Klarna's AI Fumble Exposed: The Real Cost of Ditching Humans in Design & Workflows
Klarna's attempt to replace 700 customer service employees with AI in 2023 seemed like a win for efficiency and cost savings - reducing chat times to 2 minutes and saving $40 million annually. But the move backfired. Customers faced unresolved issues, robotic interactions, and repeat inquiries surged by 25%. Satisfaction plummeted, with the Net Promoter Score dropping from +45 to +28 and churn increasing by 15%.
The lesson? Automation can't replace human qualities like empathy, judgment, and emotional intelligence. Klarna has since pivoted to a hybrid model, blending AI's speed with human oversight to fix these gaps. Businesses should prioritize customer experience over cost-cutting and adopt human-in-the-loop systems to balance efficiency with quality.
Klarna's AI-Driven Approach: What Went Wrong?
The Switch: From Human to AI-Only Support
In 2022, Klarna made a bold move to cut costs by transitioning from human-led customer support to an AI-only system. This shift resulted in the layoff of roughly 700 employees as the company leaned heavily into automation. By February 2024, Klarna's AI assistant was managing 75% of customer chats - handling approximately 2.3 million conversations across 35+ languages. This workload was equivalent to the output of 900 human agents, with the AI automating 66% of customer inquiries.
On paper, the numbers looked promising. The AI slashed average chat resolution times to just two minutes, compared to the eleven minutes it took human agents. Additionally, Klarna's revenue per employee skyrocketed by 152%, even as the company reduced its workforce by 40%, dropping from 5,527 employees in December 2022 to 3,422 by the end of 2024.
CEO Sebastian Siemiatkowski expressed excitement about the experiment, even calling Klarna OpenAI's "favorite guinea pig". However, despite these efficiency gains, the approach revealed serious cracks.
The Problems: Falling Satisfaction and Missing Context
While the efficiency metrics painted a rosy picture, the reality for customers was far less appealing. The AI struggled with intent recognition, often misunderstanding customer needs and delivering irrelevant responses. This left many users frustrated when trying to resolve important issues.
Complex problems became a sticking point. Klarna attempted to automate high-stakes tasks like resolving disputes, handling fraud cases, and addressing financial inquiries. Unfortunately, the AI frequently hit roadblocks, unable to grasp the nuance required for these situations. Customers often endured long, unproductive conversations before finally being escalated to a human agent.
A major issue was the lack of empathy in AI interactions. Responses felt robotic and failed to acknowledge customer frustration, leaving users with the impression they were speaking to a machine rather than receiving genuine support.
Adding to the frustration were handoff problems. When the AI couldn't resolve an issue, it failed to seamlessly transfer the conversation to a human agent. Customers often had to start over, repeating their concerns from scratch. This not only extended resolution times but also confused customers about whether they were interacting with a human or a machine.
Transparency was another weak spot. The AI assistant didn't clearly identify itself, leading customers to assume they were speaking with a human. When they realized otherwise, it eroded trust and left many feeling misled.
These shortcomings led to a 25% rise in repeat inquiries as customers struggled to get proper resolutions. Support teams found themselves overwhelmed with escalations that could have been handled more efficiently if routed correctly from the start.
"We went too far." - Sebastian Siemiatkowski, Klarna CEO
"As cost unfortunately seems to have been a too predominant evaluation factor... what you end up having is lower quality." - Sebastian Siemiatkowski, CEO of Klarna
AI-Only vs. Human-Led Support: Side-by-Side Comparison
The differences between Klarna's AI-only model and its previous human-led support approach became clear when comparing key performance areas:
| Dimension | AI-Only Support | Human-Led Support |
|---|---|---|
| Empathy & Understanding | Scripted responses; lacked emotional connection | Natural conversations with emotional intelligence and context |
| Complex Problem Resolution | Struggled with disputes and fraud cases | Handled nuanced situations with judgment and creativity |
| Response Time | 2 minutes average | 11 minutes average |
| Scalability | Managed 2.3M conversations in 35+ languages | Limited by human capacity and working hours |
| Customer Satisfaction | 25% rise in repeat inquiries | Higher first-contact resolution rates |
| Cost Efficiency | $40M annual savings initially | Higher costs but better outcomes |
| Transparency | Confused customers about AI vs. human interaction | Clear identification and accountability |
The data highlighted a stark contrast: while the AI excelled at speed and scale, it fell short on quality and customer satisfaction. The rise in repeat inquiries meant that the efficiency gains were undermined by the need for customers to contact support multiple times for proper resolutions.
Klarna's experience underscores a critical point: speed and automation can't replace the human touch in complex or sensitive situations. Whether dealing with disputes, fraud, or nuanced financial concerns, the AI's inability to understand context and exercise judgment often created more problems than it solved.
