Writing

• AI in Marketing
• 2 October 2026

AI in marketing case studies: verified examples, including the UK

The AI in marketing case studies worth reading are the ones with a named brand, a named agency, a clear description of what the AI actually did, and a result you can check. On that test, the strongest recent examples are O2’s Daisy scambaiter (VCCP, UK), Klarna’s AI content operation, Unilever’s product digital twins, Cadbury’s personalised Diwali ads and Heinz’s A.I. Ketchup, plus a couple of instructive failures.

The UK example that sets the bar is Daisy, an AI “granny” built for O2 by VCCP and its AI agency faith. According to Marketing Week, she kept more than 1,000 scammers on the phone, some for up to 50 minutes, earned more than 1,800 pieces of coverage, and satisfaction with O2’s fraud efforts rose 10 percentage points, from 24% to 34%.

What makes an AI in marketing case study worth trusting?

Most of what turns up when you search “AI in marketing case study” is a listicle of brand names with one flattering number each and no link to where the number came from. We have all shared one in a team Slack channel on a Friday without opening it. Every example below had to pass this filter:

• The brand and agency are named, and you can find the work.
• It says which part of the job the AI did and which part people did.
• The result has a baseline. “50% cheaper” means nothing until you know cheaper than what.
• You know who is reporting it, because a press release, an awards entry and a vendor story each have their own incentives.
• It says whether the outcome was efficiency (cost, speed, volume) or effect (sales, brand, behaviour).

Case studies on successful AI adoption in UK marketing agencies

UK agency adoption is broad and shallow. A December 2025 survey of 53 agency professionals, run by Mustard Research, twisted loop and Prolific North and mostly covering small and mid-sized independents in the North, found that 9 in 10 use AI but only 2% feel “very prepared”, 70% named efficiency as the top use case, and 17% had built bespoke tools (Prolific North). The national picture is similar: the Office for National Statistics’ Business Insights and Conditions Survey (released 20 July 2026, 38,637 responding businesses) puts self-reported AI use in UK firms with 10 or more staff at around 35%, up from around 12% in late 2023, and at 58% in the information and communication industry.

The Prolific North survey put 10-15% of respondents in the group developing real capability, and the first two agencies below show what that looks like. The third is a warning.

VCCP and faith: O2’s Daisy the AI scambaiter

VCCP, the London advertising agency, launched faith as a dedicated generative AI agency in May 2023 with a team of 14 creatives, technologists, prompt engineers, strategists, data analysts and makers (LBB Online). Its first year mixed campaigns for Virgin Media, Cadbury, easyJet and the Co-op with tools; for O2 it built the Bubl Generator, which creates imagery of O2’s brand character, and the Copy Checker, which keeps copy on tone, both live by March 2024.

Daisy launched on 14 November 2024. VCCP’s own case page describes her as a stack of generative models: a trained large language model for personality and knowledge, a custom-trained diffusion model for the photo-realistic face, and an image-to-video model for animation, with a voice modelled on a VCCP staff member’s grandmother. Marketing Week reports that she took almost a year to develop, was trained on anonymised scammer data with the online scambaiter Jim Browning, and got onto scammers’ lists through “number seeding” and “breadcrumbing”. It won Best Use of AI at the 2025 Marketing Week Awards.

What I would take from it is the brief. The AI was the product and the story, pointed at a problem O2 already owned (it blocks millions of fraudulent calls and texts a month), so the coverage did a job no stock AI-generated ad could. Satisfaction with O2’s fraud efforts is a brand metric, which puts it in a small club of AI case studies reporting something beyond cost.

faith’s one-year review is refreshingly unglamorous. Alex Dalman, VCCP’s head of social and innovation and managing partner of faith, describes a long run of “yes, but” conversations with clients: “Yes, but the training data needed to be up to scratch. Yes, but the clients’ legal teams needed to be comfortable with the risks.” The biggest practical barrier was brand data quality, especially for brands with inconsistent visual identities or ones that rely on showing real people.

