Get Started

A Digital Twin for Product Development

Use Cases & Projects, Dataiku Product, Scaling AI Renee Boerefijn, Doug Bryan

Bunge Loders Croklaan, the plant-based lipids business of Bunge, is a global leader in specialty oils and fats solutions for food manufacturing companies worldwide. As you probably know, the sauce on your favorite fast food chicken sandwich and the chocolate coating on your favorite candy bar didn’t grow on a tree. They were manufactured. Bunge Loders Croklaan makes lipid ingredients for those and many other products, from bakery and confectionery to infant nutrition and plant-based foods. Bunge has approximately 300 facilities located in more than 40 countries including port terminals, oilseed processing plants, and grain elevators.

Bunge Loders Croklaan truck

Bunge Loders Croklaan’s industry is global and competitive. (This article is written from the first author’s point of view. Renee Boerefijn, Ph.D. is the Innovation Director EMEA at Bunge Loders Croklaan.) To remain a leader, we need to continuously innovate. A few years ago, we started a journey to become more data-driven and leverage AI within more of our day-to-day processes. Our two initial steps were to engage the workforce and make data more accessible.

First, we held online brainstorming sessions to get stakeholders involved. The sessions generated a lot of excitement and over 30 ideas. The engagement and alignment of the team was just as important as individual project ideas, and it inspired everyone from the start.

Next, we implemented Dataiku, an enterprise-scale data and AI platform, with the help of our partner EyeOn. The shared platform was instrumental in overcoming organizational complexity, data silos, and data inconsistencies. By cataloging and exploring data in one place, inconsistencies became clear. For example, one manufacturing site might measure viscosity in pascal-seconds and energy in joules, while another site uses poise for viscosity and calories for energy. Once we know this, we can easily harmonize the data.

Once the team was energized and a data and AI collaboration platform was in place, we planned three phases of development focusing on product data, component data, and manufacturing data, which we’ll highlight in the remainder of the article.

Phase 1: Product Data Integration

We created a document library of over 4,000 products that includes key performance attributes. Manually extracting the data would be time consuming and error prone. Instead, we developed Python code to extract attributes and store them in a database, which works very well. The second challenge was that different methods are used for the same attribute, meaning that similar numerical values do not have the same meaning. For example, viscosity is a measure of the thickness of a fluid and can be measured with an oscillating or a linear method. Harmonizing the data was like learning to share one language, a massive benefit to enable business across our sites around the world.

That gave us trustworthy product data for expert users, but we also wanted to make it accessible for sales, supply chain, and product management. However, as it contains a lot of sensitive data, we need to carefully manage access. Imagine a technical customer support representative putting the entire database on their laptop, visiting a customer in Brazil, and accidentally leaving their laptop in a cafe. To prevent this, we developed a web app to control access at the product level and avoid replicating data to end-user devices. App users can enter partial product specifications and instantly get information on similar products that they have access to. The app very quickly generated great value via:

  • Reduced time to market: Technical customer support quickly knows if we have an existing or past product that meets the customer’s needs. If not, then they can see our history of similar products to greatly reduce the experimentation needed to develop a new one.
  • Higher initial product yields: Manufacturing can also benefit from similar past products and get it right in less time.
  • Better sales win rate: Like in many competitive industries, as soon as we give a customer a new product specification they ask, “How much will it cost? When can I get it and in what volumes?” Being able to deliver faster than our competitors is a big advantage. That was especially true during 2020-21 when consumer behavior drastically changed in just a few weeks.
  • More cross-selling: Easy access to product data, using measurement units the customer is most comfortable with, increases cross-selling. Sales teams no longer need to worry about which manufacturing sites use which units, or to delay the sales process while technical support does conversions between units for them. Conversions are done automatically on the fly.
  • Better and faster onboarding: The web app is tremendously helpful for bringing new people onboard much faster, as it makes information available instantly without needing to learn how to operate complex systems first, so they can hit the ground running.

We estimate that this web app alone increased our ROI significantly.

Phase 2: Component Data

Each product is made from components, or intermediate products, not unlike a shirt or a watch.  Components such as oil fractions are used in many final products. There are many ways to combine our components but, of course, not all of them yield a tasty product. In Phase 2, we added component data to each product to further improve efficiency. Combining component and product data allows for much faster what-if testing. What if we reduce or swap a certain component? Knowing how components are combined to create products with various attributes makes it much faster to answer these questions.

Phase 3: Optimize Manufacturing

A typical manufacturing site has a relatively small number of incoming raw material streams making many different final products. With a lot of intermediate components and a large number of products you can imagine how many possible ways there are to configure the mix of products. In Phase 3, we combine product, component, and raw material data across the entire portfolio to optimize manufacturing. Some days meeting a difficult delivery schedule for a customer might be the primary goal while other days it is revenue or profit. With the right quality data and a collaborative AI platform that overcomes organizational complexity, it will soon be feasible.

Bunge Loders Croklaan’s phased approach is a great example of digital twin development best practices, namely, engage stakeholders across a company early, pay attention to data quality, focus on app usability, and generate value every step of the way. This is the first real-world example in our series on digital twins so stay tuned for more coming soon.

EyeOn is a management consulting company specialized in integrated business planning, supply planning, demand planning, and financial planning with more than 150 customers, including KraftHeinz, Philips, Stryker or Cargill. The company has shifted their methods surrounding data processes to keep up with today’s increasingly competitive and AI-driven world.

You May Also Like

Simplify Anti-Money Laundering (AML) Alerts With Data-Backed Triage

Read More

Key Steps Involved in the Machine Learning Process: A Primer

Read More

Get Rid of Friction & Trigger an AI Chain Reaction With MLOps

Read More