Pringles maker's AI digital twin cuts waste 13% at Poland plant

Kellanova spent four years and $5 million teaching an AI model to make the perfect Pringle, reshaping factory digital twins.

An AI-powered digital twin of Kellanova's Pringles line in Poland cut waste 13 percent and lifted quality 10 percent, a sign the years-old manufacturing technology is scaling with new AI investment.

"Every potato batch you get is going to be different," Cedrik Neike, member of the managing board and chief executive of digital industries at Siemens, said. "Independent of the input, we make sure that the Pringles taste would be exactly what you want it to be."

The project, built over four years with a $4 million to $5 million investment, captures 200 data points from sensors on the production line every millisecond. The data feeds an AI model that runs real-time simulations to predict how each chip turns out and suggests machine tweaks to account for variances in raw materials, from flour particle size to where the potatoes were harvested.

The system is live on one line at the Kutno, Poland, factory, where Kellanova reports a return on investment above 40 percent. It will expand to Belgium and to U.S. lines in 2027, part of a broader push by Siemens and Nvidia to bring digital twins to factory floors.

For Jan Laenen, senior director of engineering at Kellanova, the goal is simple: every crisp, salty chip off the line should be identical — same crunch, same taste, same saddle-shaped golden sliver. No broken dough, no sticky chips caught in the fryer. "We want to make a perfect chip," he said.

The system works by creating a real-time digital version of the dough as it moves through the line. New sensors and live machine data capture minute variances in raw materials, such as the particle size of the flour. An AI model running on a Siemens Industrial Edge Device platform inside the factory processes the data and suggests adjustments — like tweaking the amount of oil or water in the recipe — before a batch goes wrong.

More than 200 parameters shape a "perfect Pringle," from dough humidity to where the potatoes were harvested. Even potatoes from the same supplier vary by season, meaning the dough is constantly changing. "A different dough could mean a different texture. It has an impact downstream," said Francesco Ielmini, a senior project manager at Siemens.

In the past, "doughmakers" adjusted machines by look and feel, grabbing dough off the line to stretch it or weighing chips fresh from the fryer. The AI model now does that work in real time, and when local adjustments are not enough, it escalates to a larger cloud-based model with access to more historical data. Both use conventional machine learning rather than generative AI.

Digital twins move from 3-D models to live production

The Pringles project is the latest sign that digital twins, a concept that has existed for years, are starting to scale in manufacturing. When they first gained traction five years ago, they were often just 3-D models with data tagged to them, said Evan Brown, a principal analyst at Gartner. Now they operate closer to real time and offer AI-powered recommendations on how to optimize production.

Siemens said it has worked on digital twins for rockets, batteries and microchips, and Nvidia has invested heavily in the technology for factory simulation. The next step, Brown said, is granting AI autonomy to change machine settings on its own — though most companies, including Kellanova, prefer a human worker to make those calls.

What the investment means

Kellanova produces 705 million pounds of Pringles annually across three factories in Poland, Belgium and Jackson, Tennessee. Late last year, the company was acquired by privately held snacks maker Mars, which now handles U.S. manufacturing of the chips; in Europe the two businesses still operate side by side.

The Poland project delivered a 10 percent improvement in quality, a 13 percent reduction in waste and a return on investment above 40 percent on the $4 million to $5 million spent. Kellanova plans to expand the system to Belgium and to U.S. lines in 2027, with the goal of eventually covering all lines. Laenen said the model could eventually handle multiple varieties of potatoes, corn and rice, using that variability to still produce the same chip.

For investors, the project shows how industrial software is becoming a bigger revenue driver. Siemens' digital industries unit, which sells the edge platform and AI tools behind the project, is a core growth engine for the German conglomerate, while Nvidia's push into digital twins positions it beyond data-center chips. A 40 percent-plus return on a mid-single-digit-million investment gives manufacturers a concrete case for spending on AI, a pitch that could accelerate adoption across food, auto and electronics plants.

This article is for informational purposes only and does not constitute investment advice.