Coca-Cola, Klarna and the Receipts on Generative AI in Creative Work

How CodoraTech is funded: CodoraTech is supported by advertising and, in some articles, by affiliate links. Where an article contains affiliate links, we say so at the top of that article.

Two companies have published enough detail about generative AI in creative production to argue with. The Coca-Cola Company has run AI-generated holiday advertising in consecutive years and has described the process publicly. Klarna issued a line-by-line account of what AI removed from its marketing budget. Both accounts come from the organisations that benefit from them, and neither has been audited from outside. Set against the criticism the Coca-Cola work attracted and the ceiling score a commercial testing panel gave the same spot, they still form the most substantial documented material available on this question.

Coca-Cola started before the hype cycle

Pratik Thakar, global vice president and head of generative AI at The Coca-Cola Company, told Marketing Dive that the work predates the public arrival of ChatGPT: “Pre-GPT launch, pre-hype cycle of generative AI, we actually started working with gen AI and we were working on a campaign called ‘Masterpiece.'” The programme has since involved a named set of partners including Stability AI, OpenAI, Microsoft Azure and Mimic.

The production description is more interesting than the tooling. For the AI Christmas campaign the company worked with three studios, in Los Angeles, San Francisco and Kuala Lumpur, and ran consumer research across North America, Europe and Latin America using at least three research agencies before launch. Thakar is direct about where the labour sits: “It’s not like you input one prompt saying ‘create an ad’ and it pops out an ad. A lot of creative decisions are all made by humans.” That is a company executive describing his own team’s contribution, but it is consistent with the headcount and timeline figures reported elsewhere.

Seventy thousand clips, five specialists, thirty days

TheWrap reported on 3 November 2025 that the 2025 AI Christmas advertisement required more than 70,000 generated clips, five AI specialists and a 30-day production timeline. Those are input figures, not savings figures, and it is worth resisting the temptation to read them as a cost reduction. Nobody has published what the same spot would have cost through conventional production, so the comparison that would make the numbers meaningful does not exist in public.

Thakar told the same outlet that the “craftsmanship is ten times better” than the previous year’s version, and that on AI in advertising, “you’re not going to put it back in.” The first is a subjective judgement from the person responsible for the work. The second is a prediction. Neither is a measurement, and an article that presents them as evidence of anything beyond the company’s intent has overstated its material.

A senior writer called it an imitation of nostalgia

The 2025 spot used animals rather than people, which TechRadar described as an explicit attempt to sidestep the uncanny-valley problem that attracted criticism the previous year. Writing on 4 November 2025, senior AI writer John-Anthony Disotto accepted the technical improvement and rejected the result: “the warmth and human connection that made the original 1995 spot so memorable is still missing,” and the advertisement “feels like an imitation of nostalgia instead of nostalgia itself.”

His conclusion was about brand risk rather than craft. Coca-Cola, he wrote, risks being “remembered not as a festive highlight but as a sign that the brand no longer cares.” This is one journalist’s assessment in one publication, clearly labelled as opinion, and it is cited here as exactly that. It matters because it is one of very few readings of this work produced by someone with no commercial stake in it.

An ad-testing panel scored the same spot at ceiling

System1 Group ran the 2025 “Holidays Are Coming” advertisement through its Test Your Ad Pro consumer panel and awarded it a 5.9-Star Rating, which the firm describes as “the maximum possible score an ad can get in the Test Your Ad model.” It added ratings of “Exceptional” on Spike, its short-term effectiveness measure, and on Brand Fluency.

Three limits apply to that score and none of them are hidden. System1 does not publish sample size, respondent demographics or testing parameters in that write-up. It offers no comparison against the 2024 AI advertisement or against Coca-Cola’s earlier non-AI holiday campaigns, which is the comparison a reader actually wants. And the firm itself notes that holiday advertising generally scores higher than ordinary advertising, so a ceiling score in December carries less information than the same score in March. What this is, precisely, is an ad-effectiveness firm’s panel score. It is not proof that the criticism was misplaced.

The honest shape of the story is that both readings are true at once. Critics found the work hollow and a commercial testing panel recorded a maximum score. Any version of this story that reports only one of those results is selling something.

Klarna’s 10 million dollars, issued by Klarna

The most specific published figures on generative AI in creative production come from a Klarna press release dated 28 May 2024. The company reported that AI had cut external marketing agency spend by 25 percent, a 4 million dollar run-rate saving, and reduced image production costs by 6 million dollars, contributing roughly 10 million dollars annualised, which it described as 37 percent of its total cost savings. Sales and marketing spend fell 11 percent in the first quarter of 2024.

For generative AI specifically the release cited 1.5 million dollars of savings in the first quarter of 2024, more than 1,000 images generated in that quarter, an image development cycle shortened from six weeks to seven days, and 80 percent of copywriting now involving AI. David Sandstrom, Klarna’s chief marketing officer, said: “AI is helping us become leaner, faster and more responsive to what our customers care about, leading to a much, much better experience.”

Every one of those figures is Klarna’s own, issued in a press release while the company was pre-IPO and actively promoting an AI narrative to investors. Several business outlets reported the numbers at the time; none published an audit of them. The six-weeks-to-seven-days production figure sounds like the most checkable claim in the set, and it is still self-reported.

• Named generation tools in Klarna’s account: Midjourney, DALL-E and Adobe Firefly
• Named refinement tools: Topaz Gigapixel and Photoroom
• Reported image development cycle: six weeks reduced to seven days
• Reported share of copywriting using AI: 80 percent

Every number here was published by a party with an interest in it

Assembled in one place, the evidence base for generative AI in creative work looks thinner than its volume of coverage suggests. Coca-Cola’s process and production figures come from Coca-Cola executives and from trade outlets reporting what those executives said. Klarna’s savings come from Klarna’s investor communications. The only two readings from outside are a journalist’s opinion piece and a commercial testing panel that does not disclose its methodology.

One finding does survive across both accounts, and it happens to be the least promoted. Coca-Cola’s own head of generative AI states that humans made the creative decisions, and the production figures back that up: five specialists, three studios, thirty days and three research agencies is not a description of automated advertising. Klarna’s fastest claimed gain is a shortened iteration cycle rather than a removed step. On the published evidence, generative AI in these two organisations changed the speed and cost of producing options. It did not replace the judgement about which option to run.

Sources: Marketing Dive · TheWrap · TechRadar · System1 Group · Klarna