Klarna Uses AI for Customer Service; Handles Two-Thirds of Chats in One Month


Klarna started using an OpenAI-based AI assistant for its customer service chat. They published a report outlining the better satisfaction rate as the AI did the job of 700 full-time agents.

A month ago, Klarna started using an AI assistant powered by OpenAI’s language models to offer customer support chat services. Two days ago, they published their findings. As expected, AI models are better at troubleshooting user queries. Here’s the full list of advantages the company talks about, as mentioned in its official blog post:

  • 2.3 million conversations, two-thirds of Klarna’s customer service chats.
  • Equivalent work of 700 full-time agents.
  • On par with human agents in customer satisfaction score.
  • More accurate in errand resolution with a 25% drop in repeat inquiries.
  • Average resolution time getting below 2 minutes compared to 11 minutes previously.
  • Available in 23 markets, online 24/7, and can handle 35+ languages.
  • Estimated to deliver $40 million in profit to Klarna in 2024.

It’s no surprise. The business community has already expected artificial intelligence models to outperform humans in rote tasks and customer service is one of the segments supposed to be hit the worst by replacements.

Note that Klarna isn’t a simple business. They offer financial technology such as payment processors and payment installment services for ecommerce stores. As such, their customers are ecommerce stores that have a dedicated technical team or person. The queries are bound to be complex as users try to figure out website connection issues or inquire about the more nuanced aspects of integrating Klarna’s tools into their stores.

Typically, a lot of queries can be handled by the same template in customer service. Many users come back with pretty much the same questions. Certain queries are asked much more than others. As such, an AI system with access to all that information is going to be faster in resolution.

With further fine-tuning such as how the model should talk to customers (like keeping responses short, having an informal touch, and so on) and proper training on previous customer chat data, these tools can become significantly superior to human beings.

In its 2023 report on “Generative AI could raise global GDP by 7%,” Goldman Sachs expects the “equivalent of 300 million full-time jobs [exposed] to automation.” Is that exposure happening right now? Well actually, this change is likely to have started long ago. For example, The Washington Post covered an Indian startup in October 2023. The CEO used good old ChatGPT to replace 27 customer service agents. That CEO had the following to say:

It was [a] no-brainer for me to replace the entire team with a bot which is like 100 times smarter, who is instant, and who cost[s] me like 100th of what I used to pay to the support team.

Suumit Shah, CEO of Dukaan

With training becoming even easier, startups can easily field LLMs to handle their customer support. These systems are often better as they seem to “care more” than your typical customer support rep. They place equal value on each query, don’t get frustrated or exhausted, and are online 24/7. Pretty much the same machine-vs-human debate. On the other end of the spectrum, a 5000-employee-strong company like Klarna is also expecting an additional $40 million in profits using these systems. And at the far end of this spectrum, a tech giant like Google is also eyeing similar replacements, though they have many more ethical battles to fight before they lay people off.

With every size of company in every industry having similar views about using AI to replace humans that can be easily replaced (especially customer support staff), the job market is going through a revolutionary shift. You’ll often hear arguments like humans won’t be replaced, but merely augmented, or we’ll work together with AI. That’s a pile of nonsense. I’m sure people during Napoleon’s reign also had a similar rejoinder when the Luddites wanted to stop automated textile equipment from taking their jobs. “But Monsieur Ravier, these new sewing machines might be fast, but they will merely augment you, so kindly don’t lead a sewing machine riot with anywhere from 150 to 200 tailors to destroy these machines, yeah?”

By Abhimanyu

Unwrapping the fast-evolving AI popular culture.