Klarna's Q1 2026 earnings — $1.0 billion in revenue, up 44 percent year-on-year, on a headcount of 2,831 full-time employees — produced the cleanest single data point in the European AI-native employer thesis: revenue per employee of nearly $1.4 million, four times the company's 2022 level and a figure no European fintech and few American ones can match. The number is not an accident of timing or a one-quarter anomaly. It is the audited output of a three-year workforce redesign that the company's CEO, Sebastian Siemiatkowski, has described in public with more numerical specificity than any other European tech leader — and that is now entering a second phase, in which the structural reduction is largely complete and a targeted rebuild of senior AI engineering and model operations roles has begun in Stockholm. What Klarna has built, and what it is now hiring to extend, is the first fully post-transition AI-native operating model at scale inside European fintech. The question for mid-2026 is not whether the model worked — the Q1 results settled that — but what rebuilding on top of it looks like, what it pays, and what it signals for the European fintech employers watching from Amsterdam, London, and Dublin.
What Happened
The headline numbers are now publicly reported. Klarna's workforce peaked at approximately 7,000 employees in mid-2022, declined to 5,527 by end-2022, fell to 3,422 at end-2024, and stood at 2,831 as of December 31, 2025 — the figure carried into Q1 2026 filings. Operating expenses declined 8 percent over that same period while revenue grew 104 percent. Adjusted operating profit in Q1 2026 reached $68 million, compared with $3 million a year earlier. Net income turned positive — $1 million — against a $99 million loss in Q1 2025.
The reduction was executed through a combination of mechanisms Siemiatkowski has detailed across multiple public appearances. A hiring freeze instituted in early 2023 allowed natural attrition to run without backfill. The OpenAI-powered customer service assistant, launched in February 2024 and described by Klarna as performing the work of 700 full-time agents at peak deployment, removed the single largest category of human workflow from the headcount. Internal SaaS consolidation — Klarna terminated its Salesforce contract and over a dozen other enterprise software subscriptions — eliminated the support and operations roles tied to those platforms. By September 2025, when the company listed on the New York Stock Exchange under the ticker KLAR at $40 per share, the transformation was the central argument in its investor prospectus: not a buy-now-pay-later consumer lender, but an AI-native financial platform with the operating leverage to prove it.
The post-IPO adjustment is the part of the story that arrived quietly. In May 2025, Siemiatkowski told Bloomberg that the aggressive AI replacement of customer service functions had gone too far in specific categories — that customer satisfaction data on complex and high-value interactions had deteriorated, and that Klarna was rebuilding a human service tier using a flexible remote-agent model targeting its own consumer base as a recruitment pool. That reversal was partial and structural: the AI-first model for high-volume, routine interactions was retained; the full elimination of human judgment on escalation and financial distress cases was not. The hybrid is now the operating baseline.
What is less discussed — and what ENTRA's tracking of Klarna's Stockholm role activity through H1 2026 shows clearly — is the parallel technical rebuild happening above the customer-facing layer. Klarna maintained more than 40 open AI infrastructure roles for extended periods through Q1 and Q2 2026. The role clusters break into four functional areas: ML platform engineering covering model serving, training pipeline infrastructure, and the internal tooling that connects Klarna's multiple OpenAI and fine-tuned model deployments into a coherent operating layer; applied LLM engineering for the customer-facing and internal workflow surfaces built on top of GPT-4-class and Mistral-class models; data engineering for AI feeding Klarna's credit-risk models and AI assistant with production-quality inputs; and AI observability and model operations, the fastest-growing category in H1, responsible for monitoring model drift, latency, and output quality across Klarna's live AI systems. The model ops function is where the post-IPO accountability lands: a publicly listed company running LLM systems across 119 million active consumers (Q1 2026 earnings) has audit and reporting obligations that a private company's internal AI use did not.
The EU AI Act dimension compounds this. Klarna's credit decisioning and BNPL eligibility tools fall within Article 6 and Annex III high-risk classification — AI systems used to assess access to financial services. The Annex III enforcement obligations — originally scheduled for August 2, 2026 but extended to December 2, 2027 under the provisional Digital Omnibus agreement reached by EU institutions on May 7, 2026 — require Klarna to maintain conformity documentation, post-market monitoring architecture under Article 72, and data quality governance under Article 10 for training data feeding those credit models. The timeline extension does not remove the compliance obligation; it shifts the acute hiring urgency from this quarter to 2027. The AI governance and compliance associate roles that have appeared on Klarna's Stockholm and Berlin careers pages since January 2026 are not reputational management hires. They are legal obligations with a deadline.
Why It Matters
The Klarna operating model at H1 2026 is not a cautionary tale and it is not a template. It is a data point — a specific company at a specific scale with a specific product — that is changing the reference frame for how European fintech employers think about workforce architecture.
