Making toilet paper run at 48 km/h

This article was originally published in German by Tagesschau. This is an English translation with our added commentary.

Artificial intelligence (AI) is transforming industries by making machines more reliable. Companies are already capitalizing on this technology, including those in the toilet paper manufacturing sector.

Tino Pietraßyk sits in a loft-style open office, passionately discussing toilet paper. He can talk on and on about the nuances of the production process: single-ply, double-ply, band speed, potential issues, and ways to improve. Pietraßyk, 36, is a Senior Data Scientist at the Berlin-based AI company FactoryPal.

FactoryPal advises tissue converting lines at numerous locations across Germany. “Toilet paper is our bread and butter,” Pietraßyk explains before diving into the essence of artificial intelligence: What does AI actually mean? What does it do in practice? And can it be monetized?

Take toilet paper, for example. In Germany, production is decentralized, meaning it takes place at various locations with different machines—each with potential for optimization. This is where AI comes into play. The goal, as Pietraßyk puts it, is to “run a paper web through a 50-meter-long machine as quickly as possible without breaking anything—ideally at a speed of 48 km/h.”

AI companies like FactoryPal have realized a key truth: every machine makes mistakes. Paper can tear or jam, rollers can get stuck, and feeds can fail. There are countless sources of errors. Typically, a toilet paper production line operates at only 60 to 70 percent of its theoretical output capacity. The rest is lost due to minor issues.

This week, “AI in Production” was the focus of a conference in Berlin, featuring real-world applications rather than futuristic visions. The event drew numerous company executives, a handful of researchers, and Berlin’s Senator for Economic Affairs.

FactoryPal’s goal is to increase output. Pietraßyk promises that the “Overall Equipment Efficiency” (OEE) can improve by at least three percentage points when FactoryPal takes on a project. That might not sound like much, but “each percentage point of OEE can cost up to a million euros per year,” he explains.

Modern industrial plants are highly complex. Even a seemingly simple production line for toilet paper rolls is too intricate for humans to fully grasp. Pietraßyk and his team measure between 200 and 2,500 input parameters, depending on the plant, along with 30 to 70 adjustable settings on the machinery.

“This results in billions of possible combinations,” Pietraßyk summarizes—billions of ways to improve OEE by a fraction of a percent. Algorithms, which learn independently, determine which adjustments lead to more efficient operations. In short, this means producing more toilet paper in less time with fewer production interruptions.

This is how AI companies earn their revenue: by optimizing processes and squeezing out the last few percentages of efficiency. Everything is automated, relying on algorithms that can detect causal relationships far beyond human cognitive abilities. “In principle, AI is just math,” says Pietraßyk. “Our AI doesn’t care what we’re producing.” Toilet paper is no different from other products. It’s an established industry with skilled professionals, though convincing everyone that improvements are still possible can sometimes be challenging.

Pietraßyk shares his experience with toilet paper production experts: “Many workers prefer to run the machine a bit slower because they believe it reduces the risk of paper tears.” He pauses briefly. “But the data analysis doesn’t support that.” In other words, the experts are wrong—the algorithm knows better.

Concerns that artificial intelligence might lead to widespread job losses are unfounded in this case, Pietraßyk says. “Replace workers? Not at all in this instance.” He isn’t aware of a single case where someone lost their job because of AI. Instead, the algorithms simply make the process more efficient.