What Difference Does It Make? Where AI Is Actually Working in Fashion

A Helmut Lang jacket appears on a resale site. It could be worth $80 or $800, depending on whether it was designed by Lang himself, by the anonymous studio team that followed his departure, or by Peter Do. The label reads Helmut Lang in all three cases. Most AI tools cannot tell the difference, and most human resellers cannot either — not without ten minutes of research per item anyway. Multiply that by a few hundred garments a day and you arrive at the basic problem: fashion’s circular economy is running on infrastructure that was not built for the volume it now carries.

AI is supposed to fix these types of problems. The question is whether it actually does, and where along the value chain its impact is real rather than rhetorical.

Upstream
Start at production. Levi Strauss uses AI-driven demand forecasting to reduce overstock at its source while Stella McCartney has partnered with biotech firms using AI-designed enzymes for biological fabric recycling. Style3D and CLO 3D eliminate the need for physical samples, compressing development timelines from months to days.

T-Fashion AI takes a different angle: trend forecasting trained on fashion-specific data – runway archives spanning over 1,500 designers and real-time social signals – rather than repurposed generic models. Its system produces demand forecasts segmented by demographic and geography, feeding them into a generative design studio where concepts become production-ready visuals in minutes. For an industry that produces 92 million tonnes of waste annually, according to the World Economic Forum, much of it from misaligned production bets, sharper forecasting is not incremental. It intervenes where overproduction decisions are actually made.

These are real advances. But they are also, overwhelmingly, brand-level tools – accessible only to companies with the budgets to integrate and advertise them. That asymmetry matters.

Unitree Robotics’ humanoid robot Unitree G1

The Gap
Fashion’s sustainability conversation fixates on production: eco-collections, recycled fibres, carbon pledges. The resale ecosystem – projected by ThredUp and BCG to reach $367 billion by 2029, growing nearly three times faster than the broader apparel market – attracts far less attention. The operators keeping garments in circulation are mostly small businesses whose constraint is not ideology but logistics.

Is the imbalance justified? If better forecasting prevents a million garments from being manufactured, that may outweigh any resale improvement downstream. But overproduction is not going away soon – McKinsey’s State of Fashion report makes clear the structural incentives remain in place – and the resale infrastructure absorbing what is already produced is badly under-tooled. Both claims can be true. The question is whether investment in one should come at the expense of the other. Right now, it largely does at the expense of downstream solutions.

Downstream
Listing an item for resale across eBay, Depop, Poshmark, Etsy, Grailed, and Shopify – each with its own taxonomy, required fields, and image standards – takes ten to fifteen minutes per garment. At scale, listing labour becomes one of the largest barriers to growing a resale business.

This is one of the problems Secnd, which I co-founded in 2024, was built to address. I am not a neutral observer, but the data is worth examining. The platform generates AI listings from a photograph and cross-posts across eight marketplaces. In deployments with Toronto-based vintage retailers: one operation saw a 33% sales increase within two weeks, processing 100 items in a single hour; another averaged 1.5 minutes per item including quality control, roughly 10x faster than before; a third, almost entirely offline previously, saved over 40 hours monthly while digitising 200+ pieces.

The limitations are real. New resellers with no listing history get less immediate benefit. Niche items still need human judgment. Authentication is not addressed. And the sustainability argument – faster listing means more garments in circulation – holds directionally but involves steps no single platform fully controls.

On the pricing side, UK-based TRUSS identifies garments from a photograph against 200 million+ items, returning pricing history and sell-through rates in under two seconds from 25+ resale platforms. Its benchmarking claims generative AI chatbots disagree with real market prices by 23% on average. TRUSS has partnered with Depop and Selfridges and secured a £1.1 million Innovate UK grant. As Business of Fashion noted in its 2024 profile, the bottleneck in resale is not just listing speed but product knowledge.

Archive powers branded resale for The North Face, Lululemon, and New Balance, and its data shows roughly half of a brand’s resale shoppers are new to the brand – reframing resale as an acquisition channel rather than a sustainability obligation. SECONDSENSE, backed by Outlander VC according to The Fashion Law’s funding tracker, surfaces real-time market values for secondhand luxury handbags across platforms where the same bag can vary by hundreds of dollars.

A Tool, Not a Narrative
T-Fashion, TRUSS, Archive, SECONDSENSE, Secnd – different positions in the value chain, different problems. None closes the loop alone.

But there is a useful test: does a given AI application change a measurable behaviour? A reseller listing 100 items per hour instead of 15 sources differently and holds less dead stock. A brand discovering half its resale customers are new allocates its budget differently. A pricing engine correcting a 23% gap changes which items get listed and at what margin. These are not transformations. They are operational shifts, and they compound.

What none of these tools answer is whether the fashion industry produces too much to begin with. If production volumes keep rising, even a perfectly functioning resale ecosystem absorbs only a fraction of the output. The most honest assessment of AI in fashion right now is that it is making specific parts of the value chain measurably better, while the system those parts belong to has not yet decided whether it wants to change.

By Eduard Härtel

Sources: