Published September 16, 2024 · By CampaignsLive · Case Study
In mid-2023, Klarna — the Swedish buy-now-pay-later company — began publicly describing a restructuring of its marketing operation around AI tooling. The headline numbers attracted attention quickly. Klarna’s communications team named an AI customer-service deployment that, it said, did the work of seven hundred full-time agents. The company stopped hiring across multiple categories. By mid-2024 the marketing-focused version of the story had attracted a different set of numbers: the company’s then-CMO described an internal marketing org producing creative work, at production volume, with a fraction of the agency spend Klarna had historically paid.
The press coverage handled the story in two registers. The trade press read it as a milestone — the first major consumer brand to credibly restructure marketing around AI tooling. The general business press read it as a warning — a high-profile case of AI displacing knowledge work. Both readings were partial.
Eighteen months later, with more visibility into what the restructured operation actually looked like in production, the picture has more texture than either of those readings allowed.
What Klarna actually did, by the public reporting
Three things, by roughly chronological order.
The first, in 2023, was a customer-service consolidation. Klarna deployed an OpenAI-powered support system that handled an increasing share of customer queries that had previously been routed to human agents. The company’s public framing was that the AI handled queries at higher quality, faster turnaround, and at a fraction of the cost. The headline number — seven hundred FTE-equivalent — was the most-cited piece of the story.
The second, in late 2023 and through 2024, was a marketing creative consolidation. Klarna’s then-CMO David Sandström described an internal marketing organization that had pulled creative production in-house, replacing meaningful portions of the agency relationships the company had historically run. The internal team used a combination of off-the-shelf AI tools — primarily Midjourney and DALL·E for image work, OpenAI’s text models for copy — and reportedly produced an outsized share of the company’s marketing creative.
The third, also through 2024, was a marketing technology stack consolidation. The company simplified its martech vendor list significantly, citing internal tools built on top of LLM infrastructure as replacing the function of several SaaS subscriptions.
The aggregate public story was that AI had let Klarna do more marketing with less headcount, less agency spend, and less software cost. The trade press and the company’s own communications converged on a version of this narrative.
What was more complicated than the narrative suggested
Three things, in retrospect.
The first is that the AI customer-service deployment, while real and significant, did not perform as cleanly in its second year as the first-year headline suggested. Public reporting in 2024 indicated that Klarna had begun reintroducing human agents on the channels where the AI’s performance had been thinnest, particularly higher-complexity disputes and emotionally sensitive interactions. The seven-hundred-FTE figure remained directionally accurate as a productivity claim, but the operational picture was less clean than the headline.
The second is that the marketing creative consolidation was, in production reality, more partial than the public framing implied. The company continued to commission outside creative for its larger campaigns. The in-house AI work, by working accounts, handled the high-volume, performance-marketing end of the production stack — paid social variants, retargeting creative, localizations, A/B test sets. The brand-equity work remained partly external.
The third is that the cost savings, while real, were not entirely AI-attributable. The company was simultaneously running broader operational efficiency programs. Disaggregating “what AI saved” from “what reorganization saved” was harder than the public narrative allowed.
The point of these complications is not that the Klarna story was misleading. It is that the simple version of the story — AI replaced an agency, an agency relationship, or a category of work — was less useful than the more specific version. AI replaced certain categories of work and not others; the categories matter; the working pattern is generalizable but only with the specificity intact.
The pattern that emerged
The pattern visible from the Klarna case — and from the smaller versions of the same restructuring at other consumer-finance and tech brands through 2024 — has three components.
High-volume, low-equity creative moves in-house with AI tooling. Performance marketing variants, retargeting creative, localizations, social variants, e-commerce product imagery. The work that was historically the most expensive piece of the agency line-item, in the sense that it absorbed the most production hours per dollar of brand equity it generated.
Brand-equity creative stays partly external. The major campaign work, the year’s tentpole creative, the work that defines the brand’s visual identity for the next eighteen months. The work that justifies the agency’s strategic value beyond production.
Customer-service and support consolidates faster than marketing creative. The AI productivity case is structurally cleaner in support than in creative. Support has well-defined success metrics, narrow conversation surfaces, and clear escalation paths. Marketing creative has none of those things. The Klarna timeline, in retrospect, showed support moving twelve to eighteen months ahead of marketing creative in the AI adoption curve.
What brand-side teams should take from this
Two things.
The first is that the in-house creative restructuring Klarna ran is replicable, but only on the categories of work it actually applied to. A brand that tries to use AI tooling to replace its entire agency relationship is likely to find that the agency was doing a piece of work the AI cannot replace — usually the brand-equity-defining piece — and that this piece is the one the brand cared most about preserving.
The second is that the cost case has to be honest about what is being saved. AI-tooling productivity in marketing creative is real, but it is uneven across work types, and the simple narrative of “AI replaced X agency hires” obscures more than it reveals. Brands that have done well with this restructuring have been the ones that disaggregated the work, applied AI to the categories where the case is cleanest, and preserved the external relationships for the categories where the case is not.
For the related restructuring on the holding-company side, see WPP, Publicis, Omnicom: How the Holding Companies Bought Into AI. For the question of where AI fits in the production stack more broadly, see How CampaignsLive Works.