Fashion's speed problem
The winners in fashion over the last thirty years were not the best designers. They were the fastest readers of demand. Agentic AI hands that speed to every brand, not only to the few that built a supply chain around it.
Speed already decided the industry once
In the mid-2000s the standard design-to-retail cycle was five to six months. Inditex did it in five weeks, and needed two weeks for repeat orders. While competitors shipped to stores every twelve weeks, Inditex did so twice a week (Crofton and Dopico, 2007).
| Model | Design to shelf |
|---|---|
| Outsourced East Asian subcontracting | up to 8 months |
| Industry standard | 5 to 6 months |
| Inditex, new design | 5 weeks |
| Inditex, repeat or slight change | 2 weeks |
Shein then moved the contest from the factory to the data. It starts each new product with only 100 to 200 units and reorders what sells (Retail TouchPoints, 2023). In the second half of 2021 it added between 2,000 and 10,000 new styles a day (Rest of World).
The lesson is not that every brand should become fast fashion. It is that the brands which learn fastest from real demand waste the least and sell the most.
The cost of being slow
Most brands still plan a season ahead and find out what worked after the markdowns. The cost shows up in three places.
- Stock that does not sell. Fashion inventory days on hand in 2024 were 14% above pre-2020 averages (McKinsey, February 2026).
- Returns from fit. About 70% of apparel returns are caused by poor fit or style (McKinsey, 2021). In the US, processing online apparel and footwear returns cost about USD 25.1bn in 2023 (Coresight).
- Content that lags the product. 54% of shoppers have abandoned a sale because product information was inconsistent across sites, and 71% have returned an item because the content was wrong (Salsify 2025). Baymard found 82% of apparel sites give too little sizing information (Baymard).
Why catalogue and assortment must now move in days
Three forces make the old calendar untenable.
- Discovery has moved to AI. In McKinsey's June 2026 survey of about 21,000 consumers, 53% had used AI for shopping in the past three months, and apparel led the categories where people completed a purchase with AI (McKinsey, October 2026). An AI shopping agent can only recommend what it can read. A product with missing attributes is invisible to it.
- Channels multiply the catalogue. Every marketplace has its own rules. Myntra asks for 3:4 images of at least 1080 by 1440 pixels; Flipkart apparel needs a light grey background with a soft shadow and front and back views (Fynd documentation). One product becomes many listings.
- Content is now the shelf. "If content was king, it's now going to be an emperor," as McKinsey's Anita Balchandani put it (McKinsey). Zalando took AI-generated product content from near zero to 90% in 2025 and cut campaign creation from six weeks to days (Zalando).
What speed with agents looks like
| Work | The old rhythm | With agents |
|---|---|---|
| Read demand | Season reviews and sell-through after markdown | A weekly city trend board from the brand's own shoppers, press and social |
| Decide what to design | Mood boards a season ahead | Concepts drafted where a theme was asked for again and again, tried by shoppers before sampling |
| Catalogue a drop | Days of tagging, shoots and copy per batch | Tags proposed from photos daily; shoots, copy and listings drafted on request |
| Fix the assortment | Monthly reports, broken sizes found late | A weekly merchandise brief by attribute, size curve and store cluster |
| Learn what failed | Returns data months later | "Asked for, not carried" and "tried, not bought" every week |
BCG's project experience puts numbers on the shape: 50% faster time to market in product design, 10 times faster assorting, three times the campaign speed (BCG, November 2025). McKinsey estimates generative AI could add USD 150bn to 275bn to the operating profits of apparel, fashion and luxury (McKinsey, 2023).
And yet most projects stall
McKinsey's Colleen Baum notes that "about 90 percent of projects have stalled at the pilot phase" in fashion (McKinsey, 2026). The reason is rarely the model. It is that point tools sit outside the way a fashion company actually works: across design, merchandising, catalogue, marketing and the shop floor, with a person approving what goes out. The next essays are about that gap.