PepsiCo
AI adaptation of a global brand platform into hyper-local creative by market, paired with automated media buying
PepsiCo states that its AI-localized Lay's campaign in the Netherlands and Belgium reached 6.5 million unique users in just over a month with a 4 percentage-point increase in ad recall and a cost per thousand impressions around 30% below its typical campaigns, and that AI-powered video campaigns now account for more than 60% of its YouTube Ads activity.
Key points
- PepsiCo EMEA adapts its brand platforms into local creative with Gemini, then distributes them through automated media buying.
- The Lay's campaign in Benelux reached 6.5 million unique users in a month, with 4 percentage points more ad recall.
- That campaign's cost per thousand impressions came in around 30% below the brand's typical campaigns.
- AI-powered video campaigns account for more than 60% of PepsiCo's YouTube Ads activity.
Objective
Make a global brand sound like a local one, without multiplying production cost by the number of markets. PepsiCo runs the same creative platform across dozens of countries where the hook that works is not the same on either side of a border. The system has AI carry the listening work (understanding what makes sense locally) and the adaptation work (producing the variants), while automated media buying places those variants. The economic stake is cost per thousand: gaining local relevance has to show up in CPM and recall, otherwise localization is only overhead.
The deployment
For its Europe, Middle East and Africa region, PepsiCo built an advertising production chain where AI intervenes in two places: before creation, to understand what makes sense locally, and during it, to adapt the same brand platform into country variants. The best documented case is Lay's in Benelux, which had to adapt the global platform "Joy is a simple recipe". Since joy is an abstract notion, the teams had Gemini analyze responses from more than 6,000 consumers, feeding in native terms to surface the nuances specific to each country. The model brought out gaps between immediate neighbors: Dutch happiness expressed more through activities such as boating on the canals, Belgian happiness leaning toward cycling culture. The global platform was then adapted into about thirty personalized three-word creatives, built on a fixed structure (action, place, twist), producing hooks as specific as "koers, kasseien, demarrage" in Belgium and "bootje, Oude Delft, familie" in the Netherlands. Published result: 6.5 million unique users reached in just over a month, 4 percentage points more ad recall, and a cost per thousand impressions around 30% below the brand's typical campaigns. At regional level, PepsiCo states that AI-powered video campaigns account for more than 60% of its YouTube Ads activity, with YouTube CPM down around 25% year over year. Other components are cited at pilot stage: attention moment detection in video for Doritos in the UK, automated campaigns for Lay's in Spain, creative prediction for Lay's in Poland. What the case does not say: the effect on sales, and what the same localized campaign would have delivered without AI.
Results Proof B
Quantified platform case study: both sources are editorial publications by Google covering PepsiCo's work, one on the AI media ecosystem of the EMEA region (March 2026), the other on the Lay's campaign in the Netherlands and Belgium. The figures measure the system itself (reach, recall, CPM of the generated campaign, share of AI-powered video campaigns) and are attributed by name to PepsiCo's EMEA media lead. Honest caveat, which rules out a higher level: both sources belong to the same publisher, which is also the supplier of the technologies used, therefore an interested party. No corroboration by an independent press source was found at collection, and no PepsiCo document repeats these figures. PepsiCo's April 2026 press release on its Google Cloud collaboration was rejected as a source: it covers supply chain and internal tooling, not this marketing system, and using it would have manufactured a corroboration that does not exist.
How it works
Inferred typical approachThe internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.
The stack in detail
- llm Gemini Used to analyze consumer responses and extract cultural nuances by country, then to adapt the global creative platform into local hooks.
- plateforme Video reach campaigns Automated video buying format on YouTube. According to PepsiCo, these AI-powered campaigns account for more than 60% of its YouTube Ads activity.
- plateforme Performance Max Automated campaigns tested by Lay's in Spain as part of the same ecosystem.
- outil Peak Points Attention moment detection technology in video, built on Gemini, tested by Doritos in the UK.
How it runs, concretely
For ops teams-
1Collecting local insight marketing
Consumer responses are gathered by market, in the local language.
-
2Analyzing cultural nuance AI
Gemini analyzes the responses, with native terms as input, to surface what differs from one country to the next.
-
3Creative adaptation AI
The global brand platform is adapted into local variants following a fixed structure.
-
4Brand validation marketing
Variants are reviewed before distribution, since multiplying versions multiplies the chances of drift.
-
5Automated distribution AI
Creative goes out in video campaigns and automated buying, with geographic and contextual targeting.
-
6Reading the results data team
Reach, recall and CPM are compared with the brand's typical campaigns.
Consumer responses collected by market, which ground the local nuance. Without them, localization falls back on national cliches, the main risk of this pattern. On the media side, the signal is CPM and recall, which say whether the local variant earns its cost.
How your customers perceive this type of use
Sourced studiesUn ecart net separe les annonceurs des consommateurs : 77% des annonceurs voient l'IA positivement contre 38% des consommateurs (Yahoo/Publicis, 2024). Les mesures implicites confirment le rejet declare : en EEG, les pubs generees par IA produisent une activation memorielle plus faible que les pubs traditionnelles et sont decrites comme agacantes, ennuyeuses et confuses (NIQ, 2024). La disclosure a un effet ambivalent : elle augmente fortement la confiance quand elle est remarquee (Yahoo/Publicis), mais 27% des jeunes consommateurs disent faire moins confiance a une entreprise dont la pub est creee par IA (IAB, 2024).
Acceptance conditions
- Une disclosure visible : quand la mention IA est remarquee, la confiance globale envers l'entreprise augmente de 96% (Yahoo/Publicis 2024)
- Une qualite visuelle suffisante : les visuels IA de basse qualite augmentent l'effort cognitif et distraient du message (NIQ 2024)
Red lines
- Le contenu IA non declare puis identifie : 72% des consommateurs disent que l'IA rend l'authenticite difficile a etablir (Yahoo/Publicis 2024) et les marques utilisant des pubs IA sont plus souvent jugees inauthentiques ou non ethiques par les consommateurs que par les dirigeants (IAB 2024)
- Les mannequins et personnes generes par IA : 46% des consommateurs n'en veulent pas dans la publicite, l'inquietude premiere etant les standards de beaute irrealistes (Attest 2025)
Sources: Yahoo / Publicis Media (terrain Ebco) 2024 · IAB (avec Attest) 2024 · NIQ (NielsenIQ) 2024 · Attest 2025
How to replicate
Inference, not sourcedData prerequisites
- Collection of consumer verbatims by market, in the local language and untranslated
- A brand platform explicit enough to be adapted without dissolving
- A reference set of brand constraints that can be enforced against generated variants
- Comparative measurement by market between localized creative and standard creative
Org prerequisites
- A clear mandate between the central level and markets on who validates what
- A brand review capable of keeping pace with variant production
- A transparency policy on generated content where the format warrants it
- A media team able to read CPM per variant, not only per campaign
Possible stack
- LLM for verbatim analysis and production of hook variants
- Asset library and brand rules to frame generation
- Automated media buying with geographic and contextual targeting
- Recall and reach measurement by market
The plan, step by step
- Step 1Collect consumer responses in the local language on the brand territory.Deliverable: A usable verbatim corpus per market.
- Step 2Surface cultural gaps between neighboring markets rather than what they share.Deliverable: A list of situations and references specific to each country.
- Step 3Define a fixed creative structure that bounds the adaptation.Deliverable: A variant template, reproducible and controllable.
- Step 4Produce the variants and put them through brand review before distribution.Deliverable: A validated set of local creative.
- Step 5Distribute through automated buying and compare reach, recall and CPM with reference campaigns.Deliverable: A quantified reading of what localization gains, market by market.
First step: Collect consumer verbatims in the language of the market, without going through translation. The core of this case is not image generation, it is having given the model untranslated local material: that is where the gaps between neighboring countries come from, and those gaps are what make the hook credible.
Sources
- S1 Inside PepsiCo's AI marketing ecosystem: A 2026 strategic vision Interested party archive pending
- S2 Lay's recipe for personalisation: Bespoke insights to media placements Interested party archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.