Speaking at the Fi Europe Conference 2023 in Frankfurt, Lahousse unpacked how these digital processes are helping to unleash a new generation of flavours. Pivotal to the process is the fact that by harnessing knowledge graphs and AI technology, development can be accelerated, while novel, sustainable and more appealing flavours can be created.
What exactly does foodpairing mean?
Foodpairing was established 14 years ago to create new and exciting flavours. Every day the company analyses around 100 products from all over the world for taste and aroma with the support of a network of 200,000 chefs. In the last seven years, it has integrated digital processes to provide more in-depth analysis these ingredients. This has also helped speed up the process significantly, enabling it to simulate the entire product development cycle in a matter of hours.
“This process helps us to create better products that are customised for a target audience, containing unmatched flavour combinations based on robust consumer and market insights,” said Lahousse. “We look at food development as being a software that is never finished and continuously needs to be updated.
“You actually need to be able to simulate everything, so instead of having a fragmented linear development process, we create a loop that is completely integrated and able to be continuously improved on.”
This results in unlimited concepts and a far easier consumer validation process, deeper insights and benchmarking and a consumer-centric process that is a fast and efficient cycle. Another advantage is that because all the development is simulated, there is no limit to how much testing can be done, avoiding time in lab endlessly testing physical samples.
How knowledge graphs serve to enhance development
The term knowledge graph was coined by Google in 2012 to describe an intelligent model that understands real life entities and their relationships to one another.
In a nutshell, Google-generated knowledge graphs serve to find the right element in a potentially complex equation, then provide the best summary – known as a knowledge panel - of that information with more factual data that allows the user to go deeper and broader in their analysis.
“What defines a knowledge graph is that it is made up of lots of notes and these notes are entities,” said Lahousse.
Taking Nestlé’s Gerber brand as an example, Lahousse demonstrated how a knowledge graph can be broken down into these notes to tell an in-depth story about any product’s formulation and the ingredients.
“When I enter the knowledge graph about this product, I can see what ingredients are in the product, the country of origin, what food category it is, and the mood states,” said Lahousse. “So, this provides one data model with all the information in one structured database. But then we can go deeper, providing all the information for the recipe, the social media data about the product, clinical data, and even e-commerce information about products in that category or market.”
Lahousse explained that when he talks to big food companies, their data tends to be siloed, whereas the knowledge graphs connect all this data as one big model and integrate it using AI technology. This means that all the information becomes interactive, making it far more valuable.
Combining knowledge graphs and AI tech to create unique flavours
“From the perspective of AI, which will become an increasingly important part of the development process, it is important you have one data model,” said Lahousse. “As part of this whole data-centric evolution, companies need to adhere to ‘FAIR’ principles. FAIR is an acronym for Findable, Accessible, Interoperable and Reusable. This means that your data is communicable with other data and reusable.”
This results in a data-centric process at the core and, instead of having a focus on analytics and metrices, food businesses can focus on having data that is accessible, accurate and can be instantly used by machine learning models.
The integration of AI technology with the knowledge graphs help to extrapolate and process data in a number of areas, including insights for ingredients and product analysis, forecasting for macrotrends, flavours and functionality, competitor benchmarking and establishing where the white spaces are, which can all be mapped out in detailed dashboards.
“Combining AI with knowledge graphs, we can typically generate millions of future concepts that can be validated according to desirability,” said Lahousse. “We can predict the purchase intent, assess if it will not cannibalise on the sales of other products to determine the viability and then finally the feasibility.”
Pointing to a concept that the company helped create for the brand Pretzels, Lahousse explained how it used knowledge graphs and AI to create a totally new concept that would not cannibalize any other products in the portfolio. This was achieved by analysing consumer data and existing products to help create concepts with unique flavour descriptions targeted at specific consumer interest and tastes.
The next step is to create the complete AI assisted formulation with appropriate ingredient substitution. Many food companies are currently trying to enhance the sustainability of their formulations, so at this point the AI model can also suggest ingredients with the lowest impact, he said.
“A lot of food companies are struggling with the EU’s requirement for ESG Reporting for 2025 and many of these businesses want to actually replace the ingredients that have the biggest impact with ingredients that have a lower impact,” said Lahousse. “So having a knowledge graph, you can also apply it to see how you can actually swap out for ingredients with less impact.”
Once the process has reached this point, the Foodpairing AI process has essentially reached the final stage of development, leaving just the packaging and the product launch to be complete in an effort to get the newly created brand on store shelves.
What are the advantages of knowledge graphs?
To conclude his presentation, Lahousse outlined five main reasons why knowledge graphs and AI can be so powerful to food businesses:
- Joining up heterogenous data from diverse sources / be FAIR
- Execute complex queries over the entirety of stored knowledge
- Integrating data and knowledge at scale => making data similar!
- Turning data into ‘a resource to exploit”
- Large language models + KG = power couple
