Clear definitions of the terms that shape fashion wholesale, retail, technology and sustainability.
Software that models how a garment's fabric drapes, stretches and fits on a virtual body, based on patterns and material properties.
3D simulation lets designers and technologists review fit and proportions before cutting a physical sample. A pattern change can be checked on several avatar sizes in minutes. The results also feed digital showrooms and virtual try-on, though fabric data must be accurate for realistic results.
An independent sales representative who sells a brand's collection to retailers in a territory in return for commission.
Agents usually represent several non-competing brands and run their own showrooms. They bring established relationships with local stockists, which helps brands enter new markets without building their own sales team. The agent takes orders, but the brand invoices and ships to the retailer directly.
A process in which AI agents plan and carry out several steps, using tools and data, with limited human input.
Rather than answering a single question, an agentic workflow pursues a goal. For example, an agent could check sell-through data, draft a re-order proposal, verify stock in the ERP and send it for approval. Clear permissions, logging and human checkpoints are essential because errors can compound across steps.
An AI system that can plan and carry out multi-step tasks, using tools and data, with limited human instruction.
In fashion, an AI agent might monitor a retailer's sell-through, propose re-orders and draft the order in a B2B portal for a buyer to approve. Agents typically combine a language model with access to business systems through APIs. Clear permissions and human review are important safeguards.
Systematic unfairness in AI outputs, often caused by unbalanced training data or design choices.
Bias can appear when a model reflects gaps or stereotypes in its data. An image generator might default to a narrow range of body types or skin tones, and a size recommendation tool might perform worse for less common sizes. Diverse data, testing across groups and human review help reduce it.
The policies, roles and controls an organisation uses to manage how AI is selected, built, used and monitored.
AI governance answers who may use which tools, with what data, and who is accountable for outcomes. A brand might keep an inventory of AI systems, require review before AI-generated images are published and define rules for sharing customer data. It also prepares the company for regulatory requirements such as the EU AI Act.
Images created wholly or partly by generative AI, such as product shots, model photos or campaign visuals.
Brands use AI-generated imagery to produce lifestyle backgrounds, show products on different models or create concept visuals quickly. Images must represent the actual product accurately, especially colour and fit, to avoid misleading buyers and consumers. Labelling and content credentials are increasingly expected for transparency.
Identifying data points or events that differ markedly from normal patterns, such as unusual orders or errors.
Anomaly detection helps spot problems early. It can flag a re-order that is ten times a retailer's usual quantity, a sudden spike in returns for one size, or an EDI file with implausible prices. Flagged cases are usually reviewed by a person rather than blocked automatically.
Application Programming Interface: a defined way for software systems to exchange data and trigger functions automatically.
APIs let a B2B portal read live stock from the ERP or send confirmed orders back without manual export and import. They are also the basis of most marketplace and retailer integrations. Compared with file-based exchange, APIs allow near real-time data flows.
Advance Shipping Notice: an electronic message telling a retailer what is being shipped, in which cartons, and when it will arrive.
An ASN is sent before goods arrive so the receiving warehouse can plan labour and check deliveries quickly. When a carton's barcode is scanned, its contents are matched against the ASN instead of being counted by hand. Inaccurate ASNs can lead to chargebacks from the retailer.
The range of products a retailer or brand offers, defined by categories, styles, colours, sizes and price points.
Assortment planning balances breadth (how many different styles) against depth (how many units per style). A flagship store might carry the full assortment while a smaller partner takes a curated selection. Good assortments reflect the target customer, the store format and the season.
An online platform where retailers can view a brand's products, check availability, place orders and manage their account.
A B2B portal moves routine wholesale tasks such as re-orders, NOS replenishment and invoice downloads into self-service. Retailers see their own prices and conditions after logging in. It complements, rather than replaces, the personal relationship with sales representatives.
The function responsible for selecting products and placing orders with brands and suppliers for a retail business.
Buyers attend showrooms and trade fairs, judge collections and negotiate prices, minimums and delivery windows. They work within budgets set with merchandising, often expressed as open-to-buy. A buyer's choices largely determine what customers will see in store each season.
A small, focused collection built around a theme, collaboration or occasion, sold alongside the main range.
Capsules allow brands to test ideas, respond to trends or partner with designers without committing to a full collection. A brand might release a ten-piece capsule for the holiday season. Because volumes are small, capsules often sell through quickly and attract media attention.
A style that continues unchanged, or with minor updates, from one season into the next.