The signal is in the ratio. At $1.4 million revenue per employee, Klarna produces more output per person than any other publicly listed European fintech and most American ones. Adyen, which generated approximately €2.0 billion in net revenue in FY2024 on a workforce of roughly 4,400 employees (4,354 per annual report), operates at approximately €458,000 (~$500,000) revenue per employee — less than a third of Klarna's Q1 2026 rate, though with a meaningfully different revenue model. Wise, at roughly 6,500 employees and £1.2 billion in revenue for FY2025 (per Wise FY2025 Annual Report), runs at approximately £185,000 (~$234,000) per employee. The comparison is not entirely clean — Klarna's BNPL transaction volume model generates different economics than Wise's transfer fee model — but the order-of-magnitude gap in labour productivity is real and observable by every CFO in European fintech.
The consequence is already showing in hiring posture. Adyen's 2026 plan called for 550 to 650 net new hires weighted heavily toward the US and specialised engineering — growth, not contraction. Wise has over 400 open roles, disproportionately in North American business development and compliance. Neither company has announced a Klarna-style structural reduction. But the Klarna productivity number is the number their boards are seeing, and the question it poses — how many of our current headcount categories could be restructured around AI augmentation — is being asked in Amsterdam and London in a way it was not being asked eighteen months ago.
For Stockholm specifically, the Klarna rebuild has a labour market effect that runs in the opposite direction from the initial reduction. The ML platform and model operations roles now open at Klarna are among the most technically demanding in Swedish fintech, and the compensation bands reflect that. Senior AI engineers at Klarna in Stockholm are currently being offered SEK 1.8M to 2.4M total compensation — base of SEK 950,000 to 1.3M (~€83,000–€114,000, ~$91,000–$124,000 at Q2 2026 EUR/SEK/USD rates) plus virtual stock options on a post-IPO equity structure that now has public price discovery. That band sits 30 to 40 percent above the Stockholm fintech ML engineering median and within 20 percent of what Spotify pays its senior ML engineers in the same city. The comp reset is pulling upward on Stockholm's entire AI engineering market, as Trustly, Anyfin, and the second-tier Swedish fintech cluster all recruit from the same KTH-and-Chalmers-trained pool.
The European fintech AI compensation thesis can now be stated with more precision than it could a year ago. Klarna's model does not close the gap to US frontier labs — a senior ML engineer at Anthropic or OpenAI receives total compensation in the range of $500,000 to $800,000, still three to four times Klarna's Stockholm band. But it has definitively closed the gap to what Wise and Adyen were paying their senior technical staff twelve months ago, and it has introduced post-IPO equity liquidity — a structural advantage over the VC-backed European AI startups where option timelines remain indefinite.
What's Next
Klarna's stated trajectory points toward 2,000 employees by 2030 — a further reduction of approximately 830 from the current 2,831. That number implies continued structural contraction in operational and support functions, offset by continued technical build in the AI engineering categories. The net headcount direction is still downward. The internal composition direction is still toward higher AI-to-headcount ratio.
The model operations function is the one to watch in H2 2026. Klarna's post-IPO obligations as a publicly listed, AI-native financial services company operating under the EU AI Act's Annex III classification create a sustained demand for engineers who can build monitoring infrastructure, generate audit-ready model performance documentation, and manage the post-market reporting cycle that the European AI Office's May 2026 conformity assessment guidance has now codified. The December 2027 enforcement date for Annex III high-risk systems (per the revised Digital Omnibus timeline) is not the end of that function's build-out — it is the deadline against which the build-out must complete. Model ops engineers hired into Klarna in H2 2026 will spend 12-18 months constructing the documentation and monitoring infrastructure that Klarna's first formal EU AI Office review will assess. That is not abstract regulatory compliance. It is the specific technical function that keeps a publicly listed AI-native fintech operating inside European law.
The peer fintech response will clarify over H2 whether the Klarna model is replicable or singular. Revolut, which secured its MiCA licence and reached a $75 billion valuation in late 2025, has the scale and the AI investment to attempt a similar structural transformation but operates across a more complex international regulatory patchwork than Klarna's primarily European and North American footprint. Adyen's Q1 2026 results showed revenue per employee climbing but from a lower base; the company's 2026 hiring plan reflects growth, not structural reduction, which suggests Adyen's board has not yet reached the Klarna conclusion. The structural question — at what revenue-per-employee ratio does a European fintech's board decide to restructure rather than grow — has Klarna's $1.4 million figure as its only empirical data point.
Siemiatkowski has framed the company's position in language that has circulated widely in the Swedish business press: that Klarna built the bank of the future, not the bank of the future's staff — a formulation that captures the distributional character of the transformation precisely: the institution has been rebuilt around AI; the workforce has been rebuilt around the institution. What Klarna is hiring now — in model ops, ML platform, applied LLM, and AI Act compliance — is the small, highly compensated, technically exacting team that maintains what the transformation built.
That is the European AI-native employer thesis at mid-2026: fewer roles, higher pay, tighter technical scope, harder credential requirements, and a regulatory compliance layer that is structurally unique to the EU. The Klarna model does not resolve whether that is the right way to run a financial services company. The Q1 2026 earnings release established, with some finality, that it works.