Founder and CEO Sebastian Siemiatkowski told Bloomberg the company is in the midst of a recruitment drive to ensure customers always have a human to talk to.

“[I]t’s so critical that you are clear to your customer that there will be always a human if you want,” he said. 

The apparent turnaround comes after years in which Klarna touted its all-in-on-AI strategy. A year ago, Klarna said that its chatbot, powered by OpenAI, does the work of 700 human agents and does it much faster. In December, the company revealed that it had hired no human workers for the previous year, with the CEO boasting that “people internally at Klarna are just rallying to deploy as much efficiency AI as they can.”

“I am of the opinion that AI can already do all of the jobs that we, as humans, do,” the CEO told Bloomberg in February.

It now seems that, while chatbots are cheaper, they’re just not as good as humans for some jobs, according to Siemiatkowski. 

“As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality,” he told Bloomberg this week. “Really investing in the quality of the human support is the way of the future for us.”

A Klarna spokesperson told Fortune the company was maintaining its policy of not replacing employees who leave, outside of hiring freelance customer-service agents for the company’s outsourcing division, noting, “we’re very much still AI-first.”

But when it comes to tempering AI expectations, Klarna is far from alone. Just one in four AI projects delivers on the return on investment it promised, according to a recent IBM survey of 2,000 CEOs. 

An even smaller portion, 16%, are scaled across the enterprise, the survey said. 

Despite this dismal success rate, companies are going all-in on AI, driven largely by the belief that everyone else is doing it. Nearly two-thirds of CEOs (64%) say “the risk of falling behind drives them to invest in some technologies before they have a clear understanding of the value they bring to the organization,” according to the study.

The tech-first-ask-questions-later approach has resulted in some memorable flubs. Air Canada once employed a chatbot that made up a refund policy when chatting with a customer; it was forced to give back $880 to the client despite trying to argue it wasn’t responsible for the bot’s actions. McDonald’s tried an AI-driven system for its drive-thrus for three years before ending the effort. During those years, the system made mistakes like trying to add bacon to an ice cream order and giving one customer an order of 260 Chicken McNuggets.

In Klarna’s case, evaluations by outside engineers suggest the bot wasn’t bad, just limited. 

Gergely Orosz, a tech writer and podcaster, tested the bot last year and found it “underwhelming. It recites exact docs and passes me on to human support fast,” he wrote on X, while praising the bot’s lack of hallucinations. 

“It acts basically as a filter to get to customer support,” he said.

Today, that customer support channel is hiring again: CEO Siemiatkowski said Klarna has two new agents acting as freelance customer support. It’s also piloting an “Uber type of setup” that lets the workers choose their own schedules and work from anywhere in Sweden, where Klarna is based, at pay starting at 400 Swedish krona ($41.17). Some of these agents, Siemiatkowski said, could even be customers. 

“We also know there are tons of Klarna users that are very passionate about our company and would enjoy working for us,” he told Bloomberg.

Update, May 9, 2025: This story has been updated with a comment from Klarna on AI strategy.