Carry-overs give retailers continuity and reduce development costs for the brand. A best-selling trouser might be carried over for several seasons, perhaps in new colourways. Carry-overs sit between seasonal fashion items and permanent NOS products.
An approach that keeps products and materials in use for as long as possible through durability, repair, reuse and recycling.
In a circular fashion model, a jacket is designed to last, repaired when damaged, resold when no longer wanted and finally recycled into new fibre. This contrasts with a linear take, make and dispose model. Circularity requires changes in design, business models and infrastructure.
A machine learning task that assigns each item to one of a set of predefined categories.
Classification models answer questions such as 'is this a dress, skirt or top?' or 'is this order at risk of cancellation?'. Automatic attribute tagging of product images is a common classification use in fashion. Its counterpart for continuous numbers, such as predicting a sales quantity, is called regression.
A machine learning method that groups similar items together without predefined categories.
Clustering finds structure in data that no one labelled in advance. A brand might cluster its retail accounts by buying behaviour and discover groups such as trend-led boutiques and replenishment-focused department stores. These groups can then shape assortment, terms and sales approach.
A recommendation method that suggests items based on what similar users chose, rather than on product attributes.
The underlying idea is that people who agreed in the past will agree again. If retailers who bought a certain knitwear style also tended to order a particular trouser, the system will suggest that trouser to a new buyer of the knit. It struggles with brand-new products that no one has bought yet, known as the cold-start problem.
A specific colour combination in which a style is produced and offered.
A single trainer design might be offered in a white, a black and a seasonal green colourway. Each colourway has its own style colour code and often its own imagery. Buyers choose which colourways to stock based on their customers and the rest of the assortment.
AI that interprets images and video, for example recognising garments, colours, patterns or defects.
Computer vision includes image recognition, which identifies what an image shows, and more detailed tasks such as locating items within it. Brands use it to tag product photos automatically with attributes like neckline or sleeve length. In warehouses it can check whether the right item has been packed.
A space within a retail store that a brand operates itself, managing stock and often staff, in exchange for a share of sales.
Concessions are common in department stores, where a brand runs its own branded area on the shop floor. The brand controls presentation, pricing and assortment, while the store takes a commission on turnover. The model gives brands more control than classic wholesale but also more cost and risk.
An arrangement where a brand places stock with a retailer but keeps ownership until the goods are sold to consumers.
Under consignment the retailer pays only for what it sells and returns the rest, so the brand carries the inventory risk. It can help new brands gain shelf space with cautious stockists. Brands need good visibility of retailer sales data to manage consigned stock.
Tamper-evident metadata, based on the C2PA standard, that records who created a piece of media and whether AI was used.
Content credentials work like a provenance label attached to an image or video, describing its origin and edits. A brand can attach them to campaign images so retailers and media can verify that the files are authentic and see if AI was involved. They are often used alongside invisible watermarking, which embeds a signal in the content itself.
The maximum amount of text, measured in tokens, that a language model can consider at once.
Anything outside the context window is invisible to the model during that request. A long contract or an entire season's assortment plan may not fit, so systems split documents or use RAG to select the relevant parts. Larger context windows help but do not guarantee the model uses every detail correctly.
Buying and selling through dialogue, such as chat, messaging apps or voice, often supported by AI assistants.
Conversational commerce lets a customer or buyer describe what they need in natural language. A retailer could ask a brand's assistant for 'all lightweight jackets deliverable in March under a set wholesale price' and receive a shortlist to order. It depends on accurate product data and integration with ordering and stock systems.
An AI assistant embedded in a software tool that helps a user with tasks while the user stays in control.
A copilot suggests, drafts or summarises, but the person decides. In a PLM system a copilot might draft a tech pack description; in a B2B portal it might suggest an order based on a store's past purchases. The term describes a design pattern and is used generically across many products.
Adding tags or annotations to raw data, such as marking garment types in images, so models can learn from it.
Supervised machine learning needs labelled examples. Someone, or a tool with human checks, has to mark which photos show a V-neck or which return reasons relate to fit. Consistent labelling rules matter: if teams use 'navy' and 'dark blue' interchangeably, the model learns confusion.
A central store that holds large volumes of raw data in its original formats until it is needed.
A data lake can collect point-of-sale exports, web logs, images and supplier files in one place without forcing them into a fixed structure first. This flexibility suits AI experiments but can become disorganised without clear ownership and documentation. Many companies combine a data lake with a data warehouse.
A structured database optimised for reporting and analysis, holding cleaned and integrated data from several systems.