Should customer service be 100% automated?
Why Humans Still Matter in Design and Workflows
Klarna's challenges with relying solely on AI support highlight an important truth: some aspects of business simply can’t function without human expertise. While AI excels at crunching data and handling repetitive tasks, it struggles with the nuanced, contextual, and emotionally intelligent work that humans naturally excel at.
What Humans Bring That AI Cannot
Humans possess emotional intelligence, creativity, and contextual judgment - qualities that are essential for creating meaningful user experiences and solving complex problems.
Empathy stands out as the most critical human advantage. Roman Krznaric describes empathy as "the art of stepping imaginatively into the shoes of another person, understanding their feelings and perspectives, and using that understanding to guide your actions". This deep understanding of user emotions and motivations is what separates good design from great design. While AI can analyze behavior patterns, it can’t replicate the genuine understanding a human brings when addressing a frustrated customer struggling with a difficult interface.
Creativity is another area where humans shine. AI might generate variations on existing patterns, but humans are the ones who bring original ideas to the table. UX design leader Lyndon Cerejo sums it up perfectly: "Using our head, heart, and hands together to make a transformative difference is what distinguishes us from AI and makes us human, creative, and innovative".
Adaptability and collaboration are also uniquely human strengths. Designers can pivot when requirements change, integrate feedback, and align their work with evolving user needs - something rigid AI algorithms simply can’t do. These qualities have a measurable impact on business outcomes. Companies that focus on human-centered design and interaction see higher customer engagement, growing revenue 1.4 times faster and increasing customer lifetime value 1.6 times more than those that don’t prioritize customer experience.
Where AI Falls Short on Complex Decisions
AI often lacks the judgment needed for tasks like curation and refinement. While it can generate countless design options, it takes human insight to identify which ones are truly relevant. As Lorselle Ann, Senior Product Owner at Slash, explains: "AI can offer a wealth of data and predictive power, but the human touch in interpreting these insights is what transforms good designs into great ones".
Complex problem-solving is another area where humans excel. Years of experience - both successes and failures - enable humans to recognize subtle patterns and navigate ambiguous situations where data alone doesn’t provide clear answers. AI, for all its analytical power, doesn’t share this depth of understanding.
Curiosity is yet another human trait that AI can’t replicate. Humans ask "what if" and explore possibilities beyond existing data, often leading to breakthroughs that purely analytical approaches might miss. Klarna’s reliance on AI-only workflows demonstrated these limitations. When customers needed help with disputes, fraud cases, or other nuanced issues, the AI often fell short, unable to fully grasp the context or exercise the judgment necessary for effective resolution. These shortcomings reinforce the importance of human oversight in even the most advanced workflows.
Klarna's Return to Human-Centered Workflows
Recognizing these limitations, Klarna shifted back to a human-centered approach. CEO Sebastian Siemiatkowski openly admitted, "we went too far", signaling a renewed focus on balancing technology with human involvement.
The company is now rebuilding its human support teams, understanding that automation’s cost savings mean little if customer satisfaction suffers. Klarna has adopted a hybrid customer service model that blends AI’s efficiency with the human qualities of empathy and judgment. This approach acknowledges that while AI can handle routine tasks, complex situations require human intervention from the outset.
This shift reflects a growing industry trend. Mustafa Suleyman, CEO of Microsoft AI, emphasizes the importance of collaboration between human experts and machine learning engineers: "To create high-quality AI experiences, we need to integrate product design expertise directly into our AI development process".
The lesson from Klarna’s experience is clear: successful AI implementation requires strategic human oversight from the very beginning. Instead of replacing humans, the most effective approach involves humans defining strategies, setting parameters, and retaining control over critical decisions, while using AI to enhance efficiency and scale.
Marc Gamet, CEO of Slash, captures this balance perfectly: "The future of product design lies in the collaboration between AI and human designers. By combining the analytical power of AI with the creativity and empathy of humans, we can create truly exceptional products".
sbb-itb-e464e9c
Best Practices for AI-Native Product Development
Klarna’s misstep with AI highlights the importance of balancing automation with human oversight. To succeed, organizations must design AI systems that prioritize user trust and satisfaction instead of focusing solely on cutting costs.
Putting User Experience Before Cost Savings
For AI to thrive, user experience needs to take precedence over cost-saving measures. Transparent and explainable outputs are central to building trust. When users understand how AI works, they’re more likely to trust and rely on it.
