RuncheyResearch

AI Absorption Ledger / KLAR

Klarna

KLAR · Q2 2026 · reported 2026-08-18 · revenue $1.04bn

Assessment

Klarna's second quarter on the ledger says less about AI than the first. The call does not use the word in prepared remarks or in questions, the press release drops the revenue-per-employee figure it gave in Q1, and the earnings release carries two AI statements: operating leverage supported by AI-enabled productivity gains and cost discipline in the same sentence as a marketing ramp (claim c1), and a placement paragraph that adds a Shopping Search app in ChatGPT to Google Pay inside Gemini and says traffic from AI platforms to retailers grew sharply and converts at higher rates (claims c2 and c3). That direction is one of the quarter's steps: distribution through AI platforms moves from described to directional. Both savings channels read exploratory, as in Q1: the one clause that attributes anything to AI shares the credit with cost discipline and names no function, and the Q1 tells (a falling cost per unit on the support line, a stated revenue per employee) are absent.

The support line reverted toward volume. Customer service and operations was against a year earlier with GMV up ; its share of GMV moved from to , and the gap to the prior-year rate narrowed to from in Q1. The line grew faster than active consumers. That is the shape a reversal would take, and the sources neither confirm nor deny one; the support channel is not sized, since AI shares the credit with cost discipline.

Non-transaction operating expenses were , up , against revenue growth of : a narrower gap than Q1, with sales and marketing and share-based payments doing the growing. AI productivity in those lines is not sized, for the same joint attribution. Revenue through AI platforms is ballparked at on a share of GMV at a take rate of . The AI vendor bill is unmentioned for a second quarter.

Read together, the two quarters show a company whose AI disclosure has thinned to a clause, whose support cost line has stopped falling against volume, and whose only forward AI statement is about distribution on other companies' surfaces. Whether the support line's reversion is the human agents added back, seasonality, or scale cannot be read from the sources; the next quarter's line is the test.

Sized channels against the income statement, Q2 2026

1 of 4 channels sized

Each blue mark is one channel's dollars for the quarter; a bar is the range of an estimate. Grey marks are the company's reported lines. The distance between them is the point: how large the AI channel is next to the line it sits in.

New money and old money, Q2 2026

2 new2 expanded

Each channel is tagged once for whether its money existed before language models, from the company's annual report and call at the start of the period. A bar splits one flow's sized dollars by that tag. The incremental total is the part that would not be there without the models: a new channel counts in full, an expanded one only for what AI added, a relabelled one at zero.

Revenue arriving through AI$1.0mn to $31mn sized

new $1.0mn to $31mn

Incremental total $1.0mn to $31mnpoint $5.2mn

The sized total counts every channel the company credits to AI, including relabelled money that existed before language models and the ledger's own estimates for it. The incremental total counts relabelled channels at zero. A flow is split by layer where its dollars sit at more than one: end use, compute sold to builders, and hardware. The same dollar can be a buyer's spend, a cloud's revenue and a chipmaker's revenue, so the layers are never added together.

Paid for AI1 channel · 1 not sized

vendor bill

AI model and compute bill

Not sized

The call, the releases and the interim financial statements are silent on AI vendor costs for a second quarter, and the line they would sit in is not split.

inscrutabledisclosure: not mentioned· motive: exploratory· before LLMs: new

A second quarter of silence. Technology and product development moved from to with no composition given, while the company added a ChatGPT app to the surfaces it runs on (claim c2 is tagged to the distribution channel, not here, because it says nothing about cost). The counterparty is mixed: model providers for models and tokens, and the compute providers behind AI-specific compute.

Evidence: 0 quotes, 2 figures, 1 confound, 4 from before coverage

What the company pays outside vendors for models, tokens and AI-specific compute behind its customer service assistant, internal tools and shopping search. It would sit inside Technology and product development, which the interim statements do not split.

Why this motive

Carried from the prior quarter, read as exploratory: no source in the ledger mentions what Klarna pays for models or compute.