A data warehouse brings together data from ERP, e-commerce and wholesale systems in consistent formats. Sell-through, margin and stock turn reports are typically built on it. It is usually the trusted base for both business intelligence and many forecasting models.
Machine learning that uses neural networks with many layers to learn complex patterns, especially in images, text and audio.
Deep learning powers most modern image and language AI. It is what allows a system to tell a trench coat from a parka in a product photo, or to draft a product description from a few attributes. It usually needs more data and computing power than simpler machine learning methods.
Realistic synthetic video, image or audio that depicts a person saying or doing something they did not.
Deepfakes raise risks of fraud and reputational damage, for example a fake video of a designer or executive endorsing a product. Using a real person's likeness for AI-generated content requires their consent and clear contracts. The EU AI Act sets transparency obligations for deepfake content.
The agreed period within which a brand must ship ordered goods to a retailer.
Each wholesale order is tied to a delivery window, for example the first two weeks of August for an autumn drop. Retailers plan floor space and marketing around these dates. Late deliveries can entitle the retailer to cancel the order or request a discount.
Predicting future sales of products using historical data, trends and other signals, increasingly with machine learning.
Forecasts guide how much to produce, where to allocate stock and when to replenish. Fashion forecasting is difficult because many products are new each season, so models often learn from similar past items. Inputs can include weather, pricing, promotions and early sell-through.
Short-term forecasting that uses very recent signals, such as daily sales or web traffic, to adjust near-term demand estimates.
Demand sensing complements longer-range demand forecasting. If a style starts selling faster than planned in its first week, demand sensing can flag it early enough to trigger replenishment or reallocate stock between stores. It depends on timely, clean data from point of sale and e-commerce systems.
A type of generative model that creates images by gradually removing noise until a coherent picture appears.
Diffusion models are behind most current text-to-image tools. A design team can describe a print idea in words and receive several visual drafts within seconds. Outputs still need checking for accuracy, originality and rights before commercial use.
A digital record attached to a product, usually via a data carrier such as a QR code, holding information on its origin, materials and sustainability.
The EU is introducing Digital Product Passports for priority product groups, with textiles among the first. A passport could show fibre composition, manufacturing locations, care and repair guidance and end-of-life options. Brands will need reliable product master data and supply chain data to create them.
Software that presents a collection to wholesale buyers online or on screen, with product data, imagery and ordering functions.
A digital showroom can support an in-person appointment on a large screen or let buyers browse and order remotely. It typically shows high-quality images, videos, prices, availability and outfit suggestions. Brands use it to reduce sample production and to reach buyers who cannot travel to a physical showroom.
A virtual representation of a physical product, process or place that is kept in sync with real-world data.
In fashion a digital twin may be a precise 3D version of a garment, linked to its materials and measurements, or a model of a warehouse or supply chain. A brand can use a garment's twin for sampling, e-commerce imagery and product passport data. The value comes from keeping it updated as the real item or process changes.
A company that buys a brand's products and resells them to retailers in a defined territory, taking on stock and credit risk.
Unlike an agent, a distributor owns the goods and invoices retailers itself. Brands often use distributors to enter distant markets where local logistics, language and regulation are challenging. The trade-off is a lower margin for the brand and less direct contact with retailers.
A release of new products into stores or online at a set time, often as one of several deliveries within a season.
Splitting a season into drops keeps the shop floor fresh and gives customers reasons to return. A brand might deliver its autumn collection in three drops between July and October. Some brands use limited drops to create scarcity and urgency around specific products.
Electronic Data Interchange: the structured exchange of business documents such as orders and invoices between company systems.
EDI replaces paper and email with standard message formats that systems can process automatically. A department store might send purchase orders to a brand by EDI and receive dispatch notes and invoices in return. Many large retailers require EDI capability from their suppliers.
Numerical representations of text, images or products that place similar items close together in a mathematical space.
Embeddings let software measure similarity by meaning rather than by exact words. A product image and the phrase 'cropped denim jacket' can be converted into embeddings and matched even if the product title says something different. They underpin visual search, RAG and many recommender systems.
Extended Producer Responsibility: policy making producers responsible for the collection and end-of-life treatment of the products they place on the market.
Under textile EPR schemes, brands and retailers pay fees based on the volume of products they sell in a country. The money funds collection, sorting, reuse and recycling. Fees may be adjusted to reward products that are more durable or easier to recycle.
Enterprise Resource Planning: core business software managing finance, orders, inventory, purchasing and logistics.
