How RumeXperts unlocks dairy farms’ data potential

Léonard Theron, Chief Technology Officer at RumeXperts – a spin-off of the University of Liège specializing in veterinary research – elaborates on this rather surprising fact about cows. “Tags, collars or other tracking devices, devices installed in milk tanks, weather data… Dairy farms monitor all kinds of parameters relating to cattle farming, such as nutrition, milking, etc. However, today, this massive and invaluable resource often goes underutilized.” 

A flurry of data to reduce response times

RumeXperts and Léonard have been working for several years to buck the trend. Their ambition? To tap into all the potential that this data can provide for the agricultural sector and dairy industry. 

“Why is this data so important? Firstly, because milk prices are based on the milk’s composition, e.g., protein content, fat percentage, and bacterial content,” Léonard explains. “Even the slightest variation can have an impact on the revenue of the farms. On average, however, there is a gap of 35 days between the moment something happens that affects these parameters and the farmer’s response. To reduce this response time, we created Salve. This online platform analyzes dairy farm data and sends out alerts, e.g., an increase in a milk tank’s bacterial content, well before the farmer realizes that there is a problem. This can drastically reduce the response time between a problem occurring and measures being taken.”

However, Léonard stresses that this is not Salve's only benefit to dairy farmers. “Data collected from 250 farms gives our users an overview of multiple herds and serves as a benchmark to compare with data from their own farm.”

Guillaume Prieur, Data scientist at delaware, Léonard Theron, Chief Technology Officer at RumeXperts and Benoît Loffet, Senior Manager Data & AI at delaware.

From a descriptive to a predictive model

While Salve allows farmers to respond more quickly, its data is mainly descriptive, meaning the system only triggers an alert when a variation already exists. So how could this tool become even more relevant? RumeXperts and Léonard decided to explore the predictive potential of this data and of AI, allowing farmers to respond as soon as any potential risk is identified, i.e., well before the anomaly occurs.

“We had copious amounts of data at our disposal. We also knew that there were recurring patterns in the data and that some of the farms were predictable,” Léonard says. “We therefore set out to determine how AI could be used to help us generate predictions that can offer real added value for these farms. That’s when we decided to reach out to Digital Wallonia.” 

A regional digital strategy

“Digital Wallonia is the name given to the region’s digital strategy and platform,” explains Antoine Hublet, Project Lead at the Agence du Numérique wallonne (AdN), the organization overseeing this strategy’s roll-out. “Our aim is to accelerate the adoption of artificial intelligence and digital technologies throughout Wallonia and support companies in their digital transformation.” 

“The DigitalWallonia4.ai program focuses on the adoption of AI through two mechanisms, both of which are 70% financed by us: 

• Start IA is a maturity assessment tool with which we help companies to identify AI opportunities. 
• Tremplin IA is an extension of Start IA and determines whether an idea can be turned into reality. 

It was, in fact, Start IA and Tremplin IA that brought together RumeXperts, Digital Wallonia and delaware.” 

A model candidate

“What we liked about RumeXperts as a candidate? They had already made a lot of headway with their project. They had a clear vision about what they wanted to achieve,” Antoine continues. “What’s more, such an initiative in the Walloon agri-food industry, where farmers are slow in adopting AI, may inspire other market players. It demonstrates how technology can lead to solutions, with real added value for the farming industry.” 

This is where delaware came in. According to Antoine, “delaware is part of a pool of some fourty AI experts that we rely on for support. The challenge when choosing an expert is to find the best possible match between the company requiring support and the expert providing it. In the case of RumeXperts, delaware was a natural fit. Why? Firstly, because they already have considerable expertise in the agri-food industry, but also because they had already overseen similar projects, therefore perfectly meeting RumeXperts' expectations.” 

According to delaware Senior Manager Data & AI Benoît Loffet, “We have been participating in these Digital Wallonia support projects since they were first launched. It’s always a pleasure to work with them because the concept’s extremely agile approach allows us to really hit the ground running. The variety of missions gives us a better understanding of the core business of companies across various industries. This understanding of the key challenges they face is fundamental to making sense of the data that is collected.” 

“We rely on a team of 140 data specialists. Thanks to their excellent command of Microsoft and SAP, these experts can define strategies, set up modern data platforms in the Cloud, develop reporting systems, create AI algorithms, and help companies manage any changes resulting from this new way of working. Léonard and his team had a very good idea of both what they had achieved so far and what their next step would be. They already had a platform to collect their own data. The starting premise was ideal, with everything crystal clear right from the off.” 

From Start to Tremplin

During the Start IA stage, Benoît’s team explored the data that Salve used to understand how they could generate added value. “We did this over three stages”, Benoît explains. “We started by looking at the architecture, analyzing the compatibility between ours and RumeXperts’ systems. We then focused on the actual data, which enabled us to define three data exploration tools based on this data analysis, which were promising in terms of their predictiveness.”

Following the exploratory stage of Start IA, RumeXperts and delaware moved things up a gear, working on a Proof of Concept and switching to the Tremplin IA. There too, the results exceeded expectations. 

Léonard: “We were able to demonstrate that two of the three objectives in terms of predictability, i.e., milk fat content and bacterial count, held up and could be used in production. In terms of the farms in Salve, 30 to 50% of the 250 companies in our cohort are predictive, so we can provide real added value. These outcomes are extremely positive. In farming, where everything is based on intuition, you could say that this is a veritable revolution, paving the way for a new form of cattle farming that consumes fewer resources.”

A broadly positive assessment

“Together, thanks to a predictive model run on the Microsoft Cloud, we were able to develop solutions for transforming data into usable information,” Léonard explains. “Our Salve platform will use this model to query data, retrieve results, and present them in such a way that users can understand them. delaware’s extremely professional, cross-sectional vision on data analytics was key in this. Their data scientists understood very quickly what kind of model we needed. They then looked at a large number of models and used our data to see what these models would generate, before defining which models would be more effective. They got to grips with our business very quickly, even without knowing our sector from the onset. This allowed us to work together in a co-construction model and learn from one another.”

“To tackle this type of challenge, you need to work very closely together,” Benoît stresses. “Data is useless without business knowledge. The only way to produce results is to share and exchange information.” 

For RumeXperts, these results already appear to be very promising. “The version with the indicators we generated is set to go into production in December. This will help us convince new users to switch to our platform, while rendering our results even more consistent,” Léonard concludes.