AI Marketing Case Studies: Real Numbers from Real Brands
The genre of "AI marketing case study" has degenerated into a kind of corporate fan fiction. Vendor blog posts list household-brand logos next to claims like "300% lift in engagement" with no methodology, no baseline, and no source. The pattern is so common that it has trained marketers to discount AI claims wholesale, which is a problem because some of the wins are real, well-documented, and actionable. The five cases below were chosen because each has a primary source — a press release, an earnings call, or an interview with a named executive — and because each carries a specific number that can be checked. The analysis afterward looks at what they have in common, which turns out to be more interesting than any individual case.
Table of contents
- Methodology and sources
- Klarna's customer service agent
- Heinz's AI ad campaigns
- Nestle's content scaling
- Unilever's research automation
- Spotify's recommendation evolution
- Common patterns in the wins
- Frequently asked questions
- The bottom line
Methodology and sources
For each case the original company source was prioritised — earnings transcripts, official press releases, blog posts attributed to a named executive, or interviews in named publications. Vendor case studies and unattributed anecdotes were excluded. Where the numbers are publicly contested, the contest is noted. The point of the exercise is not to declare which case is most impressive but to surface what is verifiable so other operators can compare against their own situation.
Klarna's customer service agent
Klarna disclosed in a February 2024 press release that its OpenAI-powered customer service assistant was, at one month after launch:
- Handling 2.3 million conversations per month, equivalent to two-thirds of the company's service chats.
- Doing the work that would otherwise have required around 700 full-time agents.
- Producing customer satisfaction scores comparable to human agents.
- Reducing repeat enquiries by roughly 25%.
- Resolving conversations in about two minutes versus 11 minutes for human agents.
- Forecast to drive a $40 million profit improvement in 2024.
The numbers were widely repeated, then partly walked back. By 2025 the company stated publicly that it was bringing some human agents back — not because the AI failed, but because customer demand for human interaction in higher-stakes cases warranted a larger human tier than the original deployment had assumed. That nuance is more interesting than the headline figure: even a successful conversational AI deployment finds its own rightful share of automation, which is rarely 100%.
For operators trying to match this case, the implementable lessons are: a single workflow scope (text-based service chat), a clear measurable baseline (CSAT, resolution time, repeat-enquiry rate), and a willingness to publish results that included costs as well as gains.
Heinz's AI ad campaigns
Heinz ran "A.I. Ketchup" in summer 2022, generating images of ketchup with various DALL-E 2 prompts ("ketchup as a Roman fresco," "ketchup street art") and discovering that almost any image-gen prompt about ketchup produced something that looked like a Heinz bottle — supporting the brand's longstanding "if you ask for ketchup, you mean Heinz" positioning.
The campaign won a Cannes Lions award and produced widely-reported earned media. Heinz said the campaign generated 1.15 billion earned impressions and a roughly 38:1 ROI on earned media value compared with paid spend, per the agency Rethink's own reporting after the campaign. The exact "ROI" figure is in the standard category of agency-self-reported metrics that should be read with scepticism, but the campaign is unambiguously real and the cultural moment was unambiguously big.
The implementable lesson: AI-generated imagery as the SUBJECT of a campaign, with the brand insight ("AI itself thinks ketchup means Heinz") doing the work, is a more durable use of the tech than AI as just a faster Photoshop. The moment passes; the insight stays.
Nestle's content scaling
Nestle has been more public than most CPG companies about its AI work in marketing. The company has stated in various 2023-2024 forums and interviews that it uses generative AI to scale content production across hundreds of brands and dozens of markets — turning a single creative asset into market-specific variants, generating product photography for ecommerce, and accelerating localisation.
The numbers Nestle has shared publicly are more about throughput than direct ROI: faster turnaround times on creative variants, reductions in production cost per asset, and increased volume of A/B test combinations possible. CMO Aude Gandon spoke publicly in 2023 about generative AI tooling supporting content production at "10x" the previous rate for some workflows, with the caveat that the figure varies by content type.
The implementable lesson: large brands win with AI in marketing not by making one impressive ad but by industrialising the long tail. Most brands have hundreds of markets, dozens of product variants, and thousands of asset versions where the marginal cost of human production was prohibitive. Generative AI changes that constraint, and the uplift compounds.
Unilever's research automation
Unilever has talked publicly about its use of AI for consumer research and creative testing. The company's internal tools, including its in-house AI assistant "Unily" and various third-party platforms, have been used to synthesise consumer research, generate first-draft creative concepts for testing, and accelerate the briefing-to-test cycle on new products and campaigns.
Specific numbers Unilever has shared: a roughly 30% reduction in content production cost on certain campaigns, faster concept-testing cycles (down from weeks to days for some workflows), and a stated ambition to make AI tools available to most of its global marketing workforce. As with Nestle, the press-release numbers are real but mostly throughput-oriented; the deeper bet is that AI removes the budget ceiling on how many concepts a brand team can test.