The ERP system is usually the single source of truth for stock, prices and order status in a fashion company. Wholesale orders captured in a showroom app or B2B portal are typically transferred to the ERP for allocation, shipping and invoicing. Fashion ERPs must handle style, colour and size matrices.
Ecodesign for Sustainable Products Regulation: the EU framework setting sustainability requirements for products sold in the EU.
ESPR allows the EU to set rules on durability, recyclability, recycled content and information requirements for product groups, including textiles. It also introduces the Digital Product Passport and restrictions on destroying unsold consumer products. Detailed rules for each product group are set in separate implementing acts.
The European Union's regulation on artificial intelligence, which sets obligations according to the risk level of an AI system.
The AI Act bans certain practices, imposes strict requirements on high-risk systems and sets transparency rules for things such as chatbots and synthetic media. It applies in stages and covers companies that place AI on the EU market or use it there, including non-EU brands. Fashion companies should map their AI uses, from recruitment tools to generated imagery, against its categories.
Methods that make it understandable why an AI system produced a particular output or decision.
Explainability builds trust and supports accountability. A planner is more likely to accept a forecast if the system shows that a recent price change and strong early sell-through drove the increase. Some regulations, including parts of the EU AI Act, require a degree of transparency about how AI is used.
Selecting and transforming raw data into inputs that help a machine learning model make better predictions.
Features are the variables a model uses. For a demand model, raw order dates might be turned into features such as weeks before season start, holiday periods or the share of a style already sold. Good features often reflect domain knowledge from buyers and merchandisers.
Further training a pre-trained model on a smaller, specific dataset so it adapts to a particular task, style or vocabulary.
Fine-tuning lets a company adapt a general model without training one from scratch. A brand could fine-tune a language model on its approved product copy so that generated descriptions follow its tone of voice and size terminology. It changes the model itself, unlike RAG, which supplies information at the moment of use.
A large AI model trained on broad data that can be adapted to many tasks, such as a general language or image model.
Foundation models are expensive to build, so most companies use existing ones and adapt them. A fashion company might use the same foundation model for copywriting, translation and customer service, each with different prompts or fine-tuning. Under the EU AI Act, many such models are treated as general-purpose AI.
Generative adversarial network: two neural networks compete, one generating content and one judging it, until outputs look realistic.
GANs were an important early method for generating realistic images, including faces and textile patterns. They have largely been overtaken by diffusion models for image creation but are still used in some specialised tasks. They are also associated with early deepfake techniques.
AI that creates new content such as text, images, video or designs based on patterns learned from training data.
Fashion companies use generative AI to produce product imagery on virtual models, explore design variations and create marketing copy. It can shorten creative cycles and reduce photo shoot costs. Questions of intellectual property, transparency and quality control remain important.
Using AI to propose many design options from a brief or constraints, which designers then select and refine.
Generative design shifts part of the designer's work from drawing to directing and editing. A team might ask for print variations within a colour palette or alternative silhouettes for a core style. Ownership, originality and the risk of resembling existing designs need careful attention.
Generative engine optimisation: making content easy for AI assistants and AI search tools to find, understand and cite.
As people ask AI assistants for advice instead of browsing search results, brands and publishers want to appear in those answers. GEO involves clear, factual, well-structured content, consistent product data and trustworthy sources. A brand's sizing guide, for example, is more likely to be used if it is precise and easy to parse.
Gross Margin Return on Inventory Investment: gross margin earned for every unit of money invested in average stock.
GMROI combines profitability and stock efficiency in a single figure. A category that earns a lower margin but sells very quickly can deliver a higher GMROI than a high-margin category that sits on the shelf. Buyers and merchandisers use it to allocate budget between brands and categories.
General-purpose AI: under the EU AI Act, models that can perform a wide range of tasks and be built into many systems.
Providers of GPAI models have obligations such as technical documentation and transparency about training content. A fashion company that uses a general-purpose model through an API is usually a deployer rather than a provider, with lighter duties. Fine-tuning or substantially modifying a model can change that role.
Tying an AI model's output to verified sources such as product master data or approved documents.
A grounded system answers from supplied facts rather than from what the model happens to have learned. For example, a customer service assistant is grounded when it quotes the return policy from the brand's current terms. Grounding usually works alongside RAG and source citations.
Global Trade Item Number, often called EAN in Europe: a unique barcode number identifying a specific product variant.
Each combination of style, colour and size receives its own GTIN, so a T-shirt in five sizes and three colours needs fifteen numbers. Retailers scan GTINs at goods receipt and at the till. Accurate GTINs are essential for EDI, marketplaces and inventory tracking.
