Data strategy for AI in fashion: what to fix first
Most fashion AI projects stall on data, not algorithms. Here is which data matters most, what to fix first and how to make product, inventory and sales data ready for AI.
KEY TAKEAWAYS Summary by the editors
- For most fashion companies the first data priority for AI is structured product master data: consistent identifiers, attributes, compositions and images for every style, colour and size.
- Gartner predicts that through 2026 organisations will abandon 60 per cent of AI projects that are not supported by AI-ready data, and found that 63 per cent of organisations lack or are unsure about the right data practices.
- Accurate inventory data is the second priority, because recommendations and assistants that suggest unavailable products damage trust; Ralph Lauren's Ask Ralph, for example, recommends items based on current stock.
- Sell-out data from retail partners is often the most valuable and least available data source for brands that sell through wholesale.
- The EU Digital Product Passport for textiles, with the delegated act planned for adoption in the fourth quarter of 2027, will reward brands that have already structured their product and supplier data.
The first thing to fix in a fashion data strategy for AI is product master data: consistent identifiers, attributes, material compositions and images for every style, colour and size. After that come accurate inventory data, clean sales history and, for brands selling through wholesale, sell-out data from retail partners. Fixing these four areas unlocks most of the AI use cases fashion companies care about.
Why is data the bottleneck for AI in fashion?
AI models learn from and act on data. If the data is incomplete, inconsistent or out of date, the model's output will be too, often stated with complete confidence. Gartner predicts that through 2026, organisations will abandon 60 per cent of AI projects that are not supported by AI-ready data. In a July 2024 survey of 1,203 data management leaders, Gartner found that 63 per cent of organisations either lacked or were unsure whether they had the right data management practices for AI.
Fashion adds its own complications. A single style can generate dozens of colour and size variants. Attributes such as fit, neckline or occasion are often stored as free text or not at all. Data sits in separate systems for design, planning, ERP, e-commerce, wholesale and stores, each with its own naming conventions. And short product life cycles mean the history for any single item is thin.
Which data does a fashion company need for AI?
| Data domain | What it includes | Use cases it enables | Common problems |
|---|---|---|---|
| Product master data | Identifiers, hierarchy, attributes, composition, care, size charts, images | Product content, search, recommendations, forecasting for new items, compliance | Free-text attributes, missing variants, inconsistent naming |
| Inventory data | Stock by location and channel, in transit, reserved | Recommendations, allocation, assistants, availability promises | Delays, phantom stock, channels counted separately |
| Sales and order history | Transactions by size, location, channel, price and promotion | Forecasting, size curves, markdowns, re-order suggestions | Lost sales not recorded, returns not linked |
| Sell-out data from partners | Retailer sales and stock by product and store | Wholesale re-orders, assortment advice, demand sensing | Not shared, irregular, in incompatible formats |
| Customer data | Profiles, behaviour, consent records | Personalisation, customer service, clienteling | Consent gaps, duplicates, privacy constraints |
| Supplier and origin data | Factories, materials, certificates | Compliance, product passports, risk monitoring | Held in documents rather than systems |
What should a fashion company fix first?
The order depends on the chosen use cases, but most fashion companies benefit from the following sequence.
- Product identifiers and hierarchy: every variant needs a unique, stable identifier and a consistent place in the category structure, across all systems.
- Attributes: define a controlled list of attributes per category (for example fit, length, material, pattern, occasion) and fill them consistently. AI can help here by extracting attributes from images and descriptions, with human review.
- Inventory accuracy: make stock data timely and comparable across channels, so that customer-facing AI does not recommend what cannot be delivered.
- Sales history quality: link returns to sales, flag stock-outs so that zero sales are not read as zero demand, and record promotions and price changes.
- Partner data: agree with key retail partners what sell-out and stock data is shared, how often and in what format.
The value of inventory data shows up clearly in customer-facing tools. Ralph Lauren launched its Ask Ralph AI stylist in its US app in September 2025, built on Microsoft's Azure OpenAI platform; according to FashionUnited, it recommends outfits based on current stock. That design choice only works if inventory data is reliable.
How do you make data AI-ready?
Gartner recommends five steps for making data ready for AI: align data sources to specific AI use cases, identify the governance requirements for AI, evolve metadata from passive to active, prepare data pipelines for model training and production, and assure and enhance data through testing and monitoring. The common thread is that readiness is defined per use case. Data that is good enough for a monthly report may be inadequate for a model making daily allocation decisions.
Language models can also help with the clean-up itself. McKinsey's analysis of generative AI in fashion includes generating personalised product descriptions; the reverse direction, extracting structured attributes from existing descriptions and images, is equally useful for filling gaps in product data. Either way, a person should check samples, because errors in composition or care data have legal and customer service consequences.
Who should own data quality in a fashion company?
Data quality fails when it belongs to everyone. Each core domain should have a business owner, for example product data with the product or merchandising team and inventory data with operations, supported by data stewards who maintain standards day to day. IT provides the systems and pipelines, but the definitions of a correct attribute or a valid size chart come from the business.
- Name an owner for each data domain and give them authority to set standards.
- Fix errors at the source system rather than in downstream reports.
- Measure completeness and accuracy of priority attributes and publish the figures monthly.
- Include data entry quality in the processes of design, sourcing and product teams.
How does regulation change data priorities for fashion?
EU regulation is turning product data into a compliance deliverable. The European Commission plans to adopt the delegated act for textiles under the Ecodesign for Sustainable Products Regulation in the fourth quarter of 2027. It will define the Digital Product Passport, which may include product identification, fibre composition, care and repair guidance, end-of-life information, origin details and the economic operators involved, accessible through a data carrier such as a QR code.
The data needed for the passport overlaps heavily with the data needed for AI. Brands that structure product and supplier data now will find both compliance and AI projects easier, while those that wait will have to do the same work under deadline pressure. A data strategy built around a few clear use cases and owners is the most practical way to prepare for both.
Frequently asked questions
What data do you need for AI in fashion?
Most fashion AI use cases need structured product master data, accurate inventory data, clean sales and order history, and customer data with proper consent. Brands that sell through wholesale also benefit from sell-out data shared by retail partners. Supplier and origin data matter for compliance use cases.
What is AI-ready data?
AI-ready data is data that is suitable for a specific AI use case: representative, consistently structured, well documented and monitored over time. Gartner stresses that readiness is defined per use case. Data that works for reporting may not be good enough for a model making daily decisions.
Why is product data so important for fashion AI?
Product data feeds almost every use case: content generation, search, recommendations, forecasting for new items and compliance. When attributes are missing or inconsistent, models cannot compare products or describe them correctly. Fixing product data often delivers benefits across several AI projects at once.
How does the Digital Product Passport relate to AI?
The Digital Product Passport for textiles will require structured product and supply chain data, such as fibre composition, care and origin information. The same data improves AI use cases like product content and search. The European Commission plans to adopt the textiles delegated act in the fourth quarter of 2027.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
