The State of Fashion AI
Where artificial intelligence works in fashion, where it doesn't, and what the next two years look like.
Thirty-five percent of fashion executives report using generative AI for business functions. Seventy-four percent say they have adopted AI for at least one purpose. At face value, the industry appears mid-transformation. Look closer. Production deployment of generative AI for core operations sits at 15-20%, and 90% of AI initiatives fail to scale beyond pilot. The AI-in-fashion market will grow from roughly $3 billion in 2025 to $8-12 billion by 2028. The question is which applications work today, which are experimental, and where the industry is being sold a future further away than the pitch decks suggest.
Key Findings
The gap between experimentation and production is vast — 48% of fashion brands have integrated some form of AI, but only 15-20% use generative AI in production. 90% of AI initiatives fail to scale beyond pilot.
Commerce AI delivers the strongest returns — Product recommendations drive 31% of e-commerce revenue. Size recommendation reduces returns by 24-40%. Dynamic pricing cuts markdowns by 18%. These are production results, not projections.
Agentic AI is the next platform shift — The agentic AI market hit $7.84 billion in 2025, growing 49% year-on-year. Average ROI is 171%. But only 5% of enterprises see returns at scale.
The 90% failure rate is a data story — Across every category, the single most cited barrier is data quality and fragmentation. Only 21% of enterprises fully meet data readiness criteria.
The mid-market is stuck — Brands with 10-250 people face enterprise pricing, fragmented tech stacks, and 18-month integration timelines they cannot afford.
The Honesty Table
A candid assessment of where each AI application in fashion actually stands. Production-ready means documented, repeated results across multiple deployments. Early production means working in a handful of cases but not yet broadly proven. Experimental means the technology functions but lacks sufficient production evidence.
31% of e-commerce revenue; conversion lifts of 50-150%; Constructor nearly doubled revenue in FY24
20-50% forecast error reduction; Nextail 30-day ROI; 65% fewer stockouts
Zara 18% markdown reduction; 7Learnings 10-13% revenue lift; Leder & Schuh EUR 3M savings
24-40% return reduction; 34% conversion lift; True Fit powers 91,000+ brands
90%+ accuracy; 20-30x faster than human inspectors; Smartex 1M kg waste prevented
Heuritech 91%+ accuracy; 40% faster time-to-market for adopters
60% faster prototyping; $80K annual savings; Raspberry AI $24M Series A
85% factory readiness (TheFWord.ai); sampling rounds cut from 4 to 1
Style3D and CLO embedding AI features; 45% surge in AI pattern-making software in 2025
Style3D AI, CLO AI features; rendering time dropped from 5 min to 90 sec
Disruption frequency up 3x since 2019; AI-powered risk prediction live at enterprise scale
LCAi offering fast assessments; Made2Flow 5,000+ suppliers; Carbonfact 50M+ LCAs
171% average ROI; but 5% success at scale; $7.84B market at 49% YoY growth
AI systems that generate and execute buy plans without human oversight
Agent systems that evaluate suppliers, negotiate, and place orders
AI for Design
The design category is where AI hype is loudest and the reality gap is widest. Every major design software company is embedding generative AI. The question is what "embedded" means in practice.
What works today
General-purpose image generators (Midjourney, DALL-E) are standard for mood boards. Trend prediction platforms like Heuritech deliver 91%+ forecast accuracy by analysing millions of social media images across 2,000+ attributes. These are production tools.
What is promising but unproven at scale
Raspberry AI raised $24 million (a16z, January 2025), the largest dedicated fashion generative AI round. Clients include Under Armour, MCM Worldwide, and Li & Fung. TheFWord.ai's tech pack generation achieves 85% factory readiness, but still requires human review.
What is further away than it looks
AI patternmaking and sketch-to-3D are not production-ready for most brands. CLO ($36M Series D) and Style3D ($145M total) are embedding AI features, with rendering times dropping from five minutes to 90 seconds. But accurate 3D simulation requires digitised pattern blocks and fabric libraries most brands lack. Sketch-to-3D will mature by 2027, not before.