Rules and technical controls that limit what an AI system can say or do, to keep outputs safe, accurate and on-brand.
Guardrails can block certain topics, enforce output formats, restrict which data an assistant can access or require approval for actions. A B2B assistant might be prevented from quoting prices outside the buyer's own price list. They reduce, but do not eliminate, the risk of errors and misuse.
When a generative AI model produces content that sounds plausible but is false or unsupported.
Language models generate likely text, not verified facts. An assistant asked about a style's fibre composition might invent a percentage if the data is missing. Grounding, guardrails and human review reduce the risk, especially for regulated information such as care labels or product passports.
An architecture that separates the customer-facing front end from the commerce back end, connecting them through APIs.
With a headless set-up, a brand can redesign its online shop, add an app or launch a new touchpoint without rebuilding pricing, stock and checkout logic. The same back end can serve consumer and wholesale front ends. The approach offers flexibility but requires more integration work.
Under the EU AI Act, an AI system in a sensitive area, such as hiring or credit, that must meet strict requirements.
High-risk systems must have risk management, quality data, documentation, human oversight and accuracy controls. Most fashion uses, such as product recommendations, are not high-risk, but an AI tool used to screen job applicants for stores or warehouses would likely be. Classification depends on the use case rather than on the technology itself.
A design in which people review, correct or approve AI outputs at defined points before they take effect.
Keeping a human in the loop balances automation with accountability. An AI may draft order confirmations or product descriptions, but a team member approves them before they reach retailers. It is especially important where errors are costly or where regulation requires human oversight.
The stage at which a trained model is used to produce an output, such as a prediction, answer or image.
Training happens once or occasionally; inference happens every time the model is used. Each time a B2B portal suggests complementary styles to a buyer, the recommendation model runs an inference. Inference cost and speed matter when AI features are used thousands of times a day.
A strategically important customer, typically a large retailer, that receives dedicated management and tailored terms.
Key accounts often account for a large share of a brand's wholesale revenue. They may receive exclusive styles, special delivery schedules, markdown support and data sharing. A key account manager coordinates everything from order planning to in-store presentation for that partner.
The full cost of a product once it arrives at the warehouse, including purchase price, freight, duties, insurance and handling.
Landed cost gives a true picture of what a product costs before it is sold. A jacket bought from a factory may look profitable until shipping, import duties and currency effects are added. Brands use landed cost to set wholesale prices and to compare sourcing options.
The time between placing an order and receiving the goods, whether from a factory to a brand or from a brand to a retailer.
Production lead times in fashion can range from a few weeks for local makers to several months for overseas sourcing. Long lead times force brands to commit to quantities early, increasing the risk of overstock or shortages. Shortening lead times is a common goal of supply chain projects.
A document listing every style in a collection with images, style numbers, colours, sizes and wholesale and retail prices.
Buyers use line sheets to review a collection and write orders, either during a showroom appointment or afterwards. Unlike a lookbook, a line sheet is a practical sales tool focused on product data rather than styling. Many brands now generate line sheets automatically from their product information system.
Large Language Model: an AI model trained on large amounts of text that can understand and generate natural language.
LLMs power chat assistants, product description writing, translation and search. A brand might use one to draft product copy in several languages from structured attributes in its PIM. Outputs should be checked for accuracy, as models can produce confident but incorrect statements.
A styled set of images presenting a collection's looks and mood, used for sales and marketing.
A lookbook shows how pieces work together in outfits and communicates the creative direction of a season. Sales teams send lookbooks to buyers before market week to generate interest. It complements the line sheet, which carries the commercial detail needed to place an order.
A branch of AI in which software learns patterns from data instead of following hand-written rules.
Machine learning models improve by being shown examples rather than being programmed step by step. A brand might train a model on several seasons of order history so it can suggest re-order quantities per store. The quality of the result depends heavily on the quality and relevance of the data used.
The difference between selling price and cost, expressed as a percentage of the selling price.
Margin shows how much of each sale is left to cover overheads and profit. If a retailer buys a shirt for 40 and sells it for 100, its gross margin is 60 percent. Margin and markup describe the same gap but use different bases, which often causes confusion.
A permanent reduction of a product's selling price, usually to clear slow-moving or end-of-season stock.
Markdowns protect cash flow by turning unsold stock into revenue, but they erode margin. A retailer might reduce a coat by a third after eight weeks of weak sell-through. Brands sometimes contribute to markdown costs for key accounts, known as markdown support.
An online platform where multiple brands or sellers offer products to customers, with the platform operator facilitating the sale.
