnaround
Siemiatkowski publicly admits that costs were "too dominant an evaluation criterion" and that quality suffered as a result. Klarna announces that it will once again employ human support staff – but not as permanent employees, rather as gig workers (remote, home office, "Uber principle"). Target group: students, rural regions.
Fall 2025 – New target vision
Klarna still employs around 3,000 people, with a target of 2,500 by the end of the year. AI remains at the core of the setup, but the narrative is shifting from "AI replacing humans" to "AI plus humans for better service."
What exactly went wrong?
The problems came as no surprise to anyone familiar with customer service dynamics:
Robotic responses in special cases: AI was able to process standard inquiries quickly, but lacked empathy and the ability to escalate issues in the case of complaints, disputes, or emotional customers.
Longer solutions for exceptions: What was faster in 80% of cases took longer in the remaining 20%—and it is precisely this 20% that shapes brand perception.
Measurable loss of quality: Industry analyses point to declining customer satisfaction, increasing frustration with complex problems, and negative brand perception.
Reputational risk: The aggressive AI strategy was increasingly perceived as purely a cost issue—at a time when Klarna's enterprise value had already fallen from $45.6 billion to $6.7 billion.
Gary Marcus, NYU professor and AI critic, coined the term "The Klarna Effect" in 2025: Companies use AI as a pretext for job cuts, then realize that quality is declining and quietly rehire people.
The "So What" for e-commerce decision-makers
If you are faced with the question of what to do with your customer service, Klarna provides three reliable insights:
1. Full automation sounds appealing, augmentation delivers
The data shows that AI works excellently for standard cases, self-service, and 24/7 inquiries. It (still) fails when it comes to nuances, empathy, and controversial situations. The hybrid model—AI for volume, humans for value—is not a compromise, but the target design.
2. Costs without quality metrics are flying blind
Klarna optimized transaction speed but neglected relational trust. First-contact resolution increased, but customer sentiment declined when it came to complex problems. The result: savings that were eaten up by damage to the company's reputation.
3. Communication makes the difference
"AI replaces 700 jobs" is a different narrative than "AI frees up our team for the important cases." The first generates headlines and investor applause, while the second builds customer trust. Klarna chose the first and had to correct itself.
Specifically: What you can do
Segment your inquiries: Which ones can be standardized (delivery status, FAQs, simple returns)? Which ones require human judgment (complaints, goodwill, complex products)?
Define quality metrics alongside cost metrics: Not just processing time and ticket volume, but customer sentiment, NPS by problem type, escalation rate.
Build escalation logic in from the outset: if AI gets stuck, the transition to human intervention must be seamless—not as an exception, but as part of the design.
Test small, scale slowly: Klarna communicated the use of AI as a strategic gamble before its quality had been validated. That's risky.
The one sentence to take away
Klarna shows that "AI first, humans second" in customer service causes reputational damage in the medium term that can exceed savings – and that "AI assists humans" is not a step backwards, but rather the more sustainable approach.