Spotify's recommendation evolution
Spotify has been one of the longest-running production examples of ML in marketing, with its recommendation system pre-dating the current generative AI wave by a decade. The 2023 launch of "DJ" — an AI host that introduces recommended music in a synthesised voice — used OpenAI's technology and Sonantic's voice cloning to create a personalised radio-host experience.
Spotify has stated that personalised recommendations drive a substantial share of total listening time on the platform — over 30% by various accounts, though the exact share varies by quarter and by how "personalised" is defined. The DJ feature, launched in early 2023, was credited in subsequent earnings calls with strong engagement among the user segment that adopted it. Discovery Weekly, a long-running ML-driven playlist, has been described by the company as one of its highest-engagement features.
Spotify is the closest thing in this list to a "marketing as product" case study: the recommendation system is both a marketing layer (nudging users toward content) and a retention engine. The numbers do not isolate "marketing ROI" in the conventional sense, but the pattern — personalisation as a daily-engagement driver — translates to most subscription businesses.
Common patterns in the wins
Across the five cases, four patterns show up consistently.
| Pattern | Why it matters |
|---|---|
| One workflow, deeply automated | The wins came from going deep on a specific job, not from broad "AI everywhere" deployments |
| Clear pre-AI baseline | Each company could state the metric the AI was meant to move (CSAT, content cost, time-to-test) |
| The brand insight does the heavy lifting | The Heinz campaign worked because the brand was already category-coded; AI just amplified the meaning |
| Numbers attributed to a named source | The cases that hold up under scrutiny are the ones where an executive went on the record with a specific number |
The cases that did NOT make this list are equally instructive. Several mid-2023 vendor case studies named major brands that, when contacted by journalists, either denied the work entirely or could not confirm the figures. The lesson for buyers: never make a budget decision based on a vendor case study without an attributed quote and a callable reference.
For deeper coverage of the operational side of these wins, see our extended brand case-study collection. For the strategy framing, the AI for business pillar covers ROI maths and project sequencing.
Frequently asked questions
Why are most AI marketing case studies unreliable?
Because the case-study genre is structurally promotional. Vendors write them to sell software; agencies write them to win awards; brands cooperate when the numbers flatter them and quietly ignore them when they do not. The unreliable cases tend to share three features: no named source, no baseline metric, and a percentage gain unanchored to absolute numbers. When all three are missing, treat the case as advertising.
What is the most replicable AI marketing win for a mid-market brand?
Content scaling. The Nestle and Unilever pattern — using generative AI to localise, vary, and test creative at higher volume — is the most accessible to brands that do not have Klarna's data infrastructure or Spotify's ML team. Tools like Adobe Firefly, Midjourney, and various enterprise creative-AI platforms have made the build cost low enough for marketing teams to pilot in a quarter.
Does AI-generated content hurt SEO?
Google has been clear that the question is quality, not provenance. AI-generated content that adds genuine value to a topic ranks fine; AI-generated content that is thin, derivative, and at scale gets penalised. The 2024 Helpful Content updates and subsequent core updates have been particularly hard on sites that scaled AI content without quality control. Use AI to accelerate human-edited writing, not to replace it.
How do I measure AI marketing ROI honestly?
Set a baseline before you start. Run the AI-augmented workflow alongside the existing one for long enough to produce reliable measurements, ideally with a controlled split. Include all costs — licences, integration, training, ongoing review — not just the visible vendor spend. Compare on the same metric you would measure any marketing investment on: CAC, conversion, LTV, or cost per qualified lead.
Are AI ad campaigns a brand-safety risk?
They can be. Image-generation tools are getting better at avoiding the most egregious failures (extra fingers, garbled text, copyrighted-character collisions), but the failure modes are different from human-created work. Build a review step into the workflow for any AI-generated asset that will appear in a brand campaign, and budget for the legal review of training-data and rights questions on platforms whose data provenance is murky.
Will AI replace marketing agencies?
It will replace the lower-margin parts of agency work — production, localisation, certain types of media planning — and put pressure on the rest. Agencies that survive will be the ones that move up the value chain to strategy, brand, and the higher-order creative work AI is still not good at. The next five years will reshape the industry; the agencies pretending otherwise will struggle.
The bottom line
The verifiable AI marketing wins of 2022-2025 share a profile: a single workflow, a clear baseline, an executive willing to put their name on the result, and a brand insight that the technology amplified rather than substituted for. The category is mature enough now that the operators getting real returns are not the loudest ones. They are running fewer, deeper projects, measuring honestly, and treating the tech as a multiplier on what was already working. The next era of AI marketing case studies will be more boring and more useful — the era of the percentage-gain victory lap is winding down. For broader strategy framing across business functions, see our AI for business hub.
Last updated: May 2026