Premium Chocolate Production Perfected: AI's Role in Quality Excellence

The "Lindor" truffles from Lindt & Sprüngli are world-renowned among chocolate lovers. These truffles have been sold since 1969, when they were introduced as a seasonal treat and they were used as actual decorations to hang on Christmas trees. Half a century later, billions of Lindor truffles are made every year. To produce at such a large scale while maintaining quality excellence, Lindt & Sprüngli runs several high-end factories in Europe and the US.

Producing Lindor truffles is a complex process that spans from ensuring the quality of the cocoa beans, through manufacturing, to logistics. The production of Lindor truffles involves multiple production steps such as molding, filling, and packaging. While the overall production setups are consistent across factories, local differences in factory infrastructure, product variants, and machine age create challenges in finding the optimal parameter settings for obtaining perfect Lindor truffles. Additionally, seasonal variations in conditions like air humidity and ambient temperature play critical roles in the production process. Therefore, understanding the interplay between process parameters and environmental factors is essential to ensure consistently high quality – and that’s where EthonAI comes in.

A complex, multi-step process

Lindt & Sprüngli has a strong commitment to excellence that goes beyond the quality and taste of their products. “Our employees have vast knowledge that we use to improve our production.“says Jessica Valli, Program Manager for Digital Transformation in Operations, “to support them, we must provide them relevant facts and information.” For this, Lindt & Sprüngli records a lot of process and quality data. “We were looking for new ways to leverage it,” explains Valli. “For example, a recurring problem was to identify and eliminate the root causes behind quality deviations and inefficiencies in production. We were happy to learn that EthonAI could help us solve this important problem.”

ETH AI Center startup affiliate EthonAI spun out of the Chair of Production and Operations Management at ETH Zurich to develop software tools that detect, monitor, and prevent quality losses in manufacturing. The tools use causal AI technology to help discover and understand the physical relationships between process parameters and quality outcomes. In this way, manufacturers can fine-tune their processes, and thereby prevent costly quality losses.

Standardizing quality management across countries

At Lindt & Sprüngli, the EthonAI tools connect the process data from the Lindor production lines with the data from factory quality control systems. This gives the line managers and production teams a holistic overview of their production lines with their dozens of parameters and quality measurements. All these data matter for optimizing local production. In addition, it provides Lindt & Sprüngli a birds-eye view on data and performance from several factories, enabling group-level learning and standardization in quality management across countries in a simple format. The EthonAI software is live in Lindt & Sprüngli factories in three countries across two continents.

By understanding the complex interactions between the process parameters and negative quality outcomes the process experts at Lindt & Sprüngli are able to take action more quickly and decisively. Ultimately, the focus is on the prevention of production issues, and discovering root causes from the real-time data fed into the EthonAI analysis tool is essential for going beyond reactive measures. For example, production experts use EthonAI’s software to fine-tune machine temperatures or mold configurations to avoid quality losses proactively. This is possible because EthonAI’s tools rely on causal AI to go beyond mere correlations, and suggest improvement actions ranked by their overall impact on the final quality. In essence, EthonAI’s causal AI can simulate the effect of potential improvement actions, so it can be tried out before adjusting the physical production.