Klarna's failed AI experiment in Customer Service

Klarna's aggressive AI adoption initially delivered significant cost savings but triggered a cascade of operational and financial challenges that forced a strategic reversal. CEO Sebastian Siemiatkowski acknowledged, "As cost unfortunately seems to have been a too predominant evaluation factor... what you end up having is lower quality", directly linking AI-driven austerity measures to degraded service standards.

_{Source: Bloomberg.com}

Financial Impact Beyond Marketing Savings

Key developments:

• Customer acquisition costs likely rose indirectly due to eroded trust, though explicit figures aren't disclosed. The company reported 25% more repeat inquiries from bot failures, suggesting hidden costs in unresolved issues.

• Valuation plummeted 85% from $45.6B (2021) to $6.7B (2022), exacerbated by AI missteps and market skepticism about fully automated lending models.

• Operational paradox: While AI slashed per-transaction service costs by 40%, the company now faces new expenses piloting an "Uber-style" remote human workforce to rebuild quality.

Siemiatkowski's admission—"We wanted to be [OpenAI's] favorite guinea pig"—reveals how experimental AI implementation compromised core lending safeguards. The company continues balancing automation gains with human oversight, recently reporting 152% higher revenue per employee despite workforce reductions. This case exemplifies the razor's edge fintechs walk when prioritizing efficiency over experiential quality in credit services.

What Actually Works in Lending + AI

The most successful approach is using AI for enhancement, not replacement.

1. Augment human underwriters, not replace them

2. Provide deeper insights into existing customers

3. Speed up document processing while maintaining human review

Prioritize AI for Core Lending Processes & Risk Management

Focus AI adoption on areas proven to enhance efficiency and decision-making in lending, such as underwriting, credit risk assessment, compliance monitoring, and portfolio management. These are critical functions for lenders, and AI can directly contribute to accuracy and speed. Use AI to bolster KYC, CDD, and overall risk management frameworks, learning from regulatory actions like the one faced by Klarna.

Adopt a Hybrid Approach to Customer Interaction

While AI can handle a high volume of routine customer service inquiries efficiently and cost-effectively, recognize its limitations. Ensure there is a clear pathway for customers to escalate to human agents for complex issues, disputes, or sensitive matters. Avoid removing human support entirely, as customer satisfaction for certain interactions requires empathy and nuanced problem-solving.

Balance Efficiency with Quality and Customer Experience

Do not let cost-cutting become the sole driver for AI adoption. Implement AI solutions strategically to improve the customer experience and the quality of service, not just reduce headcount or expenses. AI should enable the business to "do more with less" without compromising quality.

Be Transparent and Promote Responsible Lending

As Klarna's Wikipink initiative suggests, transparency about lending terms, fees, and repayment expectations is crucial. Use AI not just for assessment but also potentially to educate customers and offer personalized financial guidance. Build a business model perceived as offering "fair financing" rather than predatory or opaque products.

Implement AI Progressively and Iterate

AI adoption is a process that requires care, testing, adaptation, and continuous listening to customers. Avoid rushing to go "all-in" based on hype or aggressive short-term goals. Start with specific use cases, measure effectiveness beyond just cost savings, and iterate based on real-world performance and customer feedback.

Stay Abreast of Regulatory Developments

The financial industry, and BNPL specifically, is subject to increasing regulatory scrutiny. Ensure all AI applications, especially those involved in credit decisions and customer interactions, comply with current and evolving regulations, including potential mandates for human oversight or the "right to a human" in customer service. AI implementation must adhere to banking-specific regulations regarding capital, liquidity, internal governance, and control.

Leverage AI for Data Analysis and Underwriting

Utilize AI's capability to analyze vast amounts of data from various sources to inform underwriting decisions and assess customer ability to pay. This is fundamental to managing credit risk effectively

Podcast Video: Kapitus CEO Andrew Reiser Top Alternative Lending Insights

This is a masterclass in positioning for market cycles that many lenders miss.

⚡ KEY STRATEGIES REVEALED:

Kapitus started with manual underwriting around a table (literally rang a gong for each app). The lesson? You CAN'T scale intimacy - but you can build systems that preserve the judgment that made you successful small. Most lenders try to automate everything and lose their edge.

Every time banks retreat (2008, COVID, now with rate hikes), prepared alt lenders grow exponentially. Reiser doesn't just survive these cycles - he positions Kapitus to capitalize on them.

Kapitus equipment finance acquisition reveals the real play: Take your fast underwriting engines and apply them to slow, traditional products. This is where outsized returns live - not chasing the same merchant cash advances everyone else offers.

Context vs. Speed in Business Lending: "Two guys with same FICO - one has great business, one has horrible business." Consumer lending has trillions of data points. Business lending needs contextual intelligence. Most fintechs optimize for speed and miss the complexity.

💡 THE REALITY CHECK:

Regulation isn't your enemy (if you embrace it as competitive advantage)

Technology enables scale, but underwriting discipline drives profits

Market cycles are predictable - position for them instead of just reacting

Read more here

Headlines You Don’t Want to Miss

Oaktree Specialty Lending Terminates OSI2 Facility Agreement

Oaktree Specialty Lending Corporation terminated its OSI2 Facility Agreement on May 14, 2025, after repaying all outstanding borrowings under the loan and security agreement originally set to mature in January 2029. The termination followed the full repayment of obligations, effectively ending the facility ahead of its scheduled maturity.

Alpaca Launches Securities Lending for Broker API

Alpaca has launched its Fully Paid Securities Lending (FPSL) program for US equities through its Broker API, enabling fintech partners and brokerages to offer clients passive income by lending out fully owned securities to institutions, with fees shared between brokers and clients. This new service aims to diversify revenue streams, enhance client retention, and provide seamless integration for partners, reflecting Alpaca’s ongoing commitment to expanding financial infrastructure and yield opportunities.

JPMorgan Uses Kinexys to Settle OUSG Tokenized Treasuries on Public Blockchain With Chainlink, Ondo

JPMorgan, in collaboration with Chainlink and Ondo Finance, has completed its first cross-chain Delivery versus Payment (DvP) transaction, settling Ondo’s tokenized U.S. Treasuries Fund (OUSG) between its permissioned Kinexys Digital Payments network and the public Ondo Chain using Chainlink’s cross-chain orchestration technology. This milestone demonstrates a scalable, atomic settlement process for real-world assets on public blockchains, marking a significant step in integrating traditional finance infrastructure with decentralized finance markets.