Brands may sell on consumer marketplaces through a wholesale model, where the platform buys stock, or a partner model, where the brand keeps ownership and ships or allocates stock itself. Wholesale marketplaces also exist, connecting independent retailers with brands. Each platform has its own data and content requirements.
The amount added to cost to reach a selling price, expressed as a percentage of the cost or as a multiplier.
Fashion retailers often think in multipliers: a markup of 2.5 means a product bought for 40 is priced at 100. Expressed as a percentage of cost, that is a 150 percent markup, while the resulting margin is 60 percent. Brands set wholesale and retail prices with expected markups in mind.
The processes and tools that keep core business data, such as products, customers and suppliers, accurate and consistent.
Master data management defines who owns each data field, how it is validated and how changes flow between systems. For a fashion brand that might mean one process for creating a new style, assigning GTINs and publishing it to every channel. Clean master data is a precondition for reliable AI.
The planning and management of stock levels, allocation and pricing to maximise sales and margin.
In fashion retail, merchandisers translate financial targets into budgets by category and track performance through the season. They decide which stores receive which products, when to re-order and when to mark down. The term is also used for visual merchandising, which concerns how products are displayed.
Practices and tools for deploying, monitoring, updating and governing machine learning models in production.
MLOps applies software engineering discipline to AI. It covers version control for models and data, automated testing, monitoring for model drift and controlled retraining. For a company running forecasting and recommendation models across many markets, it keeps them reliable over time.
The gradual decline in a model's accuracy as real-world data or behaviour changes from what it was trained on.
Fashion is particularly exposed to drift because trends, channels and customer behaviour shift every season. A size recommendation model trained before a brand changed its fit block may start giving poor advice. Regular monitoring and retraining are the usual remedies.
AI that can process and combine more than one type of input or output, such as text, images, audio and video.
Multimodal models can look at a product photo and write a description, or take a text brief and produce an image. A buyer might photograph a competitor's window display and ask which styles in the current collection are similar. This combination is especially relevant to fashion, where so much information is visual.
A model made of layers of connected mathematical units that adjust their weights during training to recognise patterns.
Neural networks are loosely inspired by the brain but are statistical models, not thinking systems. Each layer transforms its input slightly, so early layers in an image model may detect edges while later layers detect collars or pockets. Large language models and diffusion models are both built on neural networks.
Natural language processing: AI techniques that let software understand, analyse and generate human language.
NLP covers tasks from sentiment analysis of reviews to translation and summarisation. A brand might use it to scan thousands of retailer comments and group recurring complaints about fit. Large language models are currently the dominant approach in NLP.
Never Out of Stock: core products a brand keeps permanently available so retailers can re-order them at any time.
NOS items are usually basics such as plain T-shirts, classic denim or underwear that sell steadily regardless of season. Retailers can order them in small quantities whenever needed, often automatically. For brands, NOS requires continuous stock holding but delivers dependable, repeatable revenue.
A computer vision task that finds and locates individual items within an image, typically with bounding boxes.
Object detection goes beyond saying what is in a picture to showing where each item is. In a street-style photo it can separate the bag, the shoes and the coat so each can be matched to similar products. It is also used to count items on shelves or on a production line.
The budget a retailer still has available to purchase stock for a given period, after existing orders and inventory are accounted for.
Open-to-buy is calculated from planned sales, planned closing stock, current stock and orders already placed. It stops buyers from overcommitting and leaves room for in-season opportunities. A buyer with remaining open-to-buy in October might use it for re-orders of strong sellers.
The total of confirmed wholesale orders a brand holds for a given season or period.
The order book shows how much retailers have committed to buy and is a key indicator of the health of a wholesale business. Brands monitor it during the selling period to decide production quantities. A strong order book for spring/summer gives management early visibility of revenue for that season.
The smallest order value or quantity a brand will accept from a retailer, per order, style or season.
Minimums protect brands from the cost of processing very small orders and help ensure a credible presentation in store. A brand might require an opening order of a set value or a minimum of six units per style. Minimums are often lower for re-orders than for first orders.
A system that tailors content, products, prices or messages to each user based on data about them.
A personalisation engine decides what each visitor sees. In B2B it might show a retailer only the price list, delivery windows and assortment relevant to their account. In consumer retail it may reorder product listings by predicted interest, which raises transparency and data protection questions.
Product Information Management: software that centralises and enriches product content such as descriptions, attributes and images.
