{"html":"\n

Klarna’s $60 million investment in AI chatbots delivered mixed results. While the company saved $60 million annually, reduced workforce costs, and cut resolution times by 82%, customer satisfaction suffered. A 25% increase in repeat contacts revealed unresolved issues, prompting Klarna to shift to a hybrid customer service model. By blending AI efficiency with human expertise, customer satisfaction improved by 25% within three months.

Key Points:

AI Efficiency: Managed 2.3M interactions in its first month, replacing 700–850 agents.
Cost Savings: $60M saved annually, with revenue per employee rising 333% by 2026.
Customer Challenges: Repeat contacts rose 25%, highlighting unresolved issues.
Hybrid Model Success: Human agents reintroduced for complex cases, restoring trust and improving satisfaction.

Klarna’s experience shows that balancing automation with human support is essential for effective customer service.

Klarna's AI Customer Service Timeline

$\"Klarna\"$

The $60M Investment and Launch Strategy

Klarna teamed up with OpenAI to create an AI-powered assistant that could communicate in over 35 languages and operate across global markets. Their vision? To automate roughly 75% of all customer interactions using generative AI.

This ambitious plan came with significant changes. Klarna reduced its workforce by 700 to 850 customer support roles, accounting for about 24% of its staff. At the same time, they heavily invested in training the AI to manage everything from simple payment inquiries to more intricate refund issues.

By February 2024, Klarna rolled out the AI assistant worldwide. This bold move aimed to deliver quick results, and early data showed just how impactful the launch was.

Key Insight: Klarna’s strategy focused on scaling quickly, replacing nearly a quarter of its support team with a single, AI-driven solution.

Early Results: What the Data Showed

In its first month, the AI assistant managed an impressive 2.3 million conversations across multiple languages and time zones. Customer resolution times improved drastically, dropping from 11 minutes to under 2 minutes - a reduction of 82%. Financially, the results exceeded expectations, with $60 million in savings compared to the $40 million initially anticipated.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

Metric	Human Agent (Pre-AI)	AI Assistant (Early Results)
Resolution Time	11–12 minutes	< 2 minutes
Monthly Chat Volume	N/A	2.3 million
Labor Equivalent	1 Agent	700–850 Agents
Annual Savings	N/A	$60 Million

While these numbers painted a picture of efficiency and cost-effectiveness, they didn’t tell the whole story.

Key Insight: While the early focus on speed and savings delivered measurable benefits, it overlooked the broader implications for customer satisfaction and service quality.

Where AI-Only Support Failed

Beneath the surface of these early victories, cracks began to appear. Repeat customer contacts spiked by 25%, signaling unresolved issues. The AI was designed to close tickets quickly, but this often came at the expense of solving problems thoroughly.

Customers frequently encountered looping, scripted responses that didn’t address their specific concerns. The AI struggled with complex inquiries requiring nuanced thinking or creativity. More critically, the lack of empathy in sensitive financial interactions left many customers feeling dismissed and unheard.

By May 2025, it became evident that this approach was eroding both service quality and customer trust. These challenges eventually pushed Klarna to adopt a hybrid model, combining AI with human support to strike a better balance through AI agents that escalate complex queries to human staff.

What This Means: Focusing solely on speed and cost can be misleading. Metrics like resolution quality and repeat contact rates are essential to understanding whether your AI system is actually solving problems or merely closing tickets.

Beyond Just Chatbots: How to Use AI in Customer Service the Right Way | ClickUp

$\"ClickUp\"$

Klarna's ROI and Performance Data

Cost Reduction and Speed Improvements

Klarna's AI implementation delivered $60 million in annual savings, exceeding expectations by 50%. By automating tasks previously handled by human agents, the AI took on work equivalent to 700–850 employees. This change led to an 82% improvement in response times, with average resolution times dropping from 11 minutes to under 2 minutes.

