How can a textile manufacturer predict what customers will want months in advance, without drowning in unsold inventory or leaving orders unfilled? That’s the challenge frottana Textil GmbH & Co. KG, a German producer of terry cloth and home textiles, faced as global supply chains grew more volatile after 2020. Their answer came from an unlikely source: an artificial intelligence tool developed by Fraunhofer IWU, the Institute for Machine Tools and Forming Technology in Chemnitz.
The project, completed in late 2024, combines historical sales data, production schedules, and something most algorithms ignore — the tacit knowledge of veteran employees. The result? A system that doesn’t just predict demand. It actively teaches planners how to improve their forecasts.
Why Textile Forecasting Is So Hard
Textile manufacturing operates on long lead times. Yarn must be spun, fabric woven, dyed, cut, and sewn — often months before a single towel hits a store shelf. A bad forecast means either piles of unsold stock (expensive) or missed orders (worse for reputation).
“The textile industry is particularly vulnerable because seasonality and fashion trends compound normal demand fluctuations,” says Dr. Matthias Putz, head of the research group at Fraunhofer IWU. “What frottana needed was a system that could learn from both data and human experience — something that adapts as conditions change.”
Traditional forecasting tools rely on moving averages or simple regression models. They break down when faced with sudden shifts — a heat wave boosting towel sales, a shipping crisis delaying raw materials, or a new competitor flooding the market with discount products.
Frottana’s planners had been doing this manually, using spreadsheets and gut feel. It worked, but barely. And it didn’t scale.
How the AI Tool Actually Works
Fraunhofer IWU’s solution is a hybrid AI system. It combines a neural network trained on 15 years of frottana’s sales data with a rule-based module that incorporates production constraints — machine capacities, material availability, and order lead times.
Here’s the clever part: the system also captures implicit knowledge from senior planners. Every time a planner overrides an AI forecast, the system logs the reason — whether it’s a special promotion, a new customer contract, or a supplier issue. Over time, the AI learns to anticipate these human adjustments.
“We didn’t want to replace the planners,” explains Dr. Anja Fischer, project lead at Fraunhofer IWU. “We wanted to make them better. The tool acts as a decision support system that gets smarter with every correction.”
The interface isn’t flashy — it’s a dashboard showing predicted demand ranges for each product line, with confidence intervals shaded in green (high certainty) to red (high uncertainty). Planners can drill down into any forecast to see which factors are driving the prediction. If the AI says orders for bath towels will spike in March, the planner can see that the model is weighting a past Easter promotion and a recent uptick in hotel bookings.
This transparency matters. Planners trust the tool more — and they catch errors early. In pilot tests, frottana reduced forecast error by 23% within six months.
But look, no system is perfect. The AI initially struggled with new products that had no sales history. The team solved that by using similarity matching — the model compares a new towel design to existing products with similar characteristics (size, material, color) and borrows their demand patterns. It’s not foolproof, but it’s better than starting from zero.
Bringing Human Know-How Into the Loop
This is where the project breaks from typical AI implementations. Most demand forecasting systems treat human input as noise to be filtered out. Fraunhofer IWU did the opposite. They explicitly designed the tool to capture planner reasoning.
Every override is tagged with a reason code: “customer order,” “marketing campaign,” “weather event,” “supplier delay,” or a free-text field for other causes. After six months, the system had recorded over 1,200 such overrides, each one a data point teaching the AI about real-world constraints that don’t appear in historical spreadsheets.
“The biggest insight was that planners weren’t just guessing,” says Dr. Fischer. “They were incorporating information that wasn’t in any database — a call from a key client, a rumor about a competitor’s price cut, news about a port strike. The AI learned to recognize these signals.”
The system now generates two forecasts: a baseline (purely data-driven) and a hybrid forecast (blended with learned planner behavior). Comparing them reveals where human intuition still beats the algorithm — and where the algorithm catches things humans miss.
This dual-forecast approach is unusual. Most companies just pick one model and run with it. Frottana plans to eventually feed the hybrid forecast directly into their ERP system, automating purchase orders and production scheduling. That’s the big prize: a fully digital supply chain that adapts in near real-time.
And it’s not just about towels. The same approach could work for any industry where demand is volatile and human expertise is deep — fashion, electronics, even pharmaceuticals. You might wonder: is this the end of the supply chain manager? Probably not. But the job is changing. The managers who thrive will be the ones who learn to collaborate with these systems, not fight them.
Meanwhile, NASA scientists working on entirely different problems are also pushing the boundaries of AI — though their focus is on the cosmos rather than consumer goods. Meet the 4 NASA scientists quietly shaping the future of space exploration — they’re using similar machine learning techniques to analyze data from distant planets.
What It Means for the Rest of Us
For consumers, better demand forecasting means fewer “out of stock” notices on your favorite towel set. For the planet, it means less waste — textiles account for nearly 10% of global carbon emissions, much of it from overproduction that ends up in landfills.
“Every towel that doesn’t get made because the forecast was accurate is a towel that didn’t consume water, energy, and dyes,” says Dr. Putz. “The environmental impact of better planning is massive.”
Frottana is already expanding the system to its other product lines and considering licensing the tool to other manufacturers. Fraunhofer IWU is working on a version that incorporates real-time supply chain data — shipping container availability, port congestion, energy prices — to make forecasts even more resilient.
The next frontier? Predicting demand not just for the next quarter, but for 18 to 24 months out — the kind of horizon that determines whether a company builds a new factory or not. That’s a much harder problem. But if the frottana project proves anything, it’s that the best AI systems don’t replace human judgment. They amplify it.
And in an industry where a 5% forecast error can mean millions in lost revenue or wasted inventory, that kind of reliability isn’t just nice to have. It’s survival.
Frequently Asked Questions
How does the Fraunhofer IWU tool differ from standard demand forecasting software?
Most tools use only historical data and statistical models. This system also captures the reasoning behind human planner overrides — effectively learning from expert experience over time. It generates two forecasts (pure data vs. hybrid) to highlight where human intuition adds value.
Can this AI system work for small manufacturers with limited data?
Yes, with caveats. The system uses similarity matching to handle new products with no sales history. However, it requires at least 2–3 years of historical data for the neural network to train effectively. Smaller firms may need to start with a simpler version and add data gradually.
What’s the biggest challenge in implementing AI demand forecasting in textiles?
Cultural resistance from experienced planners is the main hurdle. The Fraunhofer team overcame this by designing the tool as a decision support system with transparent explanations — not a black box. Planners who saw their own knowledge reflected in the model’s logic became its strongest advocates.
