Recently, we shared how Unilever has turned to 3D printing to speed up the time-consuming and expensive process of packaging development for products like dish soap—but that’s not the only operation it’s working to accelerate.
Along with a slew of home and personal care brands, the CPG giant produces food brands like Hellmann’s and Knorr, and as the company looks to add new food products—whether they be more sustainable or better-for-you formulations, vegan varieties, or just new flavors—Unilever has been looking to AI to make the product development process more efficient.
Manfred Aben, global VP of science and technology for nutrition and ice cream at Unilever, who oversees the company’s “longer-term breakthrough technology,” is leading the effort. Aben has been with the company for nearly 30 years, and was hired for his experience in AI, which looked a little different back then. The first AI models were used to assess products’ shelf lives. The AI of late is a bit more intricate, and has meant less IRL experiments and tests in the lab or pilot plants, he said.
“At that time, AI was very much about rule-based systems. So you have a number of rules and the system checks those rules for you,” he said. “Nowadays, if we talk about AI, we talk a lot about more data-driven models and data-driven analysis.”
Model behavior: The product development process at Unilever typically looks like this: It starts in the professional kitchen, with chefs testing out new ingredient and flavor combinations. Then, it tests the most successful ones with consumers, and after that determines how to make the winner a mass-produced product. The products will then be tested in pilot plants and eventually factories.
All those steps are still necessary, with or without AI, Aben noted. But without AI, there’s “less predictability,” he said.
“You have to do more trials to tweak it and find exactly the right setting of your processes, the right dosing of your ingredients, the timing of the process, etc.,” he said. “The more we can do that with digital modeling, the more effective we can be, the more successful we can be.”
Unilever also aims for its nutrition products to be “holistically superior,” Aben said, which includes being healthier for consumers, better for the planet, and affordable. Figuring all that out can be a “complex optimization problem” without AI.
Retail news that keeps industry pros in the know
Retail Brew delivers the latest retail industry news and insights surrounding marketing, DTC, and e-commerce to keep leaders and decision-makers up to date.
“Having these tools at our disposal just opens up massively our toolbox in the whole development cycle from consumer insight to final product,” Aben said.
One of Unilever’s innovations using AI modeling is its Knorr Zero Salt Bouillon Cubes. Salt is a core component of a bouillon cube, so to create a product with similar properties sans the sodium, it turned to AI. Using this modeling, it was able to predict the taste, structure, and performance of the cube on factory lines based on different ingredients used instead of trial and error formulation testing, Aben said.
Unilever also used digital modeling to create new packaging to accommodate its Hellmann’s vegan mayonnaise, he said. Since the condiment was created without animal-based ingredients, it behaved differently in a squeeze bottle, sticking more to the sides than its traditional egg-based counterpart. Therefore, it used modeling to test out various materials like oils to coat the inside of the squeeze bottle, notably cutting down on the amount of options it would have to try out IRL.
Every innovation process is different, but Aben estimated the use of AI modeling slices development time from months to weeks or even days, which in turn boosts the volume of new products.
“The nutrition area becomes much more innovative nowadays, compared to maybe many years ago,” he said. “Consumers demand a lot of innovation. So basically, it means that our innovators can deliver more innovation.”
Only human: Unilever not only uses AI to develop the products, but also to determine how consumers will feel about them—like if they’re too sweet or too savory—using data from the many consumer panels it has conducted. Aben said they can actually “analyze recipes that are written on paper and predict what they will taste like,” which he said helps the company quickly respond to trends, as well as run through different recipes to update the formulations of its current products.
Still, the use of AI doesn’t completely remove the need for humans. Aben said digital modeling often produces one recipe solution with a few alternatives that might seem the same, but “elicit a different response in the consumer,” so testing the product with actual consumers is still an essential part of the process. Despite decades of technological advances, the proof is still in the, uh, plant-based mayonnaise.