Klarna Eliminated 3,104 Jobs. Zero People Were "Laid Off."

In December 2022, Klarna employed 6,011 people.

By September 2025, that number was 2,907.

That's 3,104 positions gone. More than half the company. And here's the part that should terrify anyone who writes labor policy for a living: not a single layoff notice was filed. Not one WARN Act report. Not one unemployment claim attributed to Klarna. According to every instrument the United States government uses to track displacement, nothing happened.

Three thousand people vanished from an employer's payroll, and the statistical infrastructure of the world's largest economy recorded zero.

The Trick Is There's No Trick

Klarna didn't find a loophole. It didn't need one. The mechanism is embarrassingly simple: stop hiring.

In any company, roughly 15–20% of employees leave per year — new jobs, relocations, retirements, burnout. It's called natural attrition, and at a company of 6,000, it means 900 to 1,200 people walk out the door annually without being pushed. At Klarna, when each one left, they weren't replaced. Their work was absorbed by AI tools, or redistributed to remaining employees, or simply declared unnecessary.

No pink slips. No severance negotiations. No outplacement consultants. No reporters at the door. The federal WARN Act — the Worker Adjustment and Retraining Notification Act of 1988 — requires 60 days' notice when 500 or more workers are let go in a mass layoff. But nobody was let go. People quit. People moved on. The positions dissolved behind them like footprints in wet sand.

From a legal perspective, 3,104 acts of displacement are 3,104 voluntary resignations.

What the AI Actually Did

Klarna was more transparent than most about where the bots went. In February 2024, the company launched an AI customer service assistant built on OpenAI that handled 2.3 million conversations in its first month — work that had previously required the equivalent of 700 full-time agents. Two-thirds of all customer service chats were automated. Resolution time collapsed from 11 minutes to under two. Repeat inquiries dropped 25%.

The marketing department went from 200 to 100. Image production that had cost $10,000–$50,000 per campaign photoshoot was replaced by MidJourney and DALL-E subscriptions running about $2,000 a month. Six-week production timelines became seven-day turnarounds. AI handled 80% of copywriting at a claimed 70% lower cost. The entire external agency spend was cut by a quarter.

An internal chatbot called Kiki answered 2,000 employee queries a day. Adoption hit 93% in communications, 88% in marketing, 86% in legal.

And the remaining employees? CEO Sebastian Siemiatkowski raised their average compensation from $126,000 to $203,000 — a 60% increase. Smart. Generous, even. Also: an extremely effective way to buy silence.

The IPO

Revenue per employee went from $175,000 to $1.2 million. Wall Street loves that metric. It's clean, it's up-and-to-the-right, and it photographs well in an S-1 filing.

Klarna listed on the New York Stock Exchange in September 2025 under the ticker KLAR. The narrative: AI-powered efficiency. Revenue had doubled from $1.05 billion (2022) to $2.72 billion (2025) while operating costs stayed flat. The company told investors — and the investors told each other — that this was the future of lean enterprise.

Year	Employees	Revenue	Rev/Employee
2022	6,011	$1.05B	$175K
2023	5,545	$1.92B	$346K
2024	~3,800	$2.41B	~$634K
2025	2,907	$2.72B	$938K

Beautiful chart. Clean story. And then the CEO said the quiet part out loud.

The Reversal

"We focused too much on efficiency and cost. The result was lower quality, and that's not sustainable."

That's Siemiatkowski, to Bloomberg, in May 2025 — four months before the IPO. Customer satisfaction had declined. The AI assistant was producing answers that were, in the words of one TechGig analysis, "robotic, confusing, or just off track." Complex cases — the ones that actually mattered — overwhelmed the chatbot. Early evaluations found the AI was "basically acting as a filter" to reach the human agents who'd been mostly eliminated.

So Klarna started rehiring. But not the way you'd think.

The new positions were described as "Uber-style" — fully remote, flexible-schedule, gig-economy customer service roles targeting students, parents, and rural workers. Not the full-time positions with benefits that were eliminated. The same function, restructured into something cheaper and less stable. Siemiatkowski framed it as innovation: "From a brand perspective, it's so critical that you are clear to your customer that there will always be a human if you want."

The stock, which debuted around $45, has dropped roughly 65%.

Q3 2025 showed a $95 million loss, up from $4 million a year prior.

The Playbook

Klarna isn't interesting because it's unique. It's interesting because it formalized a pattern that's already spreading through corporate America with the quiet efficiency of carbon monoxide.