Croud: custom assistants inside the agency workflow

Croud is a media agency founded in the UK in 2011, working across performance and brand. It appears in Google Cloud’s list of real-world generative AI use cases (1,302 entries at the 22 April 2026 update), which puts it at more than 650 employees and describes it building custom Gems, Google’s configurable Gemini assistants, for email sentiment analysis, complex data analysis, coding help and supplier-specific data workflows, with 4-5x productivity gains on certain tasks.

Google chose this story, and “certain tasks” is doing heavy lifting in that sentence, but it is a realistic picture of daily agency use: repeatable analysis jobs turned into assistants an account manager can run without queueing for the data team. If that is the kind of adoption you are after, the guide to building an AI marketing workflow walks through how to scope one.

WPP: a warning about treating spend as adoption

WPP, the UK-based advertising group, has made AI its central pitch. In March 2025 its then chief executive Mark Read announced “increased annual investment of £300m in AI and technology” through WPP Open, its AI operating system, alongside an investment in Stability AI, the company behind the Stable Diffusion image model. By October 2025 the new chief executive, Cindy Rose, was calling recent performance “unacceptable”: Q3 revenue of £3.3bn, down 8.4% year on year, and full-year guidance cut to a 5.5% to 6% decline (The Drum).

That does not prove AI failed at WPP, since plenty else went wrong in 2025, but it is a reminder that £300m a year on a slide describes what went in, and the case studies worth copying describe what came out for a client.

Which brands have the best-documented AI marketing results?

These are brand-side examples, and they cover the main types of AI in marketing: production, personalisation and AI as the creative idea itself.

Klarna: the AI content factory

Klarna, the Swedish buy-now-pay-later company, said in a 28 May 2024 press release that sales and marketing spend fell 11% in Q1 2024, with AI accounting for 37% of those savings, which it put at $10 million a year annualised. Image production costs fell $6 million, using Midjourney, DALL-E, Adobe Firefly, Topaz Gigapixel and Photoroom, and Klarna says it cut the image development cycle from six weeks to seven days, used AI in 80% of copywriting, and cut external agency spend by 25%.

It is one of the most specific workflow case studies in public, and it is entirely about cost and speed. Nothing in it tells you whether the cheaper, faster campaigns sold more, which is the question your finance director will ask about ten seconds after you show them the $10 million.

Unilever: digital twins for product shoots

Unilever, owner of Dove, Vaseline and TRESemmé, said on 18 March 2025 that it was building 3D digital twins of products with NVIDIA’s Omniverse platform and the creative technology studio Collective World, then using generative AI to put them into scenes. Unilever says product imagery is now two times faster and 50% cheaper to make than with traditional shoots. At brand level, TRESemmé in Thailand reported an 87% cut in content creation costs and a 5% increase in purchase intent, and the Beauty & Wellbeing brands reported up to 55% savings, 65% faster turnaround, double the click-through rate and attention held three times longer.

The reason this one works is that a product pack is the one thing a brand can specify exactly. The AI is not being asked to invent a human emotion; it is being asked to put a correctly shaped bottle of Vaseline on a correctly lit bathroom shelf, forty times, in different formats. It is also rare in reporting effect metrics next to the savings, though the release does not say how they were measured.

Cadbury India: personalisation at silly scale

For Diwali 2021, Cadbury Celebrations in India worked with Ogilvy, Wavemaker and the synthetic video start-up Rephrase.ai on “Not Just a Cadbury Ad”. Shop owners could generate an ad in which an AI recreation of the actor Shah Rukh Khan named their shop. According to Mediabrief, 139,000 versions were created, 105,000 users generated ads on the campaign site, the work earned 30 million impressions, view-through rate was 22% higher and click-through 18% higher, and it won a Titanium Lion at Cannes (reported in June 2022).

For UK marketers the transferable question is consent. Any campaign built on a real person’s synthetic face and voice needs that person’s agreement and a legal team comfortable with the terms, which is exactly the thing faith’s clients kept raising.

