How Unilever is embracing artificial intelligence to revolutionize food development
As artificial intelligence becomes a more important part of doing business in the food and beverage space, few companies have turned to the once-futuristic technology more than Unilever.
The consumer products company uses the technology to revolutionize and expedite the way it makes food products, leading to the launch of several new products such as Knorr Zero Salt Cube, Hellmann’s Vegan Mayonnaise and the Hellmann’s Real Mayo Squeeze Bottle.
Unilever is incorporating the technology across all facets of its business. AI helps the London-based company asses shelf life, texture and taste while predicting how the product will perform on factory lines during production. Unilever also harnesses AI to forecast flavor profiles, understand consumer preferences and boost food portfolio analytics.
Manfred Aben, the head of science and technology for Unilever Nutrition and Ice Cream research and development, recently sat down to discuss AI at the company, how it’s evolved and where it’s going next.
This interview has been edited for brevity and clarity.
FOOD DIVE: How has AI been used by Unilever?
MANFRED ABEN: What you actually see over the last decade or two is that the availability of data has really increased and, of course, also the computing power.
Maybe initially, [AI] was very much used in the marketing side and the consumer understanding side. In the supply chain side, obviously where we can measure a lot of what’s happening operationally and how can we optimize that, while also in R&D, which is my area where it’s really about creating products that consumers prefer, and so how do you make sure that for instance, the products are safe, have a long shelf life? So we use digital modeling and AI to predict shelf life of products so that we then can optimize the shelf life, which is obviously good for the use in consumers’ homes, but also for reduction of food waste, and then overall optimization.
And let’s say to predict flavor profiles, to predict liking of our product so that we probably want to be faster with innovation and really respond to consumer needs fast. You want to be able to understand what consumers want and how you solve their needs optimally rather than doing lots of experiments, either in the marketplace or in the laboratories. So, data modeling and AI helps a lot in shortening that period. So we can actually deliver more innovation to our consumers.
What products have you used AI to help develop?
ABEN: One example is Knorr Zero Salt Cube. Bouillon cubes have been around for a long time. So they’re basically stock cubes. And they’re basically constructed of course of salt, flavor, herbs and spices. But … if you want to make a cube without salt or sodium, or with less sodium, then you have to replace that with something else. And also very importantly, you don’t want to compromise on taste, and also not on how it performs in terms of people crumbling it, how it performs in the soup or in the dish you use it in.
Optional Caption
Permission granted by Unilever
So what we did was actually when we wanted to design a cube with no sodium, we have to change all the parameters almost like what ingredients could give us that salty flavor? What ingredients could give us that same texture and structure of the cube? We used digital modeling for that a lot, to just go through much data about many ingredients, great ingredient combinations, to then come down to a small set of options that we then of course, developed.
Another example is our Hellman’s Vegan, for instance, where we obviously replace the egg proteins with plant-based proteins. And again, consumers want alternatives that are better for them, better for nutrition, better for the planet because obviously moving to plant-based options is also good for that. But they don’t want to compromise on taste again.
They want to really have that same creamy mayonnaise that they’re used to having. So again, there we used AI models to predict the flavor and consumer liking in various countries where we launch a product so that you can make an optimal design. It saved us a lot of experimentation in the lab and thereby being able to respond quicker to consumer need.
Does Unilever use AI across other facets of its business, like the supply chain, minimizing trial and error or predicting how the product will behave on factory lines?
ABEN: Yeah, absolutely. We want to make sure our products are better on every aspect. You want to make sure you have a product that tastes great. That is stable. That is usable by the consumers. It’s profitable for us. It’s good for the planet and doesn’t produce a lot of waste. And it’s more nutritious. So you want to score all those points. Well, first you need a lot of data about what that means.
For instance, what is the consumer liking? What is the preference of consumers? So there’s a lot of data from taste panels, from consumer research that we have. On the other end of the spectrum, say you want to make sure that the product runs well and efficiently in a factory and comes out with constantly the same high quality every time. You need to understand your processes at various stages. So again, we have our chef starting in the development project, there are scientists making it possible to create it on a traditional scale and then have the supply chain to actually produce it. And all those elements need to link up together.