A PIM holds the marketing and selling information needed for e-commerce, marketplaces, catalogues and B2B ordering. It ensures that a dress has the same name, fabric description and images on every channel. Fashion PIMs often manage translations and channel-specific requirements.
Product Lifecycle Management: software that manages a product's development from concept and design to sampling and production.
PLM systems store technical packs, bills of materials, supplier information and approval workflows. A designer's sketch becomes a structured product record that pattern makers, sourcing teams and factories all work from. Data from PLM often feeds into the ERP and PIM once a style is confirmed.
An order a retailer places with a brand ahead of the season, usually from samples, before the goods are produced.
Pre-orders form the backbone of the seasonal wholesale model. A buyer visits a showroom in January, selects styles from the autumn/winter collection and commits to quantities that will be delivered months later. The brand uses these orders to plan production volumes with its suppliers.
Using data and statistical or machine learning models to estimate what is likely to happen next.
Predictive analytics answers questions such as which accounts are likely to reduce their orders or which styles are likely to be returned. A sales team might use such scores to prioritise calls before a selling campaign. Predictions are probabilities, not certainties, and need to be read in context.
Analytics that recommends specific actions, often by weighing predicted outcomes against goals and constraints.
Where predictive analytics says what may happen, prescriptive analytics suggests what to do about it. It might recommend the markdown timing that best balances margin and sell-through, or how to split a limited production run across key accounts. Recommendations rely on the business rules and objectives that are built into the model.
The core, authoritative data describing each product, such as identifiers, attributes, sizes, prices and compositions.
Clean master data is the foundation for EDI, marketplaces, digital showrooms and compliance reporting. Errors such as a wrong fibre composition or missing GTIN spread to every connected system. Many brands appoint data owners to govern how master data is created and maintained.
Designing and refining the instructions given to a generative AI model to get reliable, useful outputs.
A prompt is the input a user or system sends to a generative model, and its wording strongly shapes the result. A merchandising team might develop a tested prompt that turns PIM attributes into a product description of fixed length with care instructions in a set order. Good prompt engineering includes examples, constraints and a clear output format.
Retrieval-augmented generation: an AI model looks up relevant documents first and uses them to ground its answer.
RAG connects a language model to a company's own knowledge without retraining it. A wholesale assistant could retrieve the current delivery terms and line sheet before answering a buyer's question about a style. Because answers are based on retrieved sources, they can be checked and kept up to date more easily.
A follow-up order for styles a retailer has already bought, placed during the season once early sales show what is working.
Re-orders let retailers chase demand rather than guess it. If a shirt sells well in its first weeks, the store buys more units in popular sizes. Brands can only fulfil re-orders if they hold stock or have flexible production capacity.
Software that suggests products or content to a user based on their behaviour, preferences or similar users.
Recommender systems drive 'you may also like' features in both retail and wholesale. In a digital showroom they can suggest styles that complete a buyer's assortment or that performed well in comparable stores. They typically combine collaborative filtering with product attributes and business rules.
The process of restocking products, typically NOS items, to maintain agreed stock levels at the retailer.
Replenishment can be manual or automated using sales data shared by the retailer. A department store might send weekly sales figures for a jeans range so the brand can ship exactly the sizes that sold. Well-run replenishment reduces both stockouts and excess inventory.
The sale of pre-owned products, either through third-party platforms or through a brand's own take-back and resale programme.
Brand-run resale lets companies capture value from second-hand trade and strengthen customer relationships. A customer might return a worn coat in store for credit, which the brand then cleans, checks and resells online. Resale is a practical step towards circularity.
Radio Frequency Identification: tags that let products be identified and counted wirelessly without line-of-sight scanning.
RFID labels sewn into or attached to garments allow a store to count its entire stock in minutes with a handheld reader. This improves stock accuracy, which is essential for click and collect and replenishment. RFID can also support anti-theft measures and traceability.
Recommended Retail Price: the consumer price a brand suggests retailers charge for a product.
RRPs help keep pricing consistent across channels and protect brand positioning. Retailers are generally free to set their own prices, as competition law in many markets prohibits brands from enforcing resale prices. Line sheets normally show both the wholesale price and the RRP.
The volume or value of goods a brand sells to its retail partners, as opposed to what those retailers sell to consumers.
Sell-in measures how successfully a brand has placed its collection with stockists. A high sell-in can look positive in the short term, but if the goods do not sell through to consumers, retailers may cut future orders or demand markdown support.
The share of stock sold to consumers within a period, usually expressed as a percentage of units received.
