Klarna’s customer service story is the rare AI case where the timeline matters as much as the technology. In 2024, the buy-now-pay-later company announced it would cut hundreds of customer service roles and use an artificial intelligence chatbot instead, a change expected to save the company millions. Then, about a year later, customers complained about degraded quality of customer service, and Klarna quietly shifted course.

But the reversal was not the “humans win” moment it first appears to be. Klarna did not simply bring back the full-time customer service agents it had cut, who it contracts through an outside agency. Instead, the company recruited human customer service agents back in a different format, one Klarna CEO Sebastian Siemiatkowski described as “an Uber type of set-up.” In this model, the AI chatbot continues to handle most basic queries, while a growing number of gig workers handle the more advanced ones.

This is where the gig-economy shift stops being a cultural talking point and starts looking like an operating strategy. When companies integrate AI into customer workflows, they can reduce headcount quickly because the chatbot can absorb repetitive questions, basic ticket triage, and standard policy explanations. The tradeoff is that AI performance can wobble when customers ask ambiguous, angry, or edge-case questions, which is exactly where “degraded quality” complaints tend to cluster. Klarna’s move suggests the company hit that boundary, then responded with a structure that keeps automation benefits while restoring human coverage where it’s most needed.

The “Uber type of set-up” matters because it changes who bears operational risk. Traditional customer service staffing is a predictable cost curve: you budget seats, training, scheduling, and coverage. A gig-style approach shifts part of that cost and capacity planning to on-demand work, theoretically letting the company scale labor up and down around demand spikes and AI limitations. It also lets the company keep the AI-first surface area for most customers while keeping humans available only when the conversation becomes complex.

For executives, there is an uncomfortable middle question: is this a temporary patch or a durable redesign of the service organization? The Guardian’s account makes clear Klarna used AI to reduce roles, then returned to human labor when customer feedback made the AI-only approach untenable. But it returned in a way that still avoids the full reinstatement of the earlier staffing levels. That pattern is increasingly relevant in a broader economy where many service functions are being automated and where the labor market is already familiar with flexible work arrangements.

There is also a second-order implication for investors and boards: AI cost savings are not a one-line story. The initial financial incentive is straightforward, and Klarna framed the AI chatbot rollout as something expected to save the company millions. Yet the customer complaints show that savings can be fragile if user experience deteriorates. Klarna’s response indicates that the “savings vs. quality” equation can be rebalanced by changing the labor model, not just the AI model. In other words, the solution to AI underperformance may be reallocating labor into a different contractual structure.

Regulatory background matters here, even when regulators are not the headline. Customer service is one of the consumer-experience layers regulators care about, because it affects disputes, billing, cancellations, and complaint handling. While the source does not mention specific regulators or penalties, it does highlight a common governance challenge: when AI handles the first line of support, companies need to ensure the handoff to humans is reliable and timely, especially for higher-complexity issues. In a gig setup, boards should ask harder questions about escalation, training standards, consistency of responses, and auditability across a distributed workforce.

Finally, Klarna’s story is a signal flare to other operators. If you are running a product or risk function at a company planning AI-driven reductions, Klarna’s experience suggests a likely failure mode: customer friction can surface after the automation goes live, forcing a reallocation of resources. And if you are considering gig labor as a buffer, the “Uber type of set-up” description underscores that this is not just a staffing tweak. It is a redesign of the customer support system, with new operational dynamics, new quality controls, and a different way of measuring whether AI integration actually improves outcomes or just changes who does the work when it breaks.