”The ongoing digitalisation and growth in online retail make product data move from being physical to digital”
On how to use AI and machine learning to improve food data
November 01, 2023
Who are you?
— I am the CEO and co-founder of FoodFacts. We create intelligent health and sustainability food data to empower the necessary transition towards better food. We are a platform where companies in the food ecosystem, such as retailers, food services, food producers, health and food experts, and organisations, can access actionable data on their food products. This to drive better business decisions and enable better food consumption within, for example, carbon footprint, health data, high sugar content, filters, personalised scores, Ferreira explains.
How do you work?
— Consumers and companies want to make better food decisions today with both health and sustainability in mind. However, the necessary information is not always available or accurate enough. There is also a growing need at companies for both data and tools to implement CSR strategies down to product level. This is quite complex due to the lack of data sharing across the value chain. So, we’ve created an intelligent B2B food data cloud, improving food data, using proprietary algorithms and AI that structures ingredient lists and automates complex calculations, including the climate impact of a product.
— Since we are able to index every single product down to ingredient level we can assign tags to products, like palm oil-free, vegan, organic, low sugar, high protein and many more. These tags can be combined with each other as well as with automated calculations like NOVA (level of processing) and Nutri-Score (nutritional rating score) to create smart filtering functionalities and scoring systems.
You mentioned calculations and you’ve just teamed up with Stockholm Resilience Centre and launched a new tool. What have you created?
— It’s a CO2 calculator that automates CO2 estimates for the whole industry at the product level based on the Life Cycle Analysis (LCA) methodology. When you need a carbon footprint on a product, the normal way is to carry out an LCA with a consultant. Those analyses are both complex and costly to do and you need to redo them every time there is a small change in your product formula or supply chain decisions. The result is that very few companies know the carbon footprint of their products. And if you cannot measure it, you cannot improve it.
— We wanted to give quick and cheaper access to reliable product carbon footprints so companies can start acting now on improving their products and limiting their GHG (Greenhouse Gas) emissions. The carbon footprint of food products usually makes up the most part of Scope 3 emissions for players in the food industry and around 80% of their total carbon footprint. So it is definitely an essential topic to address if we want to mitigate climate change caused by the food industry, which, as a reminder, is around 25% of all GHG emissions. Also, sustainability reporting is undergoing rapid change, and new directives is gearing many more food companies to more accurately report their footprints and other sustainability data, including at product level.
And you’ve also used AI and machine learning when developing it.
— Yes, by doing so we have been able to create a unique automated approach to replicate the process of full life-cycle analysis. The same calculation algorithm is being used for all products which makes it really easy to compare products. For example, if a retailer with a large portfolio wants to both access and give all their products a CO2 footprint, it’s difficult today. Many products are missing a carbon footprint, and not all existing CO2 calculations have the same scope and methodologies. With our calculator, the retailer can give all their products across the portfolio a carbon footprint and make easy comparisons. This is also true for food producers, who can then make sure the same methodology is used across their assortment and also in competitors benchmarks we can provide.
— By using models and an algorithm developed with expertise from Stockholm Resilience Centre, the tool estimates the emissions from farming, production, transport, and packaging. The basis of the automation is algorithms and AI that index products down to ingredient level, and estimate the volume of each ingredient. It then calculates the farming carbon footprint based on the country of origin, or hypothesis if data is not available. We have also incorporated data from other international public sources, such as Agribalyse, FAOSTAT, and the International Energy Agency, says Ferreira. She adds:
— We have developed our method to calculate CO2 emissions within the boundaries ’cradle-to-gate’. This means it takes into account the environmental impact of food products from the agricultural or farm stage to the distribution plant, where the product is packaged and ready to be transported to retailers.
Who’s your target group?
— Our tools are relevant for retailers, food services, and food producers. But other players connected to the food industry can also work with us, for example, health or sustainability organisations and experts, who need access to better and more intelligent food data and functionalities for better-informed decision-making. We are aiming for European expansion next year.
What was the hardest challenge in the development process?
— It took us over 9 months to develop the current version of the CO2 calculator, which already builds on other existing algorithms we have. The first step has been to define the overall methodology but also the detailed hypothesis. Being guided by experts from Stockholm Resilience Centre has really helped us a lot. The most complicated part has probably been coming up with the equations but also identifying which programming library to use to solve complex math problems, and how to incorporate constraints and generate constraints using computer code.
— We also had to tweak and finetune our hypothesis and algorithms quite a lot to fit specific category needs within food. And we are constantly working on fine-tuning it further, as we learn from specific examples and new data that we integrate and, of course, as new research comes in. Machine learning represents such a big growth potential for our data set. It can be used to finetune our hypothesis or generate intelligent insights and suggest automated improvements on products.
What else is going on in your sector?
— New regulations and standards are emerging, such as Nutriscore, European sustainability indicators Eco-score and Planet-score, ISO26000, and SBTi, says Ferreira. She continues:
— One key driver that we see is the new EU Corporate Sustainability Reporting Directive (CSRD) which will come into force in 2025, requiring large companies to report their climate impact for the year 2024. This forces large food companies to report their carbon footprint accurately and down to product level and requires new, efficient, and automated tools and smarter food data. Besides this there is higher accountability of brands and retailers on their supply chains regarding human rights and environmental impacts, doing due diligence on their supply chains in the Corporate Sustainability Due Diligence Directive.
— The ongoing digitalisation and growth in online retail make product data move from being physical to digital. This transformation allows the development and use of smarter food data. There are interesting ideas emerging around product passes and using QR codes or augmented reality to give consumers more information on their products, which a physical packaging will never allow. As mentioned there is a rising awareness among consumers to eat more healthily and sustainably, but also there is currently a lack of clear consumer information and a huge risk of companies just greenwashing. Providing trusted data can fill in those missing gaps and help organisations make better decisions.