In 2024, Klarna published a headline-grabbing announcement. It sounded almost unreal: an AI customer service assistant capable of handling two-thirds of all customer contacts. To put real numbers behind that claim, Klarna said the AI agent managed 2.3 million conversations. In other words, a chatbot that could replace the work of 700 people.
The deal of the century, or the worst decision of all time?
Klarna talked about $40 million in savings, and the financial market reacted as if the company had found the cheat code. “This is where you should invest if you want to win.”
They could not have been more wrong.
After the broader 2022 tech downturn, when several companies saw their valuations fall, including Klarna, leadership needed to prove that the business was truly a tech company and not just a call center with a fintech brand. After reducing the workforce by 7,000 employees between 2022 and 2024, and freezing hiring, Klarna had to show it could scale without adding costs. What better way than to roll out AI?
February 2024: “We implemented AI and changed customer support forever”
The message was clear: we are transforming messy, human operations into something smooth, cheap, and perfect. The goal was not only operational. It was also a narrative built to support company value and drive market confidence.
- Average calls dropped from 11 minutes to 2 minutes
- 24/7 customer support
- 35 languages
- CSAT…
CSAT was the big unknown. It would have been the most important number to share.
Less than six months later, CEO Sebastian Siemiatkowski shifted the narrative and admitted the truth: quality outcomes were terrible. That was the moment the company fully realized that humans are not so easy to replace, and that AI is not magic. AI can be a powerful complement inside large organizations, but it is still a complement.
And that is how a triumphant press release ended with Klarna hiring again.
Why Klarna’s AI automation story matters
This case is fascinating because it shows multiple failures happening at the same time:
- Management pressure clouding decision-making
- Blind trust in artificial intelligence
- The real value of human judgment inside processes
- Process automation that still does not work perfectly
- The rush to publish a “phantom” announcement to appear like something the company was not
- Accepting the mistake is more important than dying while keeping up appearances
How to succeed with an AI implementation
The key is to be objective.
Start with clarity on the KPIs you must hit to achieve the company’s OKRs. Keep the “north star” in sight. Fast, short-term decisions that only address a loud, urgent problem rarely lead to sustainable success.
Once your processes run consistently and operational errors are under control, then automation can bring you closer to your goals.
But here is the critical point: automation does not automatically mean “implement an AI tool.” That distinction is often the difference between success and disappointment.
Get the scale right
Understand the scale, understand the impact, and build a solution that fits the problem.
There is no point implementing something massive if the problem is not massive. A simple example: implementing SAP for accounting makes no sense if you are a team of 10. Something smaller, even a well-built Excel template with the right automations, might be enough.
Automate what actually matters
Before automating, ask: Does this truly affect the business?
The answer tells you how careful you need to be in implementation, and whether it is worth doing at all.
Get advice and invite disagreement
If you are not an expert, that is fine. But get input from someone who is, to design the best strategy possible.
Also make room for people who do not agree with you. Different perspectives attack problems better than a single point of view. Understand the perspective of those involved, and accept a pause when someone raises a legitimate concern.