AI Cuts Scrap 10% on Nonwovens Line, But Data Quality Blocks Scale

One nonwovens production line cut scrap by 10 percent with causal AI. That figure anchors the new 49-page white paper from the International Textile Manufacturers Federation (ITMF) and the Institut für Textiltechnik of RWTH Aachen University (ITA), which finds measurable AI returns across textile operations but identifies poor data quality—not algorithms—as the main barrier to wider adoption.

The report, titled "AI in the Textile Value Chain: A Strategic Guide to Artificial Intelligence for the Global Textile Industry," draws on webinars the two organizations held over the past two years. It is available now for sourcing and manufacturing executives weighing AI investments.

What does the research actually show?

The authors reject the framing that textiles are under-digitalized. Decades of partial digitalization have instead left the industry "digitally fragmented," with large volumes of data locked inside disconnected systems.

"Data is generated across machines, quality systems, logistics platforms and enterprise software, yet these streams are rarely fully interoperable," the report noted. "This fragmentation has several consequences: decision-making remains distributed across organizational silos, information is duplicated, manually transferred or reinterpreted and opportunities for cross-functional optimization remain limited."

The operational cost is concrete. Many mills use digital systems mainly for documentation and reporting, not active decision support. "Data is collected after the fact, not integrated in real time—limiting the industry's ability to respond to demand variability, quality deviations or supply-chain disruptions," the authors wrote.

Where is AI delivering ROI today?

Confirmed returns cluster in four targeted applications:

• Computer-vision fabric and yarn inspection
• Predictive maintenance
• Demand forecasting
• Energy optimization

The nonwovens case study, showing a 10 percent scrap reduction from causal AI, is the paper's headline measured result. The authors frame these wins as incremental, tied to well-defined problems.

"Value emerges when AI is applied incrementally to well-defined problems—particularly quality management, planning and coordination," the report stated.

What blocks adoption?

The research is blunt on sequencing: AI does not fix bad data; it amplifies existing flaws. Unstructured data and workforce skill gaps compound the problem. For engineers, IT specialists and operational managers, the authors list prerequisites for meaningful use: data quality, system integration and understandable outputs.

"Successful adoption depends less on choosing the 'right' algorithm than on investing in data foundations, skills, governance and trust," the report stated.

The report also separates current evidence from aspiration. Evidence today is strongest for decision support and bounded automation. "The appropriate level of autonomy depends on the task, its risks, the maturity of the system and the consequences of failure," the authors wrote. "Human expertise remains essential for defining objectives, interpreting unusual situations, managing exceptions and ensuring accountability."

When do autonomous workflows arrive?

The authors estimate end-to-end autonomous AI workflows will become realistic within 2 to 3 years. That timeline matters for capacity planning and capital budgeting. Mills investing now in data foundations would reach that window ahead of competitors still reconciling siloed systems.

What changes for the workforce?

The paper predicts routine analytical, administrative and coordination activities will increasingly be performed or supported by AI, while human work shifts toward judgment, orchestration, exception handling and interpersonal responsibilities. That shift raises demand for digital literacy and the ability to evaluate AI outputs critically—a training cost executives should budget alongside software.

"The impact of AI on the textile industry will depend less on technological breakthroughs than on coherence—between data and systems, between human capabilities and machine autonomy, and between individual initiatives and collective action," the authors concluded.

For sourcing and manufacturing leaders, the near-term playbook is narrow: target inspection, maintenance, forecasting and energy, and fix the data underneath before scaling anything else.