Mondelēz, the company behind brands such as Oreo and Cadbury, is preparing to make generative AI a significant part of its marketing production process. The company expects its new AI-powered platform to reduce advertising production costs by 30% to 50% and says it has invested more than $40 million in developing the technology.
The platform has been developed in partnership with global advertising group Publicis and consulting firm Accenture. The companies began working on it last year, with the goal of producing short-form commercials that are polished enough for television as early as the next holiday season. Mondelēz has even suggested that the technology could be used in advertising for the 2027 Super Bowl.
The strategy is not simply about spending less. Mondelēz also wants to shorten production timelines, allowing it to launch products, test marketing ideas, and respond to performance data much faster than before.
The Technology Is Already Being Used
This is not just a long-term experiment. Mondelēz has already begun applying the technology to real marketing campaigns.
In the United States, it has been used for social media videos featuring Chips Ahoy!, while in Germany it has supported short-form content for Milka. One example involves creating scenes in which chocolate flows across a cookie like a wave. The same product animation can then be combined with different backgrounds and visual elements to suit particular audiences.
Mondelēz also plans to introduce AI-generated content to Oreo product pages on Amazon and Walmart in the United States in November. The company expects to expand the approach to Lacta in Brazil and Cadbury in the United Kingdom as well.
This suggests that Mondelēz is moving quickly to test generative AI within retail media, where brands can directly measure product-page engagement and purchasing behavior. However, the company has drawn a clear line around the use of realistic people. For now, it does not plan to generate human faces or lifelike human figures for these advertisements.
Why Food Advertising Is Well Suited to AI Production
The shift comes at a time when food and snack companies are facing pressure from two directions. Tariffs and production costs are rising, while consumers are becoming more cautious about spending.
Marketing production is one area where companies have some flexibility to reduce expenses. Food advertising has traditionally been expensive because it often requires elaborate studio setups, carefully controlled lighting, specialist food styling, and extensive post-production work.
An AI-assisted workflow can potentially reduce or replace parts of that process through scene compositing, texture simulation, and digitally generated environments. A brand may still need real photography or footage, but it can reuse and adapt those assets without rebuilding an entire set for every new advertisement.
The greatest advantage may be the ability to produce many variations quickly. Different languages, backgrounds, headlines, aspect ratios, and platform-specific formats can be generated from the same core creative concept. That could significantly reduce the time and money required to localize campaigns for different countries and media channels.
It also makes continuous testing much easier. Instead of committing a large budget to one final version, marketers can test several combinations, study the results, and update the creative based on actual performance data.
Human Review Still Matters
As the production process becomes faster, safeguards and quality control become even more important.
Mondelēz has made it clear that people will continue to review and approve the final output. The company has also published internal standards governing how generative AI can be used in its marketing.
Those standards prohibit content that glorifies overeating, depicts vaping—including the use of electronic cigarettes—relies on emotionally manipulative language, or reinforces offensive stereotypes.
The message is straightforward: no matter how quickly or convincingly AI can produce a scene, people must remain responsible for protecting the brand’s standards around health, fairness, and responsible advertising.
Any team that has worked seriously with generative AI will recognize the importance of this principle. An image may look technically impressive while still containing misleading details, cultural problems, or messages that are inappropriate for the intended audience. Speed does not eliminate the need for judgment.
Other Major Brands Are Experimenting Too
Mondelēz is not the only consumer-goods company exploring AI-generated advertising. Kraft Heinz and Coca-Cola have also experimented with the technology.
Coca-Cola released an AI-generated holiday advertisement in 2024, but some viewers criticized the unnatural appearance of human expressions and emotions. The response showed that even as generative technology improves, creating scenes that feel emotionally authentic still requires careful quality control.
Mondelēz’s decision to avoid generating realistic human faces appears to reflect the same concern. AI is currently better suited to certain types of content than others. Product imagery, backgrounds, visual effects, and format variations are relatively practical applications, while convincing human emotion remains much more difficult to reproduce.
Four Practical Lessons for Marketing Teams
From a practical marketing perspective, Mondelēz’s approach offers four useful lessons: where to begin testing, how to organize production, which formats to prioritize, and how to maintain quality.
1. Start With Retail Media and Product Pages
Retail media and online product pages are sensible places to begin. Brands can create short looping videos, detailed product visuals, and multiple combinations of copy and backgrounds, then measure conversion rates and time spent on the page.
Because the creative content and sales data exist within the same environment, teams can learn more quickly which versions are actually helping customers make purchasing decisions.
2. Treat Creative Production as an Ongoing Process
Instead of building every advertisement as a one-time campaign, teams can establish a weekly or monthly production cycle. New variations can be introduced regularly, while underperforming versions can be removed without hesitation.
This creates a repeatable learning process and allows marketing budgets to be directed toward the creative ideas that perform best.
3. Design Short-Form Content for Multiple Channels
Short-form assets should be designed with television and connected TV in mind from the beginning. Resolution, frame rate, aspect ratio, and audio standards can be built into reusable templates.
Doing this makes it easier to move successful content from digital platforms to television without rebuilding everything from scratch, reducing both production time and additional costs.
4. Document the Rules
Prompt guidelines, prohibited-content lists, and review checklists should all be documented. Teams should also clearly define who is responsible for reviewing the output and who has final approval.
A structured review process helps maintain speed and consistency while protecting brand safety. Without it, faster production can simply create mistakes at a faster rate.
AI Is Changing Both the Cost and Speed of Marketing
Ultimately, Mondelēz’s decision is about reducing costs, but it is equally about increasing speed.
Organizations that can produce content more quickly and test it more frequently gain more opportunities to learn from data and improve their campaigns. Generative AI will not replace every part of creative production, nor will it remove the need for skilled marketers, designers, and reviewers.
What it can do is turn creative content into an operational asset—something that can be continuously adapted, tested, and improved rather than produced once and left unchanged.
Mondelēz appears to understand that the biggest gains will come from combining AI-powered production with human responsibility and a culture of constant experimentation. When those elements work together, lower costs and better marketing performance may become achievable at the same time.
Thank you for reading, and I hope you have a wonderful day!
This article is also available in Korean: Read the Korean version