Before LLMs: new

A bill for models, tokens and AI compute has no equivalent before large language models. At the anchor Klarna already ran its support assistant on OpenAI models and gave engineers an AI copilot, but neither the prospectus nor the anchor call states what it paid; the cost would sit in Technology and product development, in FY2024, which the prospectus describes as personnel, hosting, software licenses and external service providers without a split.

“We announced a partnership with OpenAI in 2023 and in February 2024 launched our AI assistant powered by OpenAI to improve customer support.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“Technology and product development expenses primarily consist of personnel-related costs for technology functions as well as other expenses, including hosting, software licenses, external service providers, hardware costs and amortization of internally developed and acquired technology assets.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“Increasing use of AI and a focus on cost optimization is also supporting our ability to reduce our expenses, more than offsetting any associated increase in salaries and technology costs.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“As of June 2025, 80% of our engineers and data scientists connected their work software to the AI copilot that can create and review code (based on data exported from the AI copilot). Our legal teams use AI to expedite document review. We also use AI to minimize external vendor costs.”
Filing, business, 424B4 periodic report, 2025-09-10

Figures

  • Technology and product development expense (negative, as presented) · 2026-CQ2
  • Technology and product development expense, prior-year quarter (negative, as presented) · 2025-CQ2

What else could explain it

  • line composition: Technology and product development holds engineering personnel, software and hosting together; the interim statements do not split it.

By quarter

  • Q1 2026not mentioned · inscrutable · exploratory
  • Q2 2026not mentioned · inscrutable · exploratory

Cost displaced by AI2 channels · 2 not sized

customer support · cheap to verify

Customer service displaced by AI

Not sized

The only attribution is operating leverage supported by AI-enabled productivity gains and continued cost discipline together (claim c1), with no function and no measure of AI’s part (the methodology rule for AI named beside another cause), so the channel is not sized.

Matched line moved : the whole line, not this channel.

described, no sizedisclosure: described· motive: exploratory· before LLMs: expanded

The support line grew against , nearly in step with GMV and faster than active consumers; the gap to the prior-year cost per unit, , is a fraction of the prior quarter's. The release repeats the attribution of operating leverage to AI-enabled productivity and cost discipline with no function or number. Nothing separates AI’s part from cost discipline, so the channel is left unsized. The counterparty is mixed: the company’s own service and operations staff and outsourced service providers.

Evidence: 2 quotes, 5 figures, 6 confounds, 6 from before coverage

Consumer and merchant service contacts handled by the AI assistant and self-service instead of in-house or outsourced agents. The displaced cost is the reported line Customer service and operations.

Why this motive

The cost-per-unit gap of Q1 is nearly gone: the line’s share of GMV moved from to , almost flat, and the line grew faster than active consumers. What remains is one clause crediting operating leverage to AI-enabled productivity gains and cost discipline together (claim c1), which names no function. AI named with no measure of its own and no line moving because of it reads exploratory, as in Q1; the sources do not contradict each other, so unknown does not apply.

Before LLMs: expanded

At the anchor this money was the reported line Customer service and operations, support personnel and outsourced assistance: in FY2022 and in FY2024. The prospectus credits the AI assistant launched in February 2024 with of savings in 2024 and with handling of service chats, while also showing the line already falling in 2023 on cost optimization before the assistant existed, and it keeps a human representative on offer to every customer. The size is the change AI made, not the whole line.

“We announced a partnership with OpenAI in 2023 and in February 2024 launched our AI assistant powered by OpenAI to improve customer support.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“Customer service and operations expenses primarily consist of personnel costs for customer support functions and outsourced assistance to help with purchases, account management, returns and merchant disputes.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“This decrease was primarily driven by a decrease in customer service costs per transaction following continued improvements to the AI assistant launched in February 2024.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“In the year ended December 31, 2023, our customer service engagement increased as a result of our higher GMV. However, customer service costs decreased by $32 million and our outsourced functions and business operations costs decreased by $15 million as we made significant efforts to optimize and manage our costs.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“We continue to see very demonstrable value from Klarna's AI assistant. This has been reported before. We used to adopt it. As you can see, it used to do about 700 full-time jobs. Now it is doing about 853 full-time jobs of a saving of $60 million.”
CEO, prepared remarks, earnings call, 2025-11-18
“At the same time, appreciating that certain consumers may nevertheless prefer to interact with human representatives, we continue to offer all of our customers that option. This reflects our dual-track approach of combining broad and continuing implementation of scalable AI in our customer service with high-quality human support.”
Filing, mdna, 424B4 periodic report, 2025-09-10

