Klarna AI Customer Service: Replacing 700 Agents — A 2026 Case Study | Blog | Perspective AI

TL;DR

Klarna's OpenAI-powered customer service assistant, launched globally in February 2024, handled 2.3 million conversations in its first month — work the company said was equivalent to roughly 700 full-time agents. Klarna reported the assistant resolved issues in under two minutes versus 11 minutes for human agents, drove a 25% drop in repeat inquiries, and was on track to add $40 million in profit in 2024. By mid-2024, Klarna had cut its total workforce from about 5,000 to 3,500, largely through attrition. In May 2025, CEO Sebastian Siemiatkowski told Bloomberg the company had cut too deep on humans and was reopening hiring for premium support roles — a public rebalancing that reframed the deployment as a lesson in AI-first scope, not a wholesale agent replacement.

The Klarna context: BNPL scale, support volume, and cost pressure

Klarna's customer-service problem was an extreme version of a problem every consumer fintech has. As of 2024, Klarna's investor materials reported 150 million active consumers and 2.5 million transactions per day across 23 countries. Buy-now-pay-later support is high-volume by structure: missed payments, refund disputes, merchant escalations, payment-method changes, account closures, and identity verification all generate inbound contacts. Multiply by 23 country-specific regulatory regimes and 35-plus languages, and you get the volume profile that pushed Klarna toward automation in the first place.

Klarna had also been under cost pressure since its 2022 down round, when its valuation fell from $45.6 billion to $6.7 billion. The company laid off 10% of staff that year and shifted to a stated goal of profitability ahead of a much-anticipated IPO. AI was not an experiment for Klarna — it was a P&L lever the CFO could underwrite.

The deployment also rode an unusually tight relationship with OpenAI. Klarna was an early ChatGPT plugin partner in 2023 and one of OpenAI's most-cited enterprise references in keynote material. That partnership context matters because most enterprises buying conversational AI in 2025 are doing so without that level of technical alignment — and the gap between what Klarna shipped and what an off-the-shelf vendor ships is real.

Inside the deployment: what the Klarna AI assistant actually does

Klarna's assistant launched February 27, 2024, and was built on OpenAI's models with custom integration into Klarna's product surfaces. According to Klarna's own announcement, the assistant covers a defined set of high-volume tasks:

Critically, the assistant is embedded in the Klarna app and web flow, not bolted on as a side chatbot — it has authenticated context about the customer's purchases, payment status, and history before the first message. Operationally, Klarna routes conversations to humans when the AI flags ambiguity, regulated topics, or escalation triggers. The publicly reported "two-thirds of all chats" figure is interaction volume; the residual one-third still requires human handling, and within the AI-handled two-thirds, an unspecified portion involves a final human review.

The headline numbers Klarna actually claimed

Klarna's first-month press release in February 2024 anchored every subsequent media cycle. The specific claims:

A few things to note for any leader benchmarking against these numbers. First, the "700 agents" figure is a workload-equivalence calculation, not a layoff count. Klarna did not fire 700 agents on the day of launch — most of the headcount reduction across 2023-2024 came through hiring freezes and attrition, not layoffs tied to the AI deployment. Second, "two-thirds of chats" is volume, and the volume distribution of consumer-fintech support is heavily skewed. Third, the customer-satisfaction parity claim has been contested.

What the press cycle missed (and why the nuance matters)

Three nuances reshape the Klarna story once you go past the press release.

  1. The CSAT-parity claim was thinner than reported. Klarna's announcement said customer-satisfaction scores were "on par" with human agents, but the company has not published the underlying survey methodology, sample size, or cut by issue type.
  2. The headcount reduction was real, just not as clean as the headline. Klarna went from roughly 5,000 employees in late 2023 to ~3,500 by late 2024, primarily through attrition under a hiring freeze.
  3. The 2025 reversal. In May 2025, Reuters and Bloomberg both reported that Klarna was reopening hiring for some customer-service roles. This was a scope correction, not a pullback on AI.

Reaction inside the industry

Klarna's announcement landed in February 2024 at the peak of the "agentic AI is going to gut support" narrative cycle, and it became the canonical reference for that argument. Within weeks, customer-service vendors were citing Klarna, public companies were name-checking it on earnings calls, and at least three major BPOs publicly reframed their AI strategies.

The skeptical read came faster than most narratives. Industry analysts pointed out that Klarna's deployment ran on a tightly scoped consumer-fintech use case with structured data, authenticated users, and a finite set of common intents.

If you're building a 2026 CX program against this evidence base, the more honest framing is that this is the shape of every AI deployment in support: massive efficiency gain on the volume tier, careful human reinvestment on the value tier.

Lessons for any company running consumer support

Six things every CX, CS, and product leader should take from Klarna's deployment cycle:

  1. Scope tightly before you scale broadly.
  2. Authenticated context is the unlock.
  3. Measure resolution, not deflection.
  4. Plan the human-AI boundary up front, not after backlash.
  5. Don't run support and research in separate stacks.
  6. Reinvest, don't just cut.

What conversational research can learn from Klarna's conversational support deployment

The same architecture that lets an AI assistant resolve a refund question can run a customer-research interview at the same scale. Most enterprise teams are still running discovery with a Typeform and a quarterly survey while their support stack absorbs millions of real conversations. That mismatch is the opportunity.

Perspective AI is built for that opportunity: AI interviewers that follow up, probe, and capture the "why" the same way a great human researcher would.

Conclusion

Klarna's AI customer service deployment is the most cited example of conversational AI working at consumer scale. The deployment proved the volume tier of consumer support can be absorbed by AI with authenticated context, tightly scoped intents, and the right routing rules. The 2025 rebalancing proved that "replacing agents" is the wrong objective and that premium support is where humans still win.