Klarna Fired 700 Workers and Replaced Them With AI. Now They Are Hiring Humans Again. Here Is What Went Wrong.

In late 2024, Klarna CEO Sebastian Siemiatkowski stood on stage and delivered one of the most quoted AI success stories in business. His company had replaced 700 customer service workers with AI. The chatbot was handling 2.3 million conversations per month. Response times improved 82%. The company froze all hiring and shrank its workforce by 40%.

Wall Street applauded. LinkedIn celebrated. The "future of work" crowd had their poster child.

Then it fell apart.

What Actually Happened at Klarna

The AI performed brilliantly on routine queries. Password resets, order tracking, refund requests: the kind of repetitive, high-volume work that AI excels at. For those tasks, the 82% speed improvement was real and measurable.

The problems emerged at the edges. When customers had complex issues. When the query did not match a standard pattern. When empathy was required. When the AI needed to know what it did not know.

Customers started getting wrong answers to nuanced questions. Complaints escalated with no human to catch them. The AI confidently provided incorrect information, because large language models do not have a built-in mechanism for saying "I am not sure about this, let me get a human."

Siemiatkowski's new line? "Cost was too predominant a factor."

That is corporate speak for "we broke it."

Now Klarna is quietly building an Uber-style freelance model to bring humans back at $41 per hour. The very people they celebrated firing are being recruited back, at a premium, to fix what the AI cannot.

The Pattern: Fire, Break, Rehire

Klarna is not an isolated case. The same pattern is playing out across industries:

Salesforce fired 4,000 people and replaced them with AI. The AI tools were not ready. Workflows broke. The people left behind were doing two jobs. Morale collapsed. This is a trillion-dollar tech company with the best AI engineers money can buy.

Block (Jack Dorsey) went from 10,000 employees to under 6,000. The official line was "AI can do their jobs now." The reality: they quietly eliminated diversity roles, policy teams, and entire departments that had nothing to do with AI. Even Sam Altman, the CEO of OpenAI, said companies are using "AI" as cover for standard cost-cutting.

Multiple tech companies announced AI-driven layoffs in 2025, only to post job listings for the same roles months later. The pattern is consistent: cut headcount, announce AI replacement, discover the AI cannot handle the full scope of the work, quietly rehire.

The common thread is not that AI does not work. It is that these companies used AI as a replacement strategy instead of an augmentation strategy.

Why "AI Instead Of" Fails and "AI Underneath" Works

The distinction is simple but critical:

"AI instead of people" means removing humans from a process and letting AI handle everything. This works for narrow, repetitive, well-defined tasks with low stakes for errors. It fails catastrophically for anything involving judgement, nuance, relationships, or consequences for getting it wrong.

"AI underneath people" means keeping humans in the process but automating the manual, repetitive work that slows them down. The human still makes the decisions. The human still handles the exceptions. The human still builds the relationships. AI just removes the administrative burden that was eating 60-70% of their time.

The data consistently supports the second approach:

• PwC found that companies achieving 340% ROI from AI recruitment tools were the ones where recruiters shifted to relationship-building and closing, not the ones that replaced recruiters
• One home care provider automated candidate screening across 296,000 candidates, saved 148,000 recruiter hours, and returned $3.29 million annually, without cutting a single recruiter
• Bullhorn's data shows firms using AI are 2x as likely to have grown revenue, and the growth came from doing more with the same team, not from doing the same with fewer people

The Recruitment Agency Parallel

Now apply the Klarna lesson to a recruitment agency.

The "AI instead of" approach: You automate candidate screening and fire the recruiters who used to do it. The AI processes CVs faster. It scores candidates against criteria. It generates briefs. But then a top candidate calls in with a question about the role. They get a chatbot. The chatbot gives a generic answer. The candidate hangs up. They call the next agency on Google. A human answers. They get placed in 10 days.

That candidate is gone. And so is the $8,500 placement fee. And the client relationship that goes with it.

The "AI underneath" approach: You keep your recruiters. You automate transcript processing, candidate scoring, and brief generation: the admin work that consumes 60-70% of their day. Your recruiters now spend their time on what they were actually hired to do: building relationships, understanding client needs, coaching candidates, and closing placements.

The same candidate calls in. A human answers immediately, because they are not buried in admin. The candidate gets placed in 10 days. The client gets a faster, better experience. The recruiter is happier because they are doing meaningful work instead of formatting documents.

Same AI. Same automation. Completely different outcome. The variable is not the technology. It is the strategy.

The Economics of Augmentation vs Replacement

The maths favour augmentation overwhelmingly:

Replacement model: Fire 3 out of 5 recruiters. Save $210,000 in salary. Lose $204,000 in placements due to reduced relationship capacity. Net saving after accounting for lost revenue: approximately zero. But now you have a team of 2 trying to handle a workload built for 5, morale problems, and clients noticing the service decline.

Augmentation model: Keep all 5 recruiters. Invest in AI systems that save each recruiter 15 hours per week. 75 additional hours per week redirected from admin to placements. Additional revenue from faster placements and reduced candidate loss: $300,000-$400,000 per year. The team costs the same. The revenue nearly doubles. Margins climb from 15% to 40-60%.

Klarna tried the replacement model and is now paying $41 per hour to bring humans back. The augmentation model would have cost less and delivered more from day one.

The 82% and the 18%

Here is the most important number from the Klarna story: 82%.

The AI genuinely improved response times by 82%. That number was real. For the routine, repetitive work, the 82% of queries that follow standard patterns, AI was objectively better, faster, and cheaper than humans.

The problem was the other 18%. The edge cases. The complex issues. The moments requiring empathy, judgement, and the ability to say "I do not know, but let me find out."

In customer service, the 18% means frustrated customers and reputational damage. In recruitment, the 18% means missed placements, lost candidates, and broken client relationships.

The winning formula is not choosing between the 82% and the 18%. It is letting AI handle the 82% while humans focus entirely on the 18%: the work that actually earns the fee, builds the relationship, and differentiates your agency from every other one in the market.

What This Means for Agency Owners

If you are hearing the pitch, "replace your team with AI, it will be cheaper," remember Klarna.

It is cheaper. Until your top candidate calls in and gets a chatbot. Until your client's campaign goes out with hallucinated data. Until the people left behind are doing three jobs and burning out.

The agencies winning right now are not replacing people with AI. They are giving their people AI systems that handle the admin so the team can focus on the work that earns revenue.

There is a massive difference between "AI replaced my team" and "AI made my team dangerous."

Klarna learned that lesson the hard way. The question is whether you will learn it from their mistake, or repeat it.