Figures

  • Customer service and operations as a share of GMV · 2026-CQ2
  • Customer service and operations as a share of GMV, prior-year quarter · 2025-CQ2
  • GMV, year-over-year growth · 2026-CQ2
  • Active consumers · 2026-CQ2
  • Active consumers, prior-year quarter · 2025-CQ2

Reported line it is matched to

Customer service and operations moved from to while GMV grew ; at the prior-year share of GMV the line would have been only higher, against in the prior quarter. The line has reverted toward volume growth, which is the shape a reversal (human agents added back) would take; the statements attribute nothing in the line to AI or to hiring.

2026-CQ2: 2025-CQ2: 2026-CQ2: 2025-CQ2:

What else could explain it

  • operating leverage: Part of any gap is scale; this quarter there is little gap to explain.
  • fx: Currency added a small part of reported cost growth (claim c4).
  • line composition: Operations staff and outsourced service providers sit in the line with customer service; the split is not reported.
  • seasonality: The prior quarter carried servicing of peak-season originations; the release notes that for processing costs, and support contacts may follow the same pattern.
  • other: Klarna said before coverage that it was adding human agents back for quality; a line growing in step with volume again is consistent with that, and the sources do not say.
  • other: Continued cost discipline is credited in the same sentence as AI-enabled productivity for the same operating leverage (claim c1), and nothing separates the two.

Quotes

“Our IFRS non-transaction-related operating expenses were $419 million, up 16% year over year, as we invested ahead of peak season and ramped marketing around the World Cup in the U.S., supported by AI-enabled productivity gains and continued cost discipline.”
c1 · Filing, press release, 6-K earnings release, 2026-08-18
“Adjusted operating expenses on the other hand grew by 24% at the headline level, but only 22% on a like-for-like basis, with 2 percentage points of the increase in Q2'26 due to FX.”
c4 · Filing, press release, 6-K earnings release, 2026-08-18

By quarter

  • Q1 2026described · described, no size · exploratory
  • Q2 2026described · described, no size · exploratory

engineering · cheap to verify

AI-enabled productivity across technology, marketing and corporate functions

Not sized

Management credits operating leverage to AI-enabled productivity gains and continued cost discipline together (claim c1) and gives no measure of AI’s part (the methodology rule for AI named beside another cause), so the channel is not sized.

described, no sizedisclosure: described· motive: exploratory· before LLMs: expanded

Management states non-transaction operating expenses of , up , and attributes the leverage to AI-enabled productivity gains and cost discipline in the same clause as a marketing ramp. Revenue per employee, stated in the prior quarter's press release, is not given this quarter, and the call does not mention AI. Nothing separates AI’s part from cost discipline, so the channel is left unsized.

Evidence: 2 quotes, 11 figures, 5 confounds, 7 from before coverage

Employee time displaced by AI tools in product development, marketing and corporate functions, which management credits for operating leverage without naming a function. The lines it would show in are Technology and product development, Sales and marketing and General and administrative.

Why this motive

The revenue-per-employee figure of Q1 is not repeated, the call does not mention AI, and the only attribution left is the shared clause on AI-enabled productivity gains and cost discipline (claim c1), given in the same sentence as a marketing ramp. AI named with no measure of its own reads exploratory, as in Q1; the sources do not contradict each other.

Before LLMs: expanded

At the anchor the work sat in the reported lines Technology and product development (), Sales and marketing () and General and administrative () in FY2024, which hold the personnel cost of those functions. The prospectus reports full-time employees falling from at the end of 2022 to at the end of 2024 and attributes the reduction to AI and attrition, with revenue per employee rising from in 2022 to in the twelve months to June 2025. The size is the change AI made, not the whole line.

