AI in fast fashion: speed, demand signals and the overproduction problem
Fast fashion lives on reading demand early and reacting fast. AI sharpens both, but it can also accelerate volume, and new EU rules make unsold stock more costly.
KEY TAKEAWAYS Summary by the editors
- In fast fashion, AI is mainly used to detect demand signals early, decide what to replenish or drop, and produce large volumes of product content quickly.
- Shein describes an on-demand model in which new designs are made in small batches of 100 to 200 pieces and customer demand signals decide whether a design is replenished or retired.
- From 19 July 2026, large companies in the EU may no longer destroy unsold apparel, clothing accessories and footwear under the Ecodesign for Sustainable Products Regulation, with medium-sized companies expected to follow by 2030.
- AI can reduce overproduction only if it is used to buy less and test more; used to launch more styles faster, it can increase total volume.
- Inditex reported more than 7 million sessions of its AI-based Zara Try-on feature in 43 markets between its December launch and its March 2026 results, according to press coverage of those results.
AI in fast fashion is used to read customer demand early, to decide quickly which styles to scale, replenish or drop, and to produce product imagery and copy at high volume. The segment differs because speed and volume are the business model: small forecasting improvements move large quantities, and the same tools that cut waste can also be used to push more product to market.
Why does AI matter differently in fast fashion?
Fast fashion competes on newness, low prices and short lead times. Ranges are wide, product lives are short and margins per item are thin, so the profit depends on selling a high share at full price before trends move on. Every week a forecast is late, and every unit bought without demand, erodes that margin.
That makes demand sensing (detecting demand from early sales, searches and engagement rather than from last season's history) the core AI capability. It also explains why the segment attracts criticism: AI-driven speed can be used to reduce waste through smaller, test-and-repeat buys, or to multiply styles and volume. Which one happens is a management decision, not a property of the technology.
What are the main AI use cases in fast fashion?
| Use case | Why it matters in this segment | Example (only if verified) | Maturity |
|---|---|---|---|
| Demand sensing and test-and-repeat buying | Short product lives leave no time for slow forecasts | Shein: small initial batches of 100 to 200 pieces, demand signals decide replenishment (company statement) | Established |
| Replenishment and allocation | Thin margins require stock in the right store or warehouse quickly | No detailed verified case used here | Established |
| Trend detection from search and social data | Newness is the core promise | No verified case used here | Emerging |
| Virtual try-on | Large online volumes and high return costs | Zara Try-on, over 7 million sessions in 43 markets (results coverage, March 2026) | Emerging |
| Generative imagery and digital twins of models | Thousands of new products need imagery fast | H&M digital twins of 30 consenting models (FashionUnited, March 2025) | Emerging |
| Markdown optimisation | Unsold stock is costly and destruction is being restricted in the EU | No verified case used here | Established |
How does AI use demand signals in fast fashion?
The most discussed model is Shein's. In its own description of its on-demand business model, the company says new designs are created in small batches of 100 to 200 pieces to measure demand, that customer demand signals guide whether designs are retired or replenished, and that it passes demand signals directly to supplier partners. Shein also claims this reduces inventory waste to single-digit percentages; that figure is a company claim that has not been independently verified.
The underlying mechanics are common across the segment:
- Launch a style in a small quantity, online first where possible.
- Measure early signals: views, add-to-basket rates, sell-through in the first days, returns and reasons.
- Use models to predict which styles will continue and in which sizes and regions.
- Reorder quickly from nearby or agile suppliers, or retire the style.
- Feed the results back into design and buying decisions.
This only works if suppliers can respond within days or a few weeks. Without flexible sourcing, a better forecast simply arrives too late to act on.
Can AI help fast fashion reduce overproduction?
It can, but only under specific conditions. AI reduces overproduction when it leads to smaller initial buys, faster cancellation of weak styles and better size curves. It does not reduce overproduction if the efficiency gained is reinvested into launching more styles. Total units produced, not the sell-through rate of each style, is the measure that matters for environmental impact.
Regulation is raising the cost of getting this wrong. According to the European Commission, the destruction of unsold apparel, clothing accessories and footwear is banned for large companies from 19 July 2026 under the Ecodesign for Sustainable Products Regulation, with medium-sized companies expected to comply by 2030. Companies must also disclose information on unsold products discarded as waste, with a standardised format from February 2027. Stock that cannot be sold therefore has to find another outlet, which makes accurate buying more valuable.
How is generative AI changing fast fashion content?
Fast fashion retailers publish thousands of new products each week, each needing images, copy and translations. Generative AI shortens this work. H&M announced in March 2025 that it would create digital twins of 30 models who consented to take part. According to FashionUnited, H&M said the models retain control over the use of their digital replicas, are paid in line with traditional shoots, and that initial images would carry watermarks identifying them as AI-generated.
Inditex has applied AI on the customer side. Coverage of its March 2026 annual results reported that Zara Try-on, which lets customers create an avatar from their own photos and see it wearing real products, had recorded more than 7 million sessions since December and was live in 43 markets, with a roll-out to other Inditex brands.
What risks are specific to fast fashion?
- Rebound effect: efficiency gains used to increase total volume rather than reduce waste.
- Design copying: trend detection and generative design can drift into reproducing other designers' work.
- Supplier pressure: faster reorder cycles can push risk and overtime onto factories.
- Model rights and representation: digital twins require consent, payment and clear labelling.
- Regulatory exposure: EU rules on unsold stock raise the cost of misjudged buys.
What should a fast fashion business do first?
Start with the decision that moves the most units: the initial buy quantity and the reorder trigger. Build a demand sensing pilot on one category, connect it to suppliers who can actually react, and set a target for total units and unsold stock, not only for forecast accuracy. Clean size-level sales and returns data is the prerequisite.
In parallel, set rules for generative content: consent and payment for any model likeness, labelling, and checks that generated designs do not reproduce existing ones. Speed is the segment's advantage; using AI to buy less and sell more of what is bought is how that advantage survives tighter regulation.
Frequently asked questions
How does Shein use data and AI?
Shein describes an on-demand model in which new designs are produced in small batches of 100 to 200 pieces and customer demand signals decide whether they are replenished or retired. It says it passes these signals directly to suppliers. Its claims about low inventory waste have not been independently verified.
Can AI reduce overproduction in fast fashion?
Yes, if it is used to make smaller initial buys, cancel weak styles quickly and improve size curves. If the efficiency is used to launch more styles, total volume can rise instead. Companies should track total units and unsold stock, not only forecast accuracy.
Is it legal to destroy unsold clothes in the EU?
Under the Ecodesign for Sustainable Products Regulation, large companies may no longer destroy unsold apparel, clothing accessories and footwear from 19 July 2026, with limited exceptions such as safety reasons or damage. Medium-sized companies are expected to comply by 2030.
Does H&M use AI models?
In March 2025 H&M announced it would create AI digital twins of 30 models who consented to take part. H&M said the models keep control over their replicas and are paid in line with traditional shoots, and that the first images would be watermarked as AI-generated.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
SOURCES
- SHEIN Group: On-Demand Fashion
- European Commission: New EU rules to stop the destruction of unsold clothes and shoes
- FashionUnited: H&M to create twins of models with AI: The possibilities are almost endless
- Express & Star: Sales jump for Zara owner Inditex as it taps into AI for virtual changing rooms