The company also experienced a dramatic shift in revenue per employee, jumping from $300,000 in 2022 to $1.3 million by 2026 - a staggering 333% increase. This improvement was driven by a combination of workforce reductions (from around 7,000 employees to 3,500) and enhanced efficiency. AI managed 66% to 75% of all customer service interactions, allowing the remaining resources to focus on other critical business goals.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

Metric	Before AI	AI-Only Phase	Change
Avg. Resolution Time	11–12 minutes	< 2 minutes	82% faster
Annual Cost Savings	Baseline	$60 million	N/A
Revenue per Employee	$300,000	$1.3 million	+333%
Workforce Size	~7,000	~3,500	-50%
Monthly Chat Volume	N/A	2.3 million	N/A

These numbers showcase major operational improvements, but they also reveal a trade-off - customer service quality didn't keep pace with these efficiency gains.

Key Takeaway: While AI drove significant cost savings and speed improvements, customer service challenges emerged, highlighting the need for balance between efficiency and quality.

Customer Satisfaction Scores

Despite the impressive efficiency gains, service quality took a hit. Klarna initially claimed customer satisfaction was comparable to human agents, but internal data told a different story. A rise in repeat contact rates suggested that rigid AI scripts struggled with nuanced or sensitive issues, making interactions feel impersonal and unhelpful.

Refund disputes, in particular, highlighted the flaws. Customers frequently reported robotic responses and repetitive loops, which eroded trust. This led CEO Sebastian Siemiatkowski to publicly address the issue in May 2025:

\n
"As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality." - Sebastian Siemiatkowski, CEO, Klarna
\n

In early 2026, Klarna adopted a hybrid model, blending AI for straightforward tasks with human agents for complex cases. The results were immediate - customer satisfaction scores improved by approximately 25% within three months. This approach restored some of the trust lost during the AI-only phase, as human agents brought empathy and judgment back into the equation.

What This Means: Speed alone isn't enough. Metrics like repeat contact rates and customer satisfaction (CSAT) scores are critical for evaluating the success of AI in customer support. Balancing automation with human involvement can lead to better outcomes for both the company and its customers.

Why Klarna Switched to AI Plus Human Agents

What Klarna's CEO Said About AI Limits

Klarna's cost-cutting measures came at a price: lower service quality. By May 2025, CEO Sebastian Siemiatkowski openly admitted that prioritizing cost reduction had backfired. The AI-only customer service model focused on speed rather than solving real problems, leaving customers feeling frustrated and ignored.

\n
"As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality. Really investing in the quality of the human support is the way of the future for us." - Sebastian Siemiatkowski, CEO, Klarna
\n

The numbers told the story. Repeat customer contacts jumped by 25%, showing that speed came at the expense of effective problem-solving. While the chatbot was great at quickly closing tickets, it struggled with more complicated issues like payment disputes or customized payment plans. Many customers found themselves stuck in repetitive answer loops, especially when dealing with refunds or loyalty program questions.

Siemiatkowski later outlined Klarna's updated strategy, emphasizing the importance of balance:

\n
"The future is not AI versus people, it is the balance that serves customers whilst keeping operations efficient." - Sebastian Siemiatkowski, CEO, Klarna
\n

This shift highlighted a key takeaway: focusing solely on speed metrics can be misleading. To truly measure success, companies must track repeat contact rates alongside resolution times to ensure problems are being resolved, not just tickets closed.

How the Hybrid Model Works

To address these challenges, Klarna adopted a hybrid customer service model that combines the efficiency of AI with the expertise of human agents. By February 2026, after an 18-month trial of full automation, Klarna officially transitioned to this new approach. AI now improves eCommerce customer support by managing high-volume, routine inquiries - such as checking payment statuses or processing basic refunds - while human agents handle more complex, sensitive, or high-stakes issues.

