The complete guide to AI in fashion merchandising and buying

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.

• 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.

Suppliers, brands and internal teams increasingly talk about models, agents and embeddings. A practical glossary for fashion buyers, with what each term means for buying decisions.

• Buyers do not need to understand how AI models are built, but they do need to understand what a tool's output means and how reliable it is.
• Terms such as forecast, confidence interval and back-test describe how certain a prediction is, which matters more than the prediction itself.
• Generative AI terms such as hallucination and prompt explain why AI-written content must be checked before it is relied upon.

Brands that sell through retailers see what they ship, not what consumers buy. Sell-out data closes that gap, and without it most AI forecasting for wholesale works half blind.

• Sell-in data records what a brand ships to retailers; sell-out data records what those retailers sell to consumers, by product, size, store and period.
• AI models for wholesale forecasting, re-orders and allocation need sell-out data because sell-in reflects retailer buying decisions and timing, not consumer demand.
• Standard formats exist: GS1 Austria's fashion guideline for the EDIFACT SLSRPT message transmits sales quantities by GTIN, outlet and period between trading partners.

Machine learning can improve fashion forecasts for carry-over lines and replenishment, but new styles, short seasons and trend shifts remain hard. What leaders should expect, and what not to.

• AI forecasting models learn from historical sales and external signals, and they are only as good as the history they are trained on.
• Forecasts for carry-over and never-out-of-stock items are generally far more reliable than forecasts for new seasonal styles.
• Recorded sales understate true demand when items sell out, so stock-outs must be corrected for before training any model.

Demand sensing updates short-term forecasts with the latest sales and external signals. In fashion it mainly improves in-season decisions: replenishment, transfers and markdowns.

• Demand sensing is short-term forecasting that uses recent, high-frequency data such as point-of-sale activity, shipments, promotions, weather and social trends to detect near-term demand shifts.
• Traditional forecasts are typically updated weekly or monthly from historical sales; demand sensing tools refresh predictions daily or even hourly.
• In fashion, demand sensing mainly improves in-season decisions such as replenishment, stock transfers and markdown timing, because most quantities are committed before the season starts.

Range planning decides how many styles, at which prices and in what depth a collection should have. AI can forecast new items and test scenarios, but the creative and strategic call stays human.

• Range and line planning translate a financial plan into a structured collection: number of options per category, price architecture, newness versus carry-over, and depth per option.
• AI contributes mainly through forecasting demand for new styles from attributes, images and similar past products, and through fast scenario comparisons of different range structures.
• Research on the VISUELLE dataset from Italian fast-fashion company Nunalie showed that adding Google Trends signals to image and attribute data improved new-product sales forecasts.

AI can sharpen depth, breadth and size decisions in assortment planning. The risk is a range that drifts towards safe averages. How to use the tools while keeping the brand's point of view.

• AI is strongest at the quantitative layer of assortment planning: depth, size curves, store clustering and identifying under-performing options.
• Models trained on past sales favour what has already worked, which can gradually erode newness and brand distinctiveness.
• Brands should protect a deliberate share of the range for strategic, image-building and experimental pieces that are judged on different criteria.

Open-to-buy tells buyers how much they can still spend without breaking sales and stock targets. AI can keep that number current and more realistic, provided the plan behind it is sound.

• Open-to-buy (OTB) is the value of inventory a business can still purchase in a period while meeting its sales and stock targets.
• A common formula is: planned sales plus planned markdowns plus planned end-of-period inventory, minus planned beginning-of-period inventory, minus stock already on order.
• AI improves OTB mainly by making its inputs better: more accurate sales and markdown forecasts, and faster re-forecasting during the season.

Fashion stock is perishable: too much ends in markdowns, too little in lost sales. Here is how AI models balance the two, what evidence exists and what they need to work.

• AI inventory optimisation combines demand forecasts at style, colour, size and location level with optimisation rules that decide how much stock to buy, hold, move or discount.
• Fashion is harder than most categories because many products are new each season, selling windows are short and demand is split across sizes and locations.
• In a peer-reviewed Interfaces paper, Zara's model for store shipments was reported to have increased sales by 3 to 4 percent, an early example of optimisation linking stock levels to demand.

Buying the wrong mix of sizes leaves fringe sizes on the rail and core sizes sold out. Here is how AI estimates true size demand, why stockouts mislead the data and what to watch for.

• A size curve is the percentage split of a buy quantity across sizes; size curve optimisation estimates that split from demand rather than habit.
• Raw sales are a biased basis for size curves, because sizes that sell out early stop recording sales while customers keep looking for them.
• Research on a footwear retailer found that nearly 25 percent of demand unmet by stockouts shifted to adjacent sizes, and that accounting for this could raise profit by up to 28 percent in low-demand settings.

Allocation decides where stock goes first; replenishment decides what follows. AI makes both decisions at size and location level. Here is how it works, what it needs and where it struggles.

• Allocation is the initial distribution of a delivery to stores, web and partner accounts; replenishment is the ongoing top-up of stock based on what sells.
• AI allocation forecasts demand at style, colour, size and location level and then optimises quantities under constraints such as pack sizes, store capacity and transport schedules.
• Zara's store replenishment model, documented in Interfaces in 2010, chose shipment quantities by linking stock levels to demand and was reported to have increased sales by 3 to 4 percent.

Markdown optimisation uses demand models to decide when and how deeply to discount. How it works, what it needs, and why brand and wholesale relationships must set its limits.

• Markdown optimisation models estimate how demand responds to price over time and recommend the timing and depth of discounts to meet a stock or margin objective.
• The objective must be defined explicitly, for example maximising margin while clearing stock by a set date, because different goals produce different recommendations.
• Reliable price response estimates require clean history of past price changes, promotions and stock levels.

Open-to-buy tells a buyer how much stock can still be purchased without breaking the plan. Here is how it is calculated, used and commonly misused.

• Open-to-buy is the purchasing budget left for a period after planned sales, markdowns, target closing stock, opening stock and existing orders are accounted for.
• Planned purchases equal planned sales plus planned markdowns plus planned closing stock, minus opening stock; open-to-buy deducts orders already placed.
• Open-to-buy should be tracked by delivery month and reforecast regularly as actual sales come in.

Three metrics answer most questions about how fashion stock is performing. How to calculate sell-through, GMROI and weeks of cover, and read them together.

• Sell-through measures the share of stock sold in a period and is the main signal for re-order and markdown decisions.
• GMROI divides gross margin by average inventory at cost, combining margin and stock turn into one measure of return on stock investment.
• Weeks of cover divides current stock by recent weekly sales and is the forward-looking metric for replenishment and allocation.

From the factory cost to the price on the swing tag: how landed cost, wholesale markup and retail markup combine, and why markdowns decide what margin is really earned.

• Fashion margin is built in layers: product cost, landed cost, wholesale price, retail price and finally the price at which the item actually sells.
• Markup and margin are different measures, and confusing them is one of the most common errors in pricing discussions.
• Landed cost includes freight, duties, insurance and handling, and ignoring these lines overstates brand margin.