Klarna CEO Sebastian Siemiatkowski on Getting AI to Do the Work of 700 Customer Service Reps

Klarna's AI Transformation and Strategic Vision

• Klarna, founded in 2010 by Sebastian Gunnarsson in Stockholm, is a global payments and commerce platform with:
  • $100 billion in annual transaction volume across 20+ countries.
  • Approximately 100 million consumers and 500,000 merchants.
  • 4,000 employees and a fully regulated neobank status.

Customer Support Implementation and Outcomes

• The initiative to deploy AI began in November 2022 after Sebastian Gunnarsson secured a direct meeting with Sam Altman by leveraging Sequoia Capital's dual investment in Klarna and OpenAI.
  • Klarna agreed to act as a "guinea pig" for OpenAI, establishing a joint Slack channel for rapid experimentation.
  • The company immediately prioritized GDPR compliance and data structure to enable safe internal experimentation.
• The AI deployment originated in the dispute resolution department to handle complex evidence gathering between merchants and consumers.
  • A team built a RAG (Retrieval-Augmented Generation) architecture within two months to create a decision-support co-pilot.
  • The system successfully cleared the backlog, prompting Klarna to ask for more tickets as the AI resolved disputes faster than human agents.
• Klarna transitioned from a co-pilot to a fully autonomous customer service agent after validating that AI resolution metrics matched or exceeded human performance.
  • The team adhered to a strict rule requiring full transparency to customers about AI interaction.
  • An initial bug where AI interaction was undisclosed led to a review of 3,000 transcripts; the AI's high quality in this instance confirmed readiness for production.
• Operational metrics from the AI implementation show dramatic improvements over previous human-agent workflows.
  • Average customer service resolution time dropped from 14 minutes (human) to 2 minutes (AI).
  • The shift allowed for the elimination of 700 full-time employee contracts (mostly with third-party contractors).
  • The efficiency gain generated approximately $40 million in annual improved profitability against $2 billion in revenue.

Strategic Insights on AI Limitations and Human Roles

• Sebastian Gunnarsson predicts AI will not fully replace customer service agents in the near future, citing that a segment of customers explicitly prefers human interaction due to past negative experiences with legacy chatbots.
  • Klarna notes that the most common trigger for customers to request a human agent is the instinctive fear of poor AI quality, a perception the company is actively working to shift.
  • Current data shows that 30% of users currently opt out of AI assistance, primarily due to trust issues rather than a desire for human contact per se.
• The company views the primary value of AI not as total replacement, but as a tool to eliminate the "average" nature of LLMs while forcing human teams to improve documentation quality.
  • Gunnarsson argues that LLMs "work towards the average" and lack true creativity, making them superior for standard copy but inferior for generating "quirky," brand-defining text.
  • He suggests that for creative copy to outperform AI, humans must focus on the "extreme" and "out of the box," whereas LLMs naturally remain "in the box."
• Future internal AI products are designed to function as a "digital financial assistant."
  • The roadmap includes an "on steroids" version of the customer service assistant that will proactively advise users on Klarna services and potential savings.
  • Within 6 to 12 months, Klarna plans to launch truly disruptive services that analyze user finances to automatically suggest bank switches or savings opportunities.
  • This vision aims to eliminate banking "excess profits" derived from customer inertia by making financial mobility frictionless.

Internal Operations and Knowledge Management

• Klarna has centralized its information silos into a proprietary internal knowledge graph powered by Neo4j, enabling the creation of "Kiki," an internal AI chatbot.
  • Kiki allows employees to query organizational data, such as team structures, system launch steps, and operational policies.
  • The system serves as a hybrid of semantic search and an internal Wikipedia-style interface.
• The company is deprecating major enterprise software systems to reduce fragmentation and improve data accessibility for AI.
  • Salesforce is being replaced for many workflows in favor of Slack-based workflows and in-house custom tools.
  • Workday is being sunsetted for HR data management, with payroll functions retained via "Deal" to ensure organizational data is accessible to Kiki.
• Gunnarsson advocates for "Build over Buy" for emerging AI use cases to foster internal learning and capability development.
  • Klarna built a deep-interview bot for employee engagement that now has commercial equivalents, but the company values the internal knowledge gained during development.
  • The company advises against relying on third-party tools if their APIs are not public, as this prevents AI models from understanding how to interact with the software.

Marketing, Creativity, and Product Innovation

• In marketing, AI is used to accelerate campaign creation from a multi-month timeline to under one week, ensuring regulatory compliance and brand consistency.
  • Creative teams use AI avatars and generated assets to produce marketing materials, a process that reduces the need for traditional photography and editing staff.
  • The company views these tools as amplifiers for creativity rather than replacements, allowing non-technical staff to execute complex visual campaigns.
• Klarna has experimented with AI-generated personalized imagery within its shopping app.
  • A test generated custom category images (e.g., a shoe) based on a specific user's purchase history and brand preferences, resulting in a reported desire from the user to purchase the non-existent item.
  • This indicates a future trend where products and images are generated on demand based on individual profiles, potentially reducing inventory waste.
• Regarding the future of e-commerce, Gunnarsson distinguishes between brand retention and curation.
  • He believes brands (e.g., Nike) will remain dominant, but retailers will increasingly rely on AI for product curation and recommendation engines.
  • He predicts that AI stylists will eventually replace human recommendation agents, though complex travel planning remains a harder challenge due to the need for deep personal context.

Societal and Ethical Perspectives

• Gunnarsson calls for the global implementation of electronic identification (e-ID) to combat the rising threat of AI-generated fraud and impersonation.
  • He highlights the risk of "deepfake" avatars, citing his own use of an AI avatar for merchant presentations as a warning sign of potential misuse.
• On the topic of employment, he rejects the binary debate between stopping AI progress or assuming new jobs will automatically appear.
  • He cites the redundancy of 10,000 EU translators due to tools like DeepL as evidence that retraining is not a guaranteed solution for all displaced workers.
  • He advocates for societal support mechanisms to aid individuals affected by AI disruption without halting technological advancement.
• Gunnarsson identifies the three job categories AI should target first to minimize public backlash: CEOs, bankers, and lawyers.
  • He notes that replacing these roles generates little public outcry compared to replacing creative or service roles, which trigger strong emotional responses.
  • He observes that long-term physical labor (e.g., truck driving) may be harder to automate than knowledge work, despite recent hype around robotics.

Rapid Fire Takeaways

• Most Admired Figure in AI: Sam Altman.
• AI Art Collection: Gunnarsson does not currently collect AI art but is open to it if the image evokes a genuine emotional reaction, regardless of its origin.
• Advice for Founders: Lean into AI, experiment, and learn; do not fear the technology or dismiss it as a temporary trend.
• AGI Risk: He acknowledges the uncertainty but suggests that fearing existential threats should not paralyze innovation.
• Best Advice for Employees: Treat AI as a tool to understand and explore, similar to how he treats the unpredictability of the future.