The steps:

1. Announce a major AI investment. (The signal.)
2. Freeze all hiring. (No WARN trigger.)
3. Let natural attrition run — 15–20% per year. (No layoff optics.)
4. Replace each departure with AI tools. (Each individual departure is invisible.)
5. Celebrate revenue-per-employee gains. (Wall Street applauds.)
6. Raise remaining employee pay. (Internal loyalty purchased.)
7. If the quality collapses, rehire — as gig workers. (Cheaper, less visible, fewer benefits.)

Shopify CEO Tobi Lütke sent an internal memo mandating that no new role can be approved unless the requesting team can demonstrate that AI can't do the job. Performance reviews now include AI fluency. Duolingo declared itself "AI-first," cut 10% of its contractor workforce, and requires proof that automation can't handle a task before approving a hire. Amazon has cut 27,000+ positions since 2022; CEO Andy Jassy confirmed to investors the company intends "fewer people" as AI scales. BT Group announced 55,000 job cuts planned by 2030, with 10,000 attributed directly to AI.

None of these will generate a WARN Act notification. The mechanism is identical: attrition plus freeze plus replacement. The positions dissolve one by one, like sugar in water.

What the Law Can See

Not much.

New York became the first state to try. Effective March 2025, an amendment to the state's WARN notice form requires employers to disclose whether layoffs were "a result of or motivated by Technological Innovation or Automation." Governor Hochul directed the Department of Labor to enforce the requirement. It was a real policy response — the first in the country.

It wouldn't have caught Klarna. The amendment applies only to mass layoff events that trigger WARN in the first place — 50 or more employees in a single event. Klarna never had a single event. It had 3,104 non-events spread across 24 months.

The federal Warner-Hawley bill, still stuck in committee, would require quarterly reporting of headcount changes correlated with AI investment. That's closer to the right shape — it would capture the slow bleed. But even Warner-Hawley can't solve the attribution problem: when a marketing coordinator quits to go to grad school and her position is backfilled by DALL-E, is that "AI displacement" or "voluntary attrition"?

The honest answer is it's both. The honest problem is our legal framework can't hold both truths simultaneously.

The Productivity Question Nobody's Asking

A July 2025 randomized controlled trial by METR — an AI evaluation nonprofit — gave 16 experienced open-source developers 246 real-world coding tasks, randomly assigning them to work with or without AI tools (Cursor Pro with Claude 3.5 Sonnet). The developers with AI were 19% slower.

They didn't think so. Before the experiment, they predicted AI would make them 24% faster. After, they reported feeling 20% faster. The gap between perception and reality: 39 percentage points.

That's one study. But the pattern repeats. A GitClear analysis of 211 million changed lines of code across Google, Microsoft, Meta, and 25 major open-source projects found that AI-assisted development increased code volume by 10% while simultaneously driving code duplication up 8× and refactoring down 60%. The code got written faster. It also got worse. Google's own DORA report showed a 7.2% decrease in delivery stability as AI usage rose 25%.

MIT's NANDA initiative found that 95% of enterprise AI pilots fail to deliver measurable ROI. Bain called real-world AI coding gains "unremarkable" — 10–15% at best, "often offset by the time spent checking AI's mistakes."

Companies are cutting workers based on productivity gains that don't survive controlled measurement. The displacement is real. The productivity improvement justifying it may not be.

What It Actually Costs

Start with the 3,104 Klarna positions. Assume an average fully-loaded cost of $150,000 per employee (conservative for a Stockholm-headquartered fintech with offices in the US, UK, and Germany). That's $465 million in annual labor cost eliminated.

But the stock lost 65% of its value — roughly $3 billion in market capitalization, gone. The Q3 loss ballooned from $4 million to $95 million. Customer satisfaction declined enough that the CEO admitted it publicly and reversed course. The rehired workers came back in gig roles — same function, worse terms, no institutional knowledge.

From the company's perspective, the math may still work. Revenue doubled. From the workers' perspective, 3,104 stable positions became some smaller number of gig slots — and our measurement systems will record the gig hires as job creation.

The Bottom Line

Klarna proved three things. First: a company can eliminate half its workforce without triggering a single labor protection in any jurisdiction on Earth. Second: the AI-as-total-replacement thesis fails within 18 months when quality matters. Third: when the replacement fails, companies don't rehire on the same terms — they degrade employment quality and call it innovation.

The scariest part isn't what Klarna did. It's that every labor metric in the world recorded it as nothing.