AI in fashion merchandising: what it does, what it needs and where it fails
Machine learning now supports forecasting, assortment, pricing and planning decisions in fashion. This guide explains the use cases, the data they depend on and how merchants stay in control.
KEY TAKEAWAYS Summary by the editors
- AI in fashion merchandising means using statistical and machine learning models to support decisions on what to buy, how much, in which sizes, where to place it and at what price.
- The most mature applications are demand forecasting, allocation and replenishment, and markdown pricing; generative AI is mainly used for analysis, reporting and content around these decisions.
- Large retailers such as Mango and Marks & Spencer have announced programmes to replace fragmented planning tools with integrated, AI-supported planning across financial, assortment and demand planning.
- Zalando has published that a transformer-based demand forecasting model, used for pricing decisions, has been in use in different versions since 2019 with a 26-week horizon.
- AI output is only as good as the product, sales and stock data behind it, so clean article master data and sales at size level are prerequisites, not optional extras.
AI in fashion merchandising is the use of forecasting and optimisation models to support the core merchant decisions: what to buy, how much, in which sizes, where to send it and when to reduce the price. It does not replace the merchant's judgement on product and brand; it gives that judgement better numbers, faster, and at a level of detail (style, colour, size, store, week) that spreadsheets cannot handle.
What is AI in fashion merchandising?
Merchandising sits between product creation and selling. A merchandiser or buyer turns a financial plan into a range, a range into quantities, and quantities into stock positions across stores, e-commerce and wholesale accounts. Each step involves predicting demand for products that often have no sales history, because a large share of a fashion range is new every season.
"AI" in this context usually means three families of technology. Predictive models (machine learning forecasting) estimate demand by product, location and period. Optimisation models turn those forecasts into decisions under constraints such as budget, minimum order quantities and store capacity. Generative AI (large language models) is newer and is mainly used to explain, summarise and query planning data in natural language, or to produce product content.
Which merchandising decisions can AI support?
| Decision | What the model does | Maturity |
|---|---|---|
| Financial planning and open-to-buy | Projects sales, margin and stock by category and month; flags budget gaps | Established in planning suites, AI layer emerging |
| Range and assortment planning | Estimates demand for new styles from attributes, images and analogous products | Emerging, quality varies widely |
| Buy quantities and size curves | Forecasts demand by style, colour and size; corrects for stockouts | Established at larger retailers |
| Allocation and replenishment | Distributes stock to stores and channels based on local demand | Established, one of the oldest optimisation uses |
| Markdown and pricing | Models how demand reacts to discounts to clear stock profitably | Established at large online retailers |
| Reporting and analysis | Answers questions about sales and stock in natural language | New, generative AI pilots |
Published examples show where large players have focused. Zalando's research team has described a demand forecasting model based on the transformer architecture that forecasts weekly demand for a horizon of 26 weeks and feeds pricing decisions; the paper states it has been in use at Zalando in different versions since 2019. The authors explain that in online fashion, pricing is a major lever to reduce the risk of being overstocked at the end of a season.
Other retailers are rebuilding the planning backbone. In March 2025 Mango announced it would modernise merchandise financial planning, assortment planning and demand planning with an AI-powered planning platform, replacing what it called fragmented, outdated systems across its nearly 2,850 stores in 120 markets. In January 2024 Marks & Spencer announced a three-year programme for clothing and home covering merchandise planning, range planning and later forecasting and replenishment, with its clothing and home managing director saying an end-to-end platform would help the business "better use data and AI".
What data does AI merchandising need?
Most failed AI projects in merchandising are data projects in disguise. Models need a consistent product hierarchy, clean attributes, and history at the level where decisions are made. In practice that means:
- Product master data: style, colour, size, category, fabric, price point and season, with stable identifiers from one season to the next.
- Sales history at size and location level: by store, web and, for wholesale brands, by account where retailers share it.
- Stock and availability history: without it a model cannot tell low demand from sold-out sizes.
- Prices and promotions: full price, markdown dates and discount depths, so the model can separate price effects from underlying demand.
- Plans and constraints: budgets, minimum order quantities, lead times and store capacities.
The quality bar is high. If the same jacket carries different codes in the planning tool and the sales system, or if attributes are filled inconsistently, a model will learn the wrong lessons from history.
How does AI change the merchandiser's role?
The realistic shift is from compiling numbers to reviewing and challenging them. Instead of building a buy plan in a spreadsheet, the merchandiser receives a recommended quantity per style and size, with the drivers behind it, and decides where to override. Overrides matter: they encode knowledge the model cannot see, such as a planned celebrity collaboration or a supplier issue. Good practice is to log overrides and review afterwards whether the human or the model was closer.
Generative AI adds a conversational layer. McKinsey and The Business of Fashion reported in The State of Fashion 2026 that more than 35 percent of executives surveyed already use generative AI in areas such as customer service, image creation, copywriting and product discovery. In merchandising, the equivalent use is asking a planning system questions in plain language and getting a summary, which is useful but only as reliable as the underlying data and the guardrails around it.
What are the limits and risks?
- New products without history. Forecasts for genuinely new styles rely on similar past products and attributes, and error is higher than for carry-over items.
- Bias towards the past. Models trained on history reward what sold before and can under-invest in new directions unless merchants deliberately test them.
- Hidden lost sales. If stockouts are not modelled, sold-out sizes look like weak sellers and the error repeats every season.
- Black-box recommendations. Planners ignore numbers they cannot explain; transparency about drivers is a practical requirement.
- Cost and change effort. Integration, data clean-up and process change typically take longer than the model build itself.
How should a fashion business get started?
Start with one decision where the payback is measurable and the data already exists, typically replenishment of carry-over items, size curves or in-season markdowns. Define a baseline (current forecast error, sell-through, markdown rate), run the model alongside the existing process for a season, and compare. Only then extend to harder problems such as new-style forecasting and range planning.
Ownership should sit with merchandising, not IT alone. The teams that benefit most treat the model as a colleague whose work is reviewed weekly, with clear rules on when to accept, adjust or reject its recommendations.
Frequently asked questions
What is AI merchandising in fashion?
It is the use of forecasting and optimisation models to support decisions on assortment, buy quantities, size curves, allocation and pricing. The models work at style, colour, size and location level and produce recommendations that merchandisers review and adjust.
Will AI replace fashion buyers and merchandisers?
Current deployments support rather than replace them. AI automates calculation and analysis, while decisions about brand direction, trend bets and supplier relationships stay with people, who also need to review and override model output.
What data do I need before using AI for merchandising?
At minimum, clean product master data, sales history at size and location level, stock and availability history, and price and markdown history. Without stock history a model cannot distinguish low demand from sold-out sizes.
Which merchandising AI use case should a brand start with?
Usually a repeatable decision with good data, such as replenishment of carry-over items, size curves or markdowns. These make it easy to measure results against a baseline before tackling new-product forecasting.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
SOURCES
- arXiv: Deep Learning based Forecasting: a case study from the online fashion industry (Zalando)
- Retail TouchPoints: Mango Ramps Up Data-Driven Planning for Global Operations
- Just Style: M&S to digitally transform clothing and home end-to-end planning system
- McKinsey & Company: The State of Fashion 2026