Sell-through shows how well a product performs at the point of sale. If a retailer receives 100 jackets and sells 60 at full price in the first eight weeks, it has achieved a 60 percent sell-through. Buyers use the figure to judge re-orders, markdowns and next season's buying.
A physical space where brands or agents present collections to wholesale buyers and take orders.
Showrooms may be permanent spaces run by a brand, multi-brand agencies or temporary set-ups during fashion weeks and trade fairs. Buyers touch the samples, check fit and fabric and discuss terms with the sales team. The appointment usually ends with a draft order that is confirmed later.
One agreed, authoritative place for each piece of data, so all systems and teams use the same values.
Without a single source of truth, a style's price or composition can differ between the PIM, the B2B portal and a retailer's system. For AI this is critical, because a model or assistant grounded in conflicting data will give conflicting answers. Establishing it is often more organisational than technical.
The planned distribution of units across sizes for a style, reflecting how demand typically varies by size.
Instead of buying equal quantities in every size, buyers use size curves based on past sales. A womenswear store might buy more mediums than extra-smalls for a knit top. A poorly matched size curve leaves stores with broken size runs and unsold outer sizes.
Stock Keeping Unit: an internal code a company uses to identify and track each distinct product variant it holds.
Unlike a GTIN, which is globally unique, a SKU is defined by each business for its own systems. A retailer might create SKUs combining a style number, colour code and size. SKU count is a common measure of assortment complexity.
A language model with fewer parameters than large models, designed to be cheaper, faster or able to run on local devices.
Small language models trade some general ability for efficiency and control. A brand could run one on store tablets to answer staff questions about stock and product care without sending data to an external service. They are often fine-tuned for a narrow task where they can perform well.
How many times a business sells and replaces its average inventory over a period, usually a year.
Stock turn is calculated by dividing sales at cost by average inventory at cost. A higher stock turn means less money is tied up in inventory and products are fresher. Basics and NOS items usually turn faster than seasonal fashion lines.
Artificially generated data that imitates the patterns of real data, used for training or testing models.
Synthetic data helps when real data is scarce, sensitive or expensive to collect. A team might generate rendered images of garments in many colourways and poses to train a recognition model. It must be checked carefully, because it can carry over or amplify flaws in the data or model that produced it.
Predicting future values from data recorded over time, such as weekly sales, returns or web visits.
Time-series models look for trends, seasonality and recurring patterns. For a NOS item they can project weekly unit sales and account for seasonal peaks. Fashion adds difficulty because many products are new each season and have little or no history of their own.
A small unit of text, often part of a word, that language models read and write; usage and limits are measured in tokens.
Language models do not process whole words but tokens, so a long product name may count as several. Pricing for many AI services and the size of the context window are expressed in tokens. Translating a full line sheet into several languages can therefore consume many tokens.
The ability to track a product and its materials through each stage of the supply chain, from raw fibre to finished garment.
Traceability helps brands verify claims about origin and working conditions and comply with due diligence rules. A brand might trace cotton from farm to spinner, mill and garment factory using supplier data and certificates. Full traceability beyond direct suppliers remains a major challenge for the industry.
The examples a model learns from, such as labelled product images, order histories or text.
A model can only learn what its training data shows it. If a visual search model was trained mostly on womenswear photos on white backgrounds, it may struggle with menswear shot on models outdoors. Rights to use images and text for training are an increasingly important legal question for brands and publishers.
A database designed to store embeddings and quickly find the items most similar to a query.
Traditional databases look up exact values; vector databases look up nearest neighbours. A brand could store embeddings of every style in its archive so designers can search for 'similar silhouettes to this sketch'. Vector databases are a common building block in RAG systems.
Technology that shows how a garment, accessory or make-up would look on a person, using images, video or 3D models.
Virtual try-on can work on a photo, a live camera feed or a body avatar. A shopper might see a jacket on an image of themselves, or a buyer might see a style on a range of body shapes. Accuracy of fit and drape varies, so it is often positioned as visual guidance rather than a size guarantee.
How many weeks current stock will last at the recent or forecast rate of sale.
Weeks of cover is calculated by dividing stock on hand by average weekly sales. If a store holds 120 units of a sweater and sells 20 per week, it has six weeks of cover. Merchandisers use the measure to spot items that need re-ordering or marking down.
The price a brand charges retailers for its products, before the retailer adds its own markup.
Wholesale prices are typically set so that the retailer can reach the RRP with its target margin. They may vary by market, currency or customer tier, and volume discounts can apply. Prices are published on line sheets and in B2B ordering systems.
One edition every weekday morning. Read in five minutes. Free for industry professionals.