This audio is generated automatically. Please let me know if you have any comments.
Oreo cookie maker Mondelēz International has created a new generative artificial intelligence tool to help personalize ads for consumers while increasing engagement for several popular brands.
The snack giant, which operates Chips Ahoy!, Ritz and Perfect Bar, began working on a generative AI tool known as AIDA (AI + Data) two years ago and has spent more than $40 million on the technology to date. AIDA allows Mondelēz to produce marketing content faster and at lower cost, often providing the opportunity to personalize material for specific consumer groups.
However, while the initial cost is currently high, Mondelēz expects that the tool will reduce the cost of creating marketing content by up to 50%. Implementing it in other parts of the food manufacturer’s business could save the company even more in the long run.
Mondelēz, which launched its initial platform in July, is still learning where and how it can best use the technology across its globally spread snack portfolio. The tool is still being tailored to understand the complexities specific to each brand and how to keep advertising accountable by avoiding encouraging unhealthy behaviors such as overindulgence.
Jennifer Mennes, Vice President and Global Head of Digital Marketing and Strategy at Mondelēz, and Tina Vaswani, Vice President of Digital Activation and Consumer Data at the company, recently sat down with Food Dive to discuss AIDA and the role of artificial intelligence in food marketing.
This interview has been edited for brevity and clarity.
Food Dive: How long has Mondelēz been working on AIDA and why did the company believe it would be useful for its business?
male: Because the entry investment is quite high, we have been thinking very carefully about how to solve this problem. So making this a priority for our marketing organization and for Mondelēz as a whole requires us to think very carefully about the different types of features we need to build and get the most value back as quickly as possible, which has led us to consider which brands we should test first.
But ultimately, the decision is that the amount of content you need to produce to really satisfy your end-to-end marketing ecosystem to drive your personalization ambitions and drive high levels of engagement and conversion actually requires a whole different level of content volume. This was something that could not be achieved using traditional methods today. So we needed to find automated solutions (AI being one of them) to ensure we could engage with consumers at the fidelity and scale needed to improve our business.
It’s a helper. This isn’t a new strategy. This allows us to work faster and better.
optional captions
Permission granted by Mondelēz International
Vaswani: Part of the process was to reimagine how we currently do work and then see where introducing AI could actually help or improve efficiency. This is very important. Because we’ve all learned from our own experience that just applying AI on top doesn’t always yield the best results.
So even as we look at features, we’re very thoughtful and mindful of assessing whether they actually add value or if they’re actually just wasting more time for our engineers trying to develop technology that isn’t ready yet.
Are there any applications where AIDA has been particularly effective? Likewise, are there any parts that need a little tweaking or are better left to humans?
male: With the maturity of the technology, it is always difficult not to get ahead of the skis in terms of what it can actually offer and what the expected results are. We learn a lot every day, such as what we can produce for biscuits or cookies and what we can achieve in chocolate is very different.
We focus our attention on areas where we can drive at greater speeds and scale. But it’s a lot of experimentation on a very large scale, but every iteration, every prompt is an experiment to see how far we can push the system.
Like Oreo, we trained it only on the original black and white sandwich cookies. Our team can pull a golden Oreo out of the system without any training (AI system). They’re pushing the system to see how far we can go. To be honest, we need to break it until we can build it. How far can we push the system to create as much value as we can?
Are there any challenges in working with AIDA because of its focus on specific foods?
For men: The product that many of my CPG colleagues show is a bottle. Internal products are not shown. We present our products to convey the charm and impulse of taste. It’s about the product. It’s less about the packaging.
Therefore, to ensure product fidelity and keep taste appeal high, expectations for these models are much higher than simply showing a bottle of shampoo. It may not feel that different, but in real-world data training, the fidelity required is like night and day.
Vaswani: One thing we are learning in real time is that while AI holds a lot of promise, it comes with a huge responsibility to stay true to the quality of images that our brands and consumers are accustomed to. This is where we realize that AI still has a long way to go before we can fully implement it in the way we want.
Are there any areas where you’ve had to train after AI suggests things that don’t fit Mondelēz or a specific brand?
For men: Responsible AI is not just about trademarks and copyrights. We also have to adhere to our principles and don’t want to seem overly generous, so we can’t print out 18 cookies or 15 chocolates.
We make sure we also have brand rules around this, so we keep it within our own framework. We have not covered some of the more highly regulated brands, such as Halls (cough drops) or Belvita (biscuits). If you do that, you’ll build rules into the platform that say you can’t say this, you can only say this, so that when the system receives a message, it already knows those rules to make sure the output meets those requirements.
Nothing gets put on the market without a legal brief. We do not have an ad approval process. So when an asset is ready to hit the market, it has to go through the same manual legal process that it does today. Therefore, it is not automated and distributed to the market. This allows us to get there faster by ensuring we don’t inadvertently be a little more lenient or use language that will show up in legal review under our current process.
Are there any plans to introduce AIDA to other parts of Mondelēz’s business?
Vaswani: If you look at AIDA and its underlying infrastructure, it’s definitely scalable. What we’re working on is defining what the next two, three, four big value cases are and then understanding where they fit into the AIDA family and where there are new platforms to build on. We are very considerate and practical. We’ve built a scalable foundation, but how we scale it depends on what the next big value is, which determines how we scale. Current infrastructure.