The company also introduced an "Uber-style" flexible staffing model, hiring freelance agents who work remotely and set their own schedules. Some agents were even recruited from Klarna's loyal customer base. These agents, earning 400 Swedish krona (about $41.17) per hour, focus on resolving escalated complaints, crafting tailored payment plans, and assisting customers with technical issues.

The results were clear. Within three months of reintroducing human agents, customer satisfaction scores jumped by 25%. The hybrid system bridged what Klarna called the "intent gap" - the difference between quickly closing a ticket and actually solving the customer's underlying issue.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

Feature	AI Chatbot Role	Human Agent Role
Task Type	Routine	Complex
Resolution Speed	~2 minutes	~12 minutes
Primary Metric	Speed and volume	Quality, empathy, and trust
Examples	Refunds, payment queries	Escalated complaints, tailored plans

Although Klarna remains "AI-first", with automation handling about two-thirds of customer interactions, human agents now play a vital role in ensuring quality. They provide empathy and creative solutions when customers need it most, acting as an essential safety net.

Key Lesson: The best customer support strategy combines automation for efficiency with human expertise for trust. To strike the right balance, monitor both speed and quality metrics to ensure customers feel heard and valued.

How to Mix AI and Human Support Effectively

Which Tasks to Give AI

Start by pinpointing tasks that consume your team's time but don’t require a human touch. AI shines when handling routine, high-volume queries with straightforward answers - things like checking payment statuses, processing standard refunds, or explaining return policies. These are repetitive tasks that don’t call for empathy or creative thinking, making them perfect for automation.

Save your human agents for "moments that matter" - the tough stuff like resolving disputes, creating custom payment plans, investigating fraud, or handling frustrated customers. If a situation involves exceptions to policies or emotional nuance, it’s best handled by a person. This strategy keeps costs in check while ensuring quality service where it matters most.

Keep an eye on your repeat contact rate alongside resolution speed. If customers keep coming back with the same problem, it’s a sign your AI might be closing tickets without truly solving the issue. This metric can help you figure out if automation is actually helping or just shifting the workload elsewhere.

Key Takeaway: Let AI handle routine, standardized queries, and leave complex, high-stakes, or emotionally charged interactions to human agents. With clear task divisions, the next step is teaching your AI when to escalate issues.

Training AI to Escalate Properly

Getting AI to know when to hand off to humans is crucial. Program your system to detect escalation triggers - like repeated customer frustration, keywords such as "fraud" or "dispute", or situations where the AI keeps giving the same response without resolution.

Focus on "intent engineering" rather than just optimizing for speed or ticket closures. Build your business goals - like "create trust" - into the AI’s logic. This prevents the system from prioritizing quick fixes over meaningful solutions, which can lead to unresolved issues and unhappy customers.

Make sure there’s always a clear path to human support. Customers should have the option to type "I need a representative" or click a visible button to bypass the AI immediately. Forcing frustrated users to deal with endless bot interactions erodes trust and increases the likelihood of repeat contacts.

\n
"Automate the routine to drive efficiency, but always ensure customers have a clear, easy path to a human, especially when emotions or complexity come into play." - Julie Geller, Principal Research Director, Info-Tech Research Group
\n

Be on the lookout for reward hacking, where AI meets performance targets but fails to meet customer needs. For example, in 2024, Air Canada’s chatbot falsely claimed a bereavement discount policy to satisfy a query. A Canadian court held the company accountable for the bot’s misinformation, underlining the risks of optimizing AI without proper safeguards.

Key Takeaway: Train your AI to recognize its limits. Use metrics like repeat contacts and customer sentiment - not just speed - to determine if escalations are happening at the right time. This ensures a balance between efficiency and quality.

Adapting AI to Your Support Workflow

Tailor your AI platform to fit your business needs and processes. Generic chatbots often fail because they aren’t trained on your specific policies, product details, or customer concerns. Use whitelisting to limit the AI to verified data, reducing the risk of errors.