“Technology and product development expenses primarily consist of personnel-related costs for technology functions as well as other expenses, including hosting, software licenses, external service providers, hardware costs and amortization of internally developed and acquired technology assets.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“Increasing use of AI and a focus on cost optimization is also supporting our ability to reduce our expenses, more than offsetting any associated increase in salaries and technology costs.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“As of June 2025, 80% of our engineers and data scientists connected their work software to the AI copilot that can create and review code (based on data exported from the AI copilot). Our legal teams use AI to expedite document review. We also use AI to minimize external vendor costs.”
Filing, business, 424B4 periodic report, 2025-09-10
“The reduction in the number of full-time employees resulted from our strategic decision to reduce our overall headcount and drive operational efficiency by leveraging AI in our business and focusing on what really matters to our mission. We expect the number of employees to continue to decrease in future periods.”
Filing, business, 424B4 periodic report, 2025-09-10
“For example, our sales and marketing costs declined to $355 million million in the the last twelve months ended June 30, 2025 from $531 million in 2022, driven by AI implementation, cost efficiencies and process centralization, while we maintained a strong GMV and revenue growth.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“We are now at $1.1 million per employee, and we hope to continue to do that acceleration. Part of that is due to AI and just a focus on operational efficiency, which, not through layoffs, but through natural attrition, as we have not hired for a few years, has now led to the number of employees to shrink by about 47%.”
CEO, prepared remarks, earnings call, 2025-11-18
“As you can see, below transaction margin, the operating costs, we have no plans whatsoever to increase any spending there currently because of the efficiency gains that we're seeing from AI. We don't believe that hiring is the right approach at this point in time.”
CEO, qa, earnings call, 2025-11-18

Figures

  • Non-transaction-related operating expenses (operating expenses less processing and servicing, provisions and funding costs) · 2026-CQ2
  • Non-transaction-related operating expenses, prior-year quarter · 2025-CQ2
  • Non-transaction-related operating expenses, year-over-year growth · 2026-CQ2
  • Non-transaction-related operating expenses, as stated in the release · 2026-CQ2
  • Non-transaction-related operating expenses growth, as stated in the release · 2026-CQ2
  • Technology and product development expense (negative, as presented) · 2026-CQ2
  • Technology and product development expense, prior-year quarter (negative, as presented) · 2025-CQ2
  • Sales and marketing expense (negative, as presented) · 2026-CQ2
  • Sales and marketing expense, prior-year quarter (negative, as presented) · 2025-CQ2
  • General and administrative expense (negative, as presented) · 2026-CQ2
  • General and administrative expense, prior-year quarter (negative, as presented) · 2025-CQ2

Reported line it is matched to

Non-transaction operating expenses rose ( to ) against revenue growth of , a narrower gap than the prior quarter's. Sales and marketing rose from to on the World Cup campaign and general and administrative from to on higher share-based payments; the release attributes neither to AI.

2026-CQ2: 2025-CQ2: 2026-CQ2: 2025-CQ2:

What else could explain it

  • operating leverage: Fixed costs growing slower than revenue is ordinary for a fast-growing network.
  • one time item: Share-based payments rose year over year inside the IFRS lines, most visibly in general and administrative, the reverse of the prior quarter.
  • mix shift: Marketing investment ahead of peak season and around the World Cup raised the base (claim c1); media is not personnel and dilutes the personnel share assumed.
  • fx: Currency added a small part of reported growth (claim c4).
  • other: Continued cost discipline is credited in the same sentence as AI-enabled productivity for the same leverage (claim c1), and nothing separates the two.

