Klarna’s CEO thought he knew his customer service culture well enough to replace 700 people with it. He was wrong, and it took him eighteen months and a public reversal to find out.
In February 2024, Klarna announced its OpenAI-built assistant was doing the work of 700 full-time agents, handling 2.3 million conversations and cutting resolution time from 11 minutes to under 2. The workforce fell from 5,500 to 3,400. Hiring froze. CEO Sebastian Siemiatkowski said publicly that AI could already do the jobs humans do.
By 2025, he was rehiring. On Bloomberg, he put it plainly: “We focused too much on efficiency and cost. The result was lower quality, and that’s not sustainable.”
Why did the AI-first strategy fail?
Because Siemiatkowski built his strategy on the culture he assumed his company had, an efficient, replaceable transaction layer, rather than the one it actually had: a customer relationship carrying real complexity, dispute, hardship, nuance, that no dashboard was measuring until it was already gone. He put the gates around a zoo he had never actually walked through. The animals he thought he was containing were not the ones causing the problem, and by the time the real ones showed up, in complaint volumes and falling satisfaction scores, the cost was already public.
Klarna is not an outlier. Gartner surveyed 321 customer service leaders and found only 20% had actually reduced staffing because of AI, and it predicts that by 2027, half of the companies that cut staff for AI will hire them back, frequently under new job titles. The pattern is common enough to have a shape: assume, act at scale, discover the assumption was wrong once the numbers force the admission.
What does the alternative look like?
I have watched this play out from the inside too, running a company where my own engineering team is AI agents. The temptation to assume competence rather than verify it is constant, and it is never the AI that catches the gap first. It is always a person who actually looked.
I once watched a leader do the opposite on his very first days in the role. He had just taken over a large business unit in New York, and instead of sitting behind a desk on day one, he spent his first two days traveling to every satellite office, meeting the people who actually ran the operation. He was not checking a box. He wanted to know who was actually in the building before he decided what the building needed.
| Leader who assumes | Leader who checks | |
|---|---|---|
| First move in the role | Acts on the org chart, or on what worked somewhere else | Goes to see what is actually happening, in person, before deciding anything |
| When the strategy is wrong | Finds out publicly, at scale, after the damage is visible in the numbers | Finds out early, quietly, before it costs anything to fix |
| What the culture becomes | Whatever the assumption already was, reinforced | Whatever is actually true, worked with directly |
Neither leader was foolish. One skipped a step the other did not. The gap between them was not intelligence or intent, it was whether they checked what was actually there before they built on top of it.
Before you scale any AI decision, ask the harder question first: are you building for the culture you have, or the one you assumed?