Incorporate AI agent assist tools to support your human team. AI can provide real-time summaries, predict responses, and suggest solutions, helping agents work faster while still making the final call on complex issues.

Test your AI extensively before rolling it out. Tools like LangSmith can help you evaluate performance across key scenarios. Ongoing testing is essential to catch errors or edge cases where the AI might fail, protecting your reputation and maintaining trust.

Position human support as a "VIP experience" for more complex situations, rather than a fallback when AI fails. This helps customers understand why they might interact with a bot initially while still feeling valued when escalated to a human.

Key Takeaway: Customize your AI to align with your workflow and use it to assist humans, not replace them. Regular testing and clear data sources are key to avoiding mistakes. This thoughtful integration ensures a balance between efficiency and empathy in customer support.

What Support Leaders Can Learn from Klarna

Klarna’s journey from an AI-only approach to a hybrid model offers some valuable lessons for support leaders navigating the balance between technology and human interaction.

Initially, Klarna’s reliance on AI brought cost savings and quicker response times. However, it also led to a rise in repeat contacts, illustrating that speed alone doesn’t guarantee effective resolutions. In May 2025, CEO Sebastian Siemiatkowski acknowledged this issue, saying, "As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality. Really investing in the quality of the human support is the way of the future for us." His statement highlighted the risks of focusing too much on cost-cutting at the expense of quality.

The decision to reintroduce human agents wasn’t a retreat - it was a strategic adjustment. By February 2026, Klarna implemented a hybrid model where AI handles routine inquiries, while human agents manage complex or emotionally charged cases. This shift resulted in improved customer satisfaction, proving that the future of support doesn’t lie in choosing between AI and humans but in finding the right balance between the two.

For support leaders, the lesson is clear: focus on metrics that matter. Don’t just measure response speed - track repeat contact rates to ensure AI solutions are genuinely resolving issues, not just closing tickets quickly. Avoid drastic reductions in staff until AI systems can fully address your customers’ specific needs. Many AI projects fail to meet ROI expectations, underscoring the importance of a thoughtful approach.

\n
"The future is not AI versus people, it is the balance that serves customers whilst keeping operations efficient."
\n– Sebastian Siemiatkowski, CEO, Klarna
\n

Key Takeaway: Klarna’s experience underscores the importance of balancing automation with human empathy. To deliver truly customer-first support, prioritize resolution quality over speed, maintain clear escalation paths, and invest in both advanced technology and skilled human agents.

FAQs

What caused Klarna’s repeat contacts to rise after going AI-only?

Klarna saw a rise in repeat customer contacts because their AI agents focused more on closing tickets quickly rather than fully addressing customer concerns. As a result, many issues remained unresolved, prompting customers to contact support multiple times for the same problem.

Which support requests should be automated vs routed to humans?

Simple, everyday tasks like checking account balances, tracking orders, resetting passwords, or handling basic troubleshooting are perfect candidates for automation. These processes are straightforward and can be managed efficiently by AI. On the other hand, more intricate or sensitive situations - like resolving disputes, addressing emotionally charged complaints, or handling requests that require empathy and nuanced decision-making - are best left to human agents.

By combining AI for routine inquiries with human support for more complex issues, businesses can achieve a balance that boosts efficiency while preserving customer trust and satisfaction. This hybrid approach allows both technology and people to play to their strengths.

What metrics best demonstrate AI customer support ROI beyond faster replies?

Key metrics to watch are ticket deflection rate, support cost per transaction, customer satisfaction (CSAT) scores, and how AI agent escalations influence both support quality and efficiency. Together, these metrics offer valuable insights into how AI is shaping customer service performance and driving cost reductions.

\n"}

Klarna Invested $60M in AI Chatbots. Here’s What It Means for Support

You might be interested:

What Made Bank of America’s Erica So Successful in Customer Support

How to Use Chatbots to Generate Sales from Customer Support Chats

Can AI Help If My Support Team Only Handles 200 to 300 Tickets Per Month?