Heinz: A.I. Ketchup

Heinz and Rethink, the Toronto agency, asked DALL-E 2 to draw ketchup, noticed the results looked like Heinz, and made that the campaign. Rethink’s creative director Zachary Bautista told The Drum it went from pitch to release in less than a month. The agency’s case film, published on Ads of the World, claims more than 850 million earned impressions, earned media worth over 2,500% more than the media investment, and engagement 38% higher than past campaigns. Those are awards-entry numbers, but the idea is neat: the AI’s output becomes proof of how completely Heinz owns the category.

Which AI marketing case studies went wrong, and why?

Coca-Cola’s 2024 remake of its 1995 “Holidays Are Coming” Christmas ad, made with the AI studios Secret Level, Silverside AI and Wild Card, is the best-known backlash. Forbes collected the complaints, from truck wheels that do not turn to oddly proportioned background figures. Secret Level’s founder Jason Zada said they ran a single squirrel through the AI “a couple hundred times”, which is a fairly expensive squirrel for a process sold on efficiency.

The best evidence on the wider question comes from Ipsos, working with Adam Peruta and Carrie Riby of Syracuse University. They took 20 real ads, rebuilt 10 with AI (Google Gemini to reverse-engineer the briefs, OpenAI’s Sora 2 for the shots) and tested them on 3,000 US consumers. Only 25% of viewers were confident the AI ads were AI-made, yet the human-made ads were 14% stronger on short-term effectiveness and 17% stronger on long-term brand equity. That is the gap in nearly every success story above: the AI saves money on the making, and any cost shows up in effectiveness, which most case studies never measure and which you find out about a year later when the brand tracker dips and we all blame the category, the pricing, the media mix and probably the weather.

AI in marketing case studies compared

Read down the “type of metric” column: only O2 and Unilever report anything about how customers responded to the brand, and neither publishes the method. For your own business case, the guide to measuring the ROI of AI in marketing covers how to separate time saved from effect, and the AI in marketing statistics page has the wider survey numbers.

Where to find case studies on AI workflows used in digital marketing campaigns

• IPA Effectiveness Awards papers. The IPA set the awards up in 1980, entrants have to show payback on marketing spend, and WARC hosts all the case studies. They are the most rigorous source on effect, so check here first for any AI-assisted campaign claiming a sales result.
• Marketing Week Awards and The Drum Awards write-ups, good for UK work though the numbers are entrant-supplied.
• Cannes Lions winners, including the Titanium Lion that Cadbury’s work won. These are strong on the idea and usually thin on method.
• Agency case pages and interviews, such as VCCP’s Daisy page and LBB Online’s faith retrospective, which give you the tech stack and process, told by the people selling it.
• Vendor customer libraries, such as Google Cloud’s use case list. Best for workflow detail (which tool, which task, which team), worst for independent results.
• Company press releases, as with Klarna and Unilever, where the numbers are specific but chosen by the company.
• UK industry bodies. IAB UK’s State of AI in Advertising report, summarised in the IPA’s June 2026 AI forum, found 75% of its members experimenting with agentic AI but only 4% describing themselves as “agent-first”, and 70% naming lack of skills as a barrier.

For a smaller-scale example, the ACTAsia AI in marketing case study shows what adoption looks like in a charity marketing team with no agency budget.

How to read an AI case study in five steps

1. Find the original source. If a blog quotes Klarna’s $10 million, check Klarna’s release, where it turns out to be annualised savings.
2. Write down the baseline. Faster than the old shoot, the old agency, or last year’s campaign?
3. Split efficiency from effect. Cost and speed go in one column, sales, brand or behaviour in another. If the second column is empty, you have learned something about cost and nothing yet about whether the work sells.
4. Look for the human part. Daisy needed a professional scambaiter, a real grandmother’s voice and almost a year of development; Unilever needed exact product data. That human part is what your team would have to copy.
5. Check the conditions. Legal sign-off, likeness consent, brand data quality and category rules (in pharma the ABPI Code and MHRA rules make fully generated claims copy a non-starter) decide whether a case transfers at all.

An AI assistant can do steps 1 to 3 if you paste in the source text; asking it to recall a case from memory is how invented statistics end up in decks.