That’s where AI is really great is that you are starting to work with so much different data, and that only through smart statistical modeling or other technologies, even your natural language processing at the beginning to understand what consumers are saying about the products. Combining all of that together in models that helps you kind of go through this massive search space of potential solutions to consumer needs.
How important is AI for Unilever?
ABEN: The company has evolved with technology so to imagine doing what we’re doing without use of the technology is impossible almost. We’ve been able to develop solutions because of those technologies that we otherwise might not have been able to develop, either at all or practically because it’s just too time consuming. I think it’s really permeating in all parts of the business.
The financial benefits in the end is when we are more successful in addressing consumer needs so if we have products that are more likely to be successful, as we launch them, products that are faster on the market because we design them right the first time.
A whole other side of things is obviously with the challenges you’ve been facing in the market with the climate, and with the availability of materials. Also, the ability to quickly shift input materials, let’s say ingredients, in order to deliver the same quality product is something that if you wouldn’t have it, it would be a very high cost to do business. You saw there are both gains in terms of being able to be better and faster, bring innovation to market, which hopefully translates to successful market performance. On the other end, cost avoidance is a big, big thing as well.
You mentioned AI has helped with ingredients, given the disruptions in the supply chain in recent years. Can you provide more details?
ABEN: We have seen fluctuations in commodities and the availability of vegetable oils, etc. For instance, switching either the source of a vegetable oil or a vegetable oil together allows us to remain on the shelf. … Food is based on stuff that is grown anywhere around the world, but it’s not the same anywhere around the world. It might have slightly different textures, slightly different tastes and you want to make sure that consumers still like the product. That’s where we use these tools too, because replacing doesn’t have to be difficult but actually making sure you’re replacing in a way that consumers will still recognize they’re the products they love. This is where using these models come in.
Have you seen times when you’ve tried AI but you found a person could do it better?
ABEN: AI is not magic. Essentially it’s building on data and a smart ability to draw conclusions from vast amounts of data that take a lot of time and maybe just not so easy for human beings to do. There’s no magic there. And so the quality of the data is fundamental, and in some areas, we don’t have enough data to draw conclusions. Again in foods, there’s such a wide variety of ingredients and recipes and formulations, that some of them have never been seen or are not available in data.
You always need the creativity of a chef somewhere in the process, or you need the judgment of a taste panel. In the end, if you make something really new it might predict it, but you want to make sure it does predict it. In many cases, I think the value of AI is really there when you have the symbiosis of the human expert using these tools to actually expand their intelligence rather than then the whole notion of AI doing it all by itself automatically.
Even being originally trained in AI myself many years ago in the 90s, I’m a very big optimist about where the technology is going to bring us but I’m very, very confident as well that human judgment and human creativity is hard to replace by data-driven models.
Is there an example of something you went through where, like you were able to go through millions of recipes or you were able to go through all this data quickly using AI?
ABEN: The gut microbiome, obviously a very interesting space into understanding how you know all the microbes in your gut impact your health, not only your gut health, but actually have an impact on your mental state because of all kind of neurotransmitters that link the brain and the gut together.
And we’ve been working with the partner Holobiome for some time to find out are there any ingredients that have a positive impact on people’s minds and work across the variety of microbiomes that you have. That would not be possible by doing by hand. Together with them, we went through literally millions of edible ingredients to see and try to figure out which one does what in order to get that benefit. That looks incredibly promising.
Those are examples where you need databases of ingredients and their properties and the very complex models of how that interacts with the guts and then narrow it down, and that’s where AI really plays a role. Obviously, if you talk about recipes for mayonnaise, those are not millions, those are maybe a few hundred, but still going through them and saying ‘Okay, I want to reduce, for instance, this level of sugar, how do I make all the possible combinations of ingredients, which then turns out to be, of course, a big number. It’s just not possible to do that physically.
Are you surprised at the speed at which AI has evolved?
ABEN: It’s evolving fast and there is kind of an exponential nature to it because as more data is generated more can be created. I think it moves really fast. One of the challenges is, of course, is that we must, in order to make sensible decisions based on AI, we, of course, have to understand how it’s come to its conclusions. And because it’s fundamentally a statistical process, it’s based on correlations, not always on causations. So that’s again, where you always need some human expertise to make sure that the conclusions you draw are actually correct. It’s an exciting time.
Source: fooddive.com