Klarna's AI support reversal is the clearest data point yet on chatbot fatigue

In February 2024, Klarna’s CEO said its OpenAI-powered assistant was doing the equivalent work of 700 customer service employees. By May 2025, Klarna was publicly recruiting humans back, piloting an Uber-style flexible-shift model for remote support staff. It’s one of the most-watched AI-support case studies precisely because Klarna didn’t quietly walk it back — it explained why, on the record.

CEO Sebastian Siemiatkowski put the reasoning bluntly: “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” He went further on brand trust: “From a brand perspective, a company perspective … I just think it’s so critical that you are clear to your customer that there will be always a human if you want,” as reported in CustomerExperienceDive’s coverage of the reversal. Notably, the AI assistant wasn’t scrapped — it still handles roughly two-thirds of inquiries, and Klarna reports an 82% improvement in response times. The shift is about restoring a human option, not abandoning AI.

The Hacker News reaction is the “let me talk to a human” genre in miniature

The 1,300+ point HN thread on Klarna’s reversal is a near-perfect specimen of the frustration this category generates. One commenter, having actually tested Klarna’s bot after the initial 700-person announcement, wrote: “I did what seemingly no other publication reporting on it did: signed up for Klarna, bought one item and used this bot. I was… not impressed?” Another described a specific, almost absurd failure mode: “A few days ago Klarna’s AI bot lied to me that it was human, lied about speaking Finnish as its native tongue, and wrote me a Haiku about Klarna’s cashback.” A third, self-identified as ten years deep in contact-center AI, offered the numbers behind the disappointment: “Rough order of magnitude you can expect to deflect around 30% of inbound contacts with an AI chatbot… The problem is that vendors are telling companies they can eliminate” far more than that.

That gap — between what’s sold and what a chatbot can actually resolve — is the entire “I just want a human” complaint genre in one thread. It shows up constantly on Reddit’s customer-service communities and in HN threads like Scaling customer support destroys it, where a founder reported that “Added a chatbot → customer satisfaction dropped 40%” even as dashboard metrics looked better, concluding: “Customers would rather wait 2 hours to talk to someone who gives a damn than get an instant reply from someone reading a script.”

What it means for buyers evaluating AI support agents

Klarna’s walk-back isn’t an argument against AI support agents — the company kept the AI handling most volume and the vendor category has matured considerably since 2024. It’s an argument for architecture: the vendors worth evaluating are the ones that make escalation to a human fast, visible, and not adversarial, rather than ones optimized purely to suppress ticket volume. When you’re comparing options on Crail’s AI customer-facing support agents category, ask any vendor directly how their product handles the moment a customer explicitly asks for a person — and how it’s measured internally, since “deflection rate” and “customer satisfaction” can move in opposite directions, exactly as Klarna found out in public. If you’re specifically weighing an agent-first vendor against an incumbent adding AI on top of existing tooling, our Intercom Fin vs. Ada comparison is a relevant next read.