Here is the full text of a marketing case study. Using only this text:
1. List every number, what it measures, and the baseline it is compared with.
   If no baseline is stated, write "no baseline".
2. Label each result as efficiency (cost, time, volume) or effect
   (sales, brand, behaviour, engagement).
3. List what humans did versus what the AI did.
4. Say who published this and what incentive they have.
Do not add any information that is not in the text.

FAQs

What is the best example of AI in marketing in the UK?

O2’s Daisy, built by VCCP and its AI agency faith and launched in November 2024, is the most complete UK example. Its AI grandmother, built from language, image and video models, wasted scammers’ time on O2’s behalf. Marketing Week reports more than 1,000 scammers engaged and satisfaction with O2’s fraud efforts rising 10 points to 34%.

Do AI-generated ads work as well as human-made ones?

The best recent evidence says not yet. Ipsos and Syracuse University researchers rebuilt 10 of 20 real ads with AI and tested them on 3,000 US consumers in 2026. Only 25% of viewers were confident the AI versions were AI-made, but the human-made ads were 14% stronger on short-term effectiveness and 17% stronger on long-term brand equity.

How much money has AI saved marketing teams in real cases?

Klarna reported in May 2024 that AI accounted for $10 million a year in annualised sales and marketing savings, including $6 million less on image production and 25% lower external agency spend. Unilever said in March 2025 that digital twins made product imagery twice as fast and 50% cheaper. Both are company-reported cost figures with no sales result attached.

How are UK marketing agencies adopting AI?

Widely but shallowly. A December 2025 survey of 53 UK agency professionals by Mustard Research, twisted loop and Prolific North found 9 in 10 use AI, but only 2% feel very prepared and 17% have built bespoke tools. VCCP’s faith unit and Croud’s custom Gemini assistants are examples of the minority building real capability.

Are awards entries reliable AI marketing case studies?

Use them with care. Awards entries, such as Heinz’s A.I. Ketchup claim of 850 million earned impressions, are written by the agency to win, so numbers are selected and rarely independently checked. IPA Effectiveness Awards papers are the most rigorous because entrants must show payback on marketing spend, and WARC hosts them.

Sources

1. Marketing Week, “How O2’s AI granny stopped 1,000 scammers targeting the public”, 20 January 2025
2. VCCP, “O2 takes on scammers with Daisy, the AI granny scambaiter”, November 2024
3. LBB Online, “One Year of faith: Lessons from VCCP’s Gen AI Agency”, 17 October 2024
4. Prolific North, “New AI benchmark reveals UK agencies are ‘all in’, but only 2% feel prepared”, 12 December 2025
5. Office for National Statistics, “Artificial intelligence in UK businesses: 2023 to 2026”, 20 July 2026
6. Google Cloud, “1,302 real-world gen AI use cases from the world’s leading organizations”, updated 22 April 2026
7. Croud, “About us”, accessed October 2026
8. WPP, “WPP announces investment in Stability AI and new partnership to shape the future of media and entertainment production”, 5 March 2025
9. The Drum, “Cindy Rose outlines 4-point turnaround plan as WPP posts ‘unacceptable’ Q3 results”, 30 October 2025
10. Klarna, “AI helps Klarna cut marketing agency spend by 25% and run more campaigns”, 28 May 2024
11. Unilever, “Unilever reinvents product shoots with digital twins and AI”, 18 March 2025
12. Mediabrief, “Rephrase AI’s NotJustACadburyAd featuring Shahrukh Khan wins Titanium Lion”, 29 June 2022
13. The Drum, “Rethink shares secret sauce behind Heinz AI campaign that won at The Drum Awards Festival”, 8 February 2024
14. Ads of the World, “Heinz: A.I. Ketchup” (agency case study), May 2023
15. Forbes, “Coca-Cola’s AI-Generated Ad Controversy, Explained”, 16 November 2024
16. Ipsos, “AI Ads Are Good Enough, And That’s the Problem”, 18 May 2026
17. IPA, IPAi Forum June 2026, “IAB UK’s State of AI in Advertising report”, June 2026
18. WARC, “IPA Effectiveness Awards”, accessed October 2026