AI in Textiles: Why Data Integration Comes First

The barrier to artificial intelligence in textiles and clothing is not the technology. It is the groundwork underneath it.

A white paper produced by the International Textile Manufacturers Federation (ITMF), based in Zurich, together with the Institut für Textiltechnik (ITA) of RWTH Aachen University, identifies poor data integration, skills shortages and governance gaps as the decisive obstacles. The findings were presented on 15 October at a conference in Fortaleza, Brazil, staged by ITMF with the Brazilian textile and apparel association ABIT and the International Apparel Federation. The work draws on ten webinars held between 2024 and 2026, involving more than 20 experts from companies, research institutes and policy bodies.

The central argument is that the industry’s problem is not a lack of digital investment. After decades of computerisation, textile businesses generate enormous volumes of information. What they lack is coherence: systems that talk to one another, standards that hold across departments and supply chain partners, and data clean enough to act on. The paper’s phrase for this condition is blunt — the sector is not under-digitalised, but digitally fragmented.

Why the pressure is structural, not fashionable

Interest in AI across the value chain is being pushed by forces more durable than enthusiasm for new tools. Demographic decline is shrinking the available workforce in many producing regions. Skills shortages persist in technical and supervisory roles. Markets swing quickly, and buyers expect more on quality, traceability and sustainability than legacy processes can reliably deliver.

AI appeals in this context because it can absorb complexity and stretch scarce expertise further — but only where the underlying information is trustworthy. Participants in the research were direct about this: AI magnifies the defects in poor data instead of repairing them. Many businesses also hold far less usable, structured information than they believe they do.

Where results are already visible

None of this means AI is unproven. In tightly defined applications, the returns are measurable. Computer-vision inspection of fabric and yarn is a mature example, catching defects faster and more consistently than manual checks. Predictive maintenance keeps critical machinery running. Energy optimisation trims consumption, and demand forecasting sharpens planning.

One case cited from a nonwovens production line showed causal AI cutting scrap by 10 per cent. That kind of figure matters because it demonstrates that AI performs best where the problem is narrow, the data is specific and the outcome can be measured.

From support to supervised action

The frontier is shifting from advice to execution. Agentic systems are beginning to coordinate multi-step processes such as sourcing, while causal AI adjusts production settings directly. In both cases, humans remain in the loop. The paper’s authors judge that genuinely end-to-end AI workflows are realistic within two to three years.

The conditions for trust

Industry voices in the research set out clear terms for adoption. Companies want control over sensitive data, robust cybersecurity, transparency about what a system can and cannot do, and continued human accountability for decisions. Without these, enthusiasm stalls at the pilot stage.

A practical route forward

Rather than chasing scale immediately, the paper recommends a sequence. Put people first — train and recruit for the skills that data work demands before rolling technology out. Treat data as shared infrastructure rather than a private hoard. Begin with a small number of limited, well-chosen use cases where success can be proven.

Collective action carries particular weight for smaller firms that cannot build everything alone. Shared data standards lower the cost of connecting systems. Data spaces that keep ownership with the originator allow collaboration without surrendering commercial advantage. Sector-specific tools, such as a ‘Textile GPT’ tailored to industry language and knowledge, are also held up as a realistic goal for associations and pre-competitive partnerships.

Olivier Zieschank, director at ITMF, framed the message plainly: AI is not a shortcut around the structural challenges the industry faces, but it can help companies manage complexity and make the most of scarce expertise. Thomas Gries, director of ITA at RWTH Aachen University, added that firms need a dedicated task force to embed data integration into their technical and business processes, and to work with external experts on a transparent assessment of their data and automation.

What this means in practice

For manufacturers and brands weighing up where to start, the implications are concrete:

• Audit your data before you buy tools. Establish what is structured, accessible and reliable — and what is not.
• Choose bounded problems. Inspection, maintenance, energy and forecasting have a track record; sprawling transformation programmes do not.
• Invest in skills early. Data literacy and process understanding are prerequisites, not afterthoughts.
• Set governance in advance. Decide who controls data, how security is managed and where human sign-off sits.
• Engage beyond your own walls. Standards, shared data spaces and industry tools reduce cost and risk, especially for smaller operations.

The technology is ready enough. The work that decides whether it succeeds lies in integration, people and rules — the less glamorous foundations on which everything else stands.