Can AI help the secondhand textile sector prove its value?

The secondhand textile sector has a wealth of operational records, but no easy way to turn them into a collective account of its contribution. That was the challenge Chris McLellan addressed in his presentation, “Closing the Data Gap: The Case for Data Collaboration in the Age of AI & EPR,” at the SMART 2026 Annual Convention in Montréal. In an interview with FashionUnited following the event, he explained how AI could help independent businesses assemble that evidence, and why trust matters more than the technology.

For McLellan, the missing information is how much material businesses handle, where it goes, how much is reused or recycled, and how many jobs their activities support. Independent collectors, sorters, graders and exporters hold pieces of that picture, but their records remain dispersed across filing cabinets, spreadsheets and accounting systems.

“Multiply that across an industry and the evidence exists, it’s just scattered,” he said. The consequence, he argues, is a sector whose economic and environmental contribution is difficult to present convincingly when policymakers ask for numbers.

According to him, AI could help bridge that gap by extracting agreed information from documents businesses already produce, including scale tickets, invoices, container packing lists and export declarations.

McLellan proposes making existing paperwork easier to use collectively. A business could upload a PDF or photograph a document, with an application extracting selected details such as weight, grade, origin and destination. The system could standardize formats and convert units, helping records from different operators become comparable.

To demonstrate the approach, McLellan built a basic prototype for his SMART presentation. With AI’s help, he said, it took less than an hour and cost virtually nothing. That demonstrates a possible starting point, rather than a fully operational, sector-wide reporting system.

He favors beginning with a visible process in which businesses review and approve extracted information before submission. More autonomous AI agents could eventually reduce the workload further, but their adoption would require greater confidence in the system.

Trust before technology

The more difficult obstacle is persuading competitors to participate. Information about volumes, buyers, grades and prices is commercially sensitive, and business owners have legitimate concerns about who might access it.

“Owners worry, very reasonably, that shared data could be seen by competitors, used against them, or simply slip out of their control,” McLellan said.

His proposed approach begins with minimizing the information collected, removing identifying details at source, and let contributors decide which projects their data can support. A company might permit its information to inform policy advocacy without agreeing to its use in industry benchmarking.

McLellan described this as a “Bill of Rights for Data Owners.” Contributors should also be able to withdraw their information, he argues, while copies should be minimized so that control remains meaningful.

An association or nonprofit would be the natural organizer, in his view, but participation should rest on clear governance rather than institutional reputation alone. He advocates a written charter, independent scrutiny and rules that prevent larger businesses from dominating.

McLellan sees particular urgency around extended producer responsibility, or EPR, and decisions affecting textile exports. His argument is that businesses need evidence of their activities if they want those activities properly reflected in policy and funding discussions.

“Who receives money from EPR programs is an economic question. Whether export markets remain open is a demand question,” he said. “Where infrastructure money goes depends on what volumes people believe exist.”

Data collaboration would not resolve the sector’s underlying commercial pressures. McLellan acknowledges shipping costs, infrastructure constraints, contamination and market demand as substantive barriers. “Data itself doesn’t sort a single bale,” he said. Its purpose is to strengthen the industry’s participation in decisions about those problems.

AI needs checking, too

Using AI to produce regulatory evidence brings risks. McLellan warns that systems can misread documents, confuse units or invent missing information. A mistake repeated across submissions could undermine confidence in the wider dataset.

He therefore calls for every data entry to remain traceable to an original document held by the contributing business. Human approval before submission and checks for anomalies would provide additional safeguards.

There is another risk, however: leaving the industry poorly documented while AI systems increasingly mediate how people learn about it. McLellan worries that limited studies can become disproportionately influential when broader evidence is unavailable.

“When an industry has no numbers, other people’s numbers become evidence by default,” he said.

His case is for collective visibility without surrendering individual control. AI could make assembling the evidence easier, but businesses need a reason to contribute, and confidence that doing so will not compromise them.