Some companies are already leading the way. Cisco, for instance, joined the Rome Call for AI Ethics in 2024, committing to transparency, inclusivity, and accountability. They’ve built internal processes to ensure fairness in AI technologies before deployment. This kind of intentional design prioritizes clarity, user control, and collaboration. Users should always know what the AI is doing, retain control over critical decisions, and feel like collaborators rather than being replaced.
Transparency alone isn’t enough, though. Human oversight plays a crucial role in refining and validating AI outcomes, ensuring the system remains reliable and fair.
Building Human-in-the-Loop Models
Human-in-the-loop (HITL) models create a feedback loop that enhances AI’s accuracy and reliability. Instead of aiming for perfection from the start, these models focus on continuous improvement through human intervention.
The process often begins with full human oversight, which can gradually decrease as the AI demonstrates consistent performance. For example, Illumina, a healthcare company based in Massachusetts, uses Vertex AI for clinical workflows. Domain experts guide the AI’s outputs, helping to identify subtle signals in patient data. This collaboration not only improves accuracy but also builds confidence in the system.
To make HITL models effective, organizations need a clear strategy. This includes governance structures, ethical guidelines, and decision-making frameworks that integrate HITL into workflows seamlessly. Triggers for human intervention - like low AI confidence scores or complex scenarios - should be well-defined. UPS Capital, for instance, uses machine learning for delivery confidence but ensures human experts step in when needed, reducing errors and improving the system.
A tiered response system is particularly effective. Automating routine tasks while involving humans for complex issues avoids the pitfalls of full automation that led to customer frustration in other cases. Continuous feedback ensures the AI adapts and evolves to meet user needs over time.
Using Feedback to Improve AI Systems
User feedback is a powerful tool for transforming decent AI systems into exceptional ones. Data shows that 88% of users report better experiences when their feedback is implemented, and 74% remain loyal when their suggestions are acted upon.
Every time a human intervenes to address an AI shortcoming, that interaction becomes a learning opportunity for the system. Over time, this creates a cycle where the AI grows more capable while still relying on human input for edge cases.
Feedback doesn’t just enhance satisfaction - it also delivers measurable results. Leading models have seen a 35% boost in user satisfaction and a 25% reduction in response times by leveraging feedback effectively. Organizations can gather this input through surveys, feedback forms, and in-app reporting tools. Natural language processing can then analyze the data to uncover patterns and recurring issues that might otherwise go unnoticed.
Tracking metrics like error reduction rates, customer satisfaction trends, and return on investment is essential for measuring progress. For example, an educational platform that revamped its interactive features based on real-time learner feedback saw a 50% increase in user engagement and a 60% improvement in course completion rates.
How Bonanza Studios Builds Reliable AI-Driven Solutions
Bonanza Studios has taken lessons from past AI challenges to heart, crafting a process where human involvement is central to AI integration. As a prominent product innovation studio, their approach is refreshingly balanced. Rather than fully automating workflows, they focus on creating AI-native products that complement human skills while ensuring strategic oversight remains firmly in human hands.
At the core of their philosophy is a belief that AI should elevate human creativity and decision-making - not replace it. This principle is embedded in every step of their product development. Their track record speaks volumes, with multiple client projects showcasing how advanced AI tools, paired with a human-first mindset, can deliver tangible outcomes for businesses.
Lean UX and Agile Methods at the Core
Bonanza Studios thrives on a structured yet adaptable framework, combining weekly design sprints and monthly delivery sprints to keep projects on track. Their process begins with targeted research into user needs and business goals. For instance, during a collaboration with ProBackup (October 2024 to July 2025), they used tools like ClickUp, Loom, and Figma for asynchronous feedback and weekly meetings. This approach significantly boosted project efficiency.
"They made sure that they fully understood the value that we are bringing to our customers."
– Willem Dewulf, CEO & Founder, ProBackup
In another example, a math and science learning app project in April 2024 demonstrated how their agile approach could fast-track results. Within just four weeks of receiving new designs, the client successfully onboarded their first users. These tightly managed sprints ensure a seamless integration of AI into workflows without compromising control.
AI Tools with Human Control
Bonanza Studios employs a "human-in-the-loop" strategy, ensuring that people - not automation - make the critical decisions. This approach was evident during a UX/UI design project for a B2B sales pipelines platform (July 2022 to January 2023), where AI tools sped up the design and prototyping stages.
Their commitment to this philosophy is also explored in their "UX for AI" podcast. In one episode, seasoned UX designer Adam shared his experience building a LinkedIn alternative called B150. He used AI-powered tools like Builder.IO to convert Figma designs into React code and Neon DB for backend infrastructure. While AI handled technical tasks, Adam's expertise drove the strategic and creative decisions.
This example highlights Bonanza Studios' core belief: AI can streamline execution, but human insight is irreplaceable when it comes to strategy, empathy, and creativity.