Quotes

“Our IFRS non-transaction-related operating expenses were $419 million, up 16% year over year, as we invested ahead of peak season and ramped marketing around the World Cup in the U.S., supported by AI-enabled productivity gains and continued cost discipline.”
c1 · Filing, press release, 6-K earnings release, 2026-08-18
“Adjusted operating expenses on the other hand grew by 24% at the headline level, but only 22% on a like-for-like basis, with 2 percentage points of the increase in Q2'26 due to FX.”
c4 · Filing, press release, 6-K earnings release, 2026-08-18

By quarter

  • Q1 2026quantified · described, no size · exploratory
  • Q2 2026described · described, no size · exploratory

Revenue arriving through AI1 channel · $1.0mn to $31mn sized · $1.0mn to $31mn incremental

distribution

Volume arriving through AI platforms and agent checkouts

0.1% to 3% of the quarter’s revenue

Incremental total: counts in full.

our inferencedisclosure: direction only· motive: channel-defensive· before LLMs: new

A step from described to directional: the release says traffic from AI platforms to retailers grew sharply and converts at higher rates, and adds a Shopping Search app in ChatGPT to the placements inside Google Pay and Stripe Link. No level for Klarna's own volume through these surfaces is given. The size shown is the ledger's own: , a share of GMV at the quarter's take rate of , widened upward from the prior quarter for the direction and the new surface.

Evidence: 2 quotes, 1 figure, 3 confounds, 4 from before coverage

Purchases paid with Klarna that start on an AI platform or inside an agent flow: Google Pay inside Gemini and Search, Stripe Link inside agent checkouts, and the Shopping Search app in ChatGPT. Revenue is the take rate on that volume.

Why this motive

The release's own words, moving with where consumers search (claim c2), are the methodology's channel-defensive tell: integrations to stay present where customers now start. Higher conversion from AI-platform traffic (claim c3) is an offensive tell with no measured level; the less durable motive is kept.

Before LLMs: new

No equivalent existed before large language models, and the anchor shows no purchase volume arriving from an AI platform or an agent checkout. The prospectus and the anchor call describe the carriers these placements ride on (Google Pay and Apple Pay as wallet distribution, Stripe Link as a checkout partnership, a shopping search engine inside the Klarna app) as ordinary distribution with no AI surface involved.

“Digital wallets like Apple Pay and Google Pay are essential distribution channels for Klarna. These platforms are widely adopted by consumers and frequently used at physical terminals that support contactless Near Field Communication (NFC) payments.”
Filing, business, 424B4 periodic report, 2025-09-10
“The benefit now is that Klarna is the main provider of that, which means that everyone getting a Stripe Link gets the option to use our Buy Now Pay Later, which becomes an even faster way to reach all of the Stripe merchants that offer that already. All of this is just like different ways of distribution of Klarna.”
CEO, qa, earnings call, 2025-11-18
“In the three months ended June 30, 2025, on average, 47 million of our active Klarna consumers opened the Klarna app every month to track their purchases, use our budgeting and banking tools, or to shop taking advantage of our AI assistant, our specialized shopping search-engine, our offers, or our inspirational product catalog.”
Filing, mdna, 424B4 periodic report, 2025-09-10
“They're going to go from strategy to nostalgia because now AI agents will be able to move all of my so-called proprietary data, preference, and all of that between different providers.”
CEO, prepared remarks, earnings call, 2025-11-18

Figures

  • Revenue take rate (total revenue as a share of GMV) · 2026-CQ2

What else could explain it

  • bundling: Klarna inside Google Pay, Stripe Link or a ChatGPT app is one option inside a wallet or surface; volume that arrives through an agent is not separated from ordinary checkouts.
  • mix shift: Management attributes GMV growth to Fair Financing, the card and PSP distribution, and attributes the volume guidance cut to Germany and currency; no part of either is attributed to AI platforms.
  • other: The direction given (traffic to retailers grew sharply last holiday season) is about AI platforms' traffic to retailers in general, not Klarna's own volume through them.

Quotes

“We are also moving with where consumers search. Klarna's flexible payments are coming to Google Search and the Gemini app within Google Pay, and our AI-powered Shopping Search app is live in ChatGPT, putting our dataset of over 100 million products, and our payments, inside the world's largest AI surfaces.”
c2 · Filing, press release, 6-K earnings release, 2026-08-18
“Traffic from AI platforms to retailers grew sharply last holiday season and converts at higher rates.”
c3 · Filing, press release, 6-K earnings release, 2026-08-18

By quarter

  • Q1 2026described · described, no size · channel-defensive
  • Q2 2026direction only · our inference · channel-defensive · $1.0mn to $31mn