The problem
Klarna serves consumers in many markets through its app. In its annual report, the company describes customer support as covering purchases, account management, returns and merchant disputes, in any language, around the clock (Klarna Group plc). The same report says customer service and operations expenses fell by $37 million, or 15%, in 2024, driven mainly by lower customer service costs.
The company framed AI as a way to serve a growing customer base at lower cost. Its first-quarter 2025 earnings release says AI is “enabling us to do more with less” across functions, with customer service as the clearest example (Klarna).
What they did
According to its annual report, Klarna developed the assistant with OpenAI in 2023 and launched it in the Klarna app in February 2024 (Klarna Group plc). The assistant handles customer service tasks such as refunds and returns, and is available around the clock in 23 markets and more than 35 languages.
Klarna moved fast. In its launch announcement on February 27, 2024, it said the assistant had held 2.3 million conversations in its first month, two-thirds of all customer service chats, and was doing the equivalent work of 700 full-time agents (Klarna). It also estimated a $40 million profit improvement for 2024.
The rollout came alongside a hiring freeze. Klarna’s full-time headcount fell from about 4,352 at the end of 2023 to about 2,831 at the end of 2025, a reduction the annual report attributes to a strategic decision to drive efficiency by using AI (Klarna Group plc).
Results
The strongest figures come from Klarna’s annual report on Form 20-F. It says the assistant handled 80% of customer service chats in 2025, according to the company’s chat log data, did the work equivalent of over 850 full-time agents, and delivered approximately $59 million in cost savings in 2025 (Klarna Group plc). The agent figure is the company’s own estimate, based on the reduction in conversations handled by agents.
The same filing says AI-handled chats resolved in two minutes on average in 2024, against 12 minutes for human agents, and that repeat inquiries fell 25% after launch. It reports satisfaction “on par with human agents”, based on internal surveys. For 2024, it puts savings at $39 million, close to the $40 million the company had projected at launch.
Klarna’s first-quarter 2025 earnings release adds a unit-cost view: customer service cost per transaction fell 40% between the first quarter of 2023 and the first quarter of 2025, from $0.32 to $0.19, “with no drop in consumer satisfaction” (Klarna).
All of these numbers are self-reported. They appear in regulatory and investor documents, but no independent study has measured them.
What didn’t go to plan
Quality became the issue. In a May 2025 interview with Bloomberg, reported by Entrepreneur, chief executive Sebastian Siemiatkowski said the AI customer service, while cheaper, had produced “lower quality” (Entrepreneur). He said customers must know “there will always be a human if you want”, and called investing in human support “the way of the future for us”.
Klarna then resumed hiring customer service staff after a freeze of more than a year, recruiting fully remote agents and targeting groups such as students, people in rural areas and Klarna users (Entrepreneur).
The company’s own documents and its chief executive’s remarks pull in different directions. The filings report satisfaction on par with human agents; the chief executive described a drop in quality. The surveys behind the satisfaction claim are internal and unpublished, so readers can’t settle the difference. What is clear is that Klarna kept the assistant at the center of its support while adding a guaranteed route to a person.
Lessons
- Measure quality with more than a satisfaction average. Klarna’s internal surveys showed parity while its chief executive saw lower quality; track the outcomes that matter for hard cases too.
- Keep a visible route to a person. Klarna’s correction was not to drop the assistant but to promise that a human is always available.
- Separate projections from results. Klarna projected $40 million at launch and later reported $39 million for 2024; report each with its basis.
- Treat headcount as a decision, not a by-product. The freeze cut staff by about a third in two years, which made adding human capacity back a recruiting project.
- Read unit costs, not just totals. Klarna’s cost per transaction fell 40% over two years, a measure that stays comparable as volumes grow.
Adapt it in your organization
Is this for you?
- Most of your support contacts are routine and can be answered from your own systems and policies.
- You can let the assistant act, such as issuing refunds or rebooking, not just answer questions.
- You can measure outcomes per customer, including repeat contacts and complaints, not only per chat.
- Customers can reach a person quickly when the assistant can’t help.
- You are prepared to set a quality threshold that slows the rollout if it is crossed.
A 90-day version
Pilot: deploy an assistant on two or three routine contact types, such as refund status and returns, for a share of customers, with a matched group served the current way. Owner: head of customer service. Success measure: cost per resolved contact down at least 20% with repeat contacts and complaints no worse than the matched group.
Metric to track
Cost per resolved contact, read together with repeat contacts within seven days. Klarna reported a 40% fall in cost per transaction over two years; a first pilot should expect a smaller gain and watch repeats closely.
Failure modes to pre-empt
- Treating an internal satisfaction average as proof of quality for complex cases.
- Hiding the route to a human, so frustrated customers churn instead of escalating.
- Cutting human capacity faster than quality data justifies.
- Announcing projected savings to the board as results.
Questions to take to your leadership team
- Which customer issues should always reach a person on the first try?
- What quality measure would tell us our assistant is failing before customers leave?
- How quickly could we add human support capacity back if we needed it?
Executive FAQ
Did Klarna stop using AI for customer service?
No. Its 2025 annual report says the assistant handled 80% of chats that year; Klarna added a guaranteed route to a human rather than removing the assistant.
How much did Klarna save with its AI assistant?
Klarna estimates about $59 million in cost savings in 2025 and $39 million in 2024, according to its annual report; the cost of building it is not disclosed.
About the curator & AI disclosure
Curated by Luciano Ayres
Luciano Ayres is Director of Artificial Intelligence at ArcTouch, with more than 20 years of experience in software and technology leadership and specialized studies in deep learning. Connect with him on LinkedIn.
AI-generated briefing. This case was produced through an automated AI research and writing workflow curated by Luciano Ayres. Sources are linked for independent verification; summaries and analysis may contain errors and do not replace professional due diligence.