Building AI-Native Products for Business Growth
By combining agile methods with their human-in-the-loop approach, Bonanza Studios develops enterprise solutions designed to fuel growth. Their focus on enterprise-grade AI-native products requires a deep understanding of both cutting-edge technology and business priorities. One standout project was for a green energy company (April to December 2024), where they adapted timelines to accommodate feedback while consistently delivering high-quality results.
Another example is their work on a gifting-as-a-service platform (September to November 2022). Here, they crafted branding and website structures, blending AI tools with strategic human insights to meet market needs.
Their process revolves around research-driven strategies, user-focused design, and AI-enhanced development frameworks. Each step is guided by thoughtful human oversight to ensure the final product aligns with real-world business goals. As Manuel Dedio, CEO of Epapa, put it:
"We needed a counterpart who understands that speed, quantity, and quality are everything. Bonanza Studios does!"
This balance between swift execution and strategic depth underscores Bonanza Studios' commitment to avoiding the pitfalls of over-automation. Their success across industries - ranging from fintech to green energy to education - proves that AI achieves its full potential when paired with human expertise. By blending the strengths of AI and human creativity, Bonanza Studios consistently delivers solutions that not only meet business objectives but also provide exceptional user experiences.
The Future of AI in a Human-Centered World
Klarna's decision to lay off 700 employees in favor of AI-driven support - an action CEO Sebastian Siemiatkowski later admitted was excessive - highlights the risks of sidelining human oversight. Missteps like this can harm customer satisfaction and tarnish a brand's reputation.
Although 75% of workers now use AI in some capacity, only 21% of companies have formal policies governing its use. This suggests many organizations are deploying powerful AI tools without the proper safeguards in place.
A smarter approach involves adopting human-in-the-loop systems. These systems allow AI to handle repetitive tasks while humans step in for critical decision-making. This setup leverages AI's ability to process data efficiently while relying on human judgment and empathy for more nuanced situations. As Timothy Yeung from UBC Sauder puts it:
"AI can significantly scale the speed of decision-making, but ultimately, humans must make the high-stakes calls. The future belongs to those who know how to use AI while keeping creativity and critical thinking at the core of what they do."
This balanced approach not only enhances decision-making but also boosts productivity. Industries with strong AI integration report labor efficiency that's 4.8 times higher. Even more promising, 89% of full-time workers report feeling more fulfilled in their jobs after automation is introduced, with 91% attributing this to time savings and improved work-life balance.
To maximize AI's potential, organizations should treat it as a tool for supporting decisions - not making them outright. This requires investing in AI literacy training for employees, establishing governance frameworks to ensure accountability, and maintaining transparency about how AI operates within the company.
Looking ahead, AI systems must prioritize adaptability and remain centered on human needs. Companies that succeed will focus on creating modular systems that evolve with technological advances while preserving human expertise and dignity. These businesses will aim for solutions that enhance human capabilities rather than simply chasing efficiency.
As Tim King from Solutions Review aptly states: "AI without empathy is not just inhumane - it's bad business". The future will belong to organizations that combine AI's speed and precision with human insight to build trust and amplify strengths.
FAQs
Why did Klarna return to a hybrid customer support model after replacing human agents with AI?
Klarna shifted back to a hybrid customer support model after realizing that their AI-driven system fell short when it came to managing complex problems. The system also lacked the empathy required to deliver the kind of customer experience people expect, which resulted in a clear decline in satisfaction levels.
To tackle these issues, Klarna brought human agents back into the mix. By combining the speed and efficiency of AI with the understanding and personal touch of human support, they aimed to create a more balanced and effective approach to customer care.
How can businesses strike the right balance between AI automation and human involvement to ensure great customer experiences and efficiency?
To strike the perfect balance, businesses should let AI take care of repetitive, routine tasks, freeing up humans to focus on areas that demand empathy, creativity, or nuanced decision-making. It's equally important to regularly assess how well AI is performing to ensure it aligns with customer expectations and make adjustments as needed. Encouraging a team mindset that values innovation and flexibility can help integrate AI effectively, while still preserving the human element where it truly counts.
What can businesses learn from Klarna’s use of AI-only customer support about balancing automation and human involvement?
Klarna’s journey highlights a crucial lesson: depending entirely on AI for customer support can sometimes go wrong, particularly when empathy or nuanced problem-solving is needed. Their move to replace human agents with AI resulted in a noticeable drop in customer satisfaction, ultimately forcing them to reintroduce human support into the mix.
The main point for businesses is clear: automation should complement, not replace, human input. To maintain a strong customer experience, involve humans in key interactions, keep a close eye on how AI performs, and set achievable expectations for what AI can realistically handle.