AI for Commerce
This is where AI in fashion has its strongest production evidence, hardest numbers, and clearest ROI.
Demand forecasting
AI forecasting reduces inventory by 20-30%, translating to working capital improvements of $15-20 million per billion in revenue. Nextail clients see 5-10% sales increases, 30% less stock coverage, and 60% fewer stockouts within 30 days.
Product discovery and personalisation
AI recommendations drive up to 31% of e-commerce revenue. Constructor ($550M valuation, revenue nearly doubled FY24) and Algolia ($2.25B valuation, $100M revenue 2024) lead the enterprise tier. Shoppers who receive relevant AI suggestions convert at rates up to 369% higher.
Dynamic pricing
Zara's AI-driven stock allocation reduced markdowns by 18% and cut stockouts by 25%, boosting operating margin by 1.8%. 7Learnings clients report 10-13% revenue increases. The company reached profitability ahead of its EUR 10M+ Series B (May 2025).
Size recommendation
True Fit powers 91,000+ brands ($152M raised) and reduces fit-related returns by 24-50%. A European retailer documented 32% fewer size-related returns, saving EUR 2.3 million annually. Products with virtual try-on see 94% higher conversion.
AI for Operations
On the factory floor and in the supply chain, AI is solving different problems with different economics.
Quality control
Smartex (Portugal, $27.6M total funding, H&M strategic investor) has prevented 1 million kilograms of fabric waste in three years. Its "Golden Stop" technology halts production within 10cm of a defect's creation. AI QC systems detect 40+ fabric defect types with 90%+ accuracy, operating 20-30x faster than human inspectors who achieve only 60-70% accuracy.
Training data requirements — AI QC requires 10,000+ labelled defect images per fabric type. Fabric variability means each new material needs retraining.
Adoption concentration — Large mills and tier-1 factories benefit most. Mid-market factories (50-250 people) remain dramatically underserved.
Supply chain communication
An estimated 85-95% of fashion brand-supplier communication happens on WhatsApp or WeChat. Only 15-25% use structured collaboration platforms with audit trails. AI is beginning to help with automated translation, spec extraction from messages, and intelligent routing.
The translation problem
Every handoff in the design-to-production chain loses data fidelity. AI can compress these handoffs: tech pack generation reduces sampling rounds from 3-5 to 1; automated BOM costing eliminates currency conversion errors. But these gains require upstream data to be clean and structured.
AI for Compliance
The EU's regulatory framework is creating a distinct category of AI application in fashion.
LCA automation — Life cycle assessment is being compressed by AI platforms. Carbonfact has built a database of 50 million+ LCAs. Made2Flow integrates with PLM platforms and has gathered data from 5,000+ suppliers globally.
Digital Product Passports — The ESPR was approved June 2024. Delegated act for textiles expected late 2026, enforcement from mid-2027 to 2028. Only 5-10% of fashion brands have DPP systems in production today.
Supply chain tracing — AI-powered supplier discovery uses satellite imagery, trade data, and NLP to map supply chains. But 80% of independent brands lack the supplier data needed for 2026 EU regulatory requirements.
Agentic AI: The Next Wave
Agentic AI is the category the industry is least prepared for and the one most likely to reshape how fashion companies operate. Unlike generative AI that augments individual tasks, agentic systems orchestrate entire workflows. An AI agent does not help you write a tech pack; it generates it, sends it to the factory, follows up on the sample, and flags deviations from spec.
Why 95% are not ready
Agentic AI has the highest data requirements of any AI category, because autonomous agents make decisions that trigger real-world actions. Poor data produces wrong actions at scale. The prerequisites are connected PLM, ERP, and supplier systems with clean, structured data. Only 21% of enterprises fully meet data readiness criteria. For mid-market brands with 3-7 disconnected tools and no dedicated IT team, the gap is structural.
The infrastructure prerequisite
Agentic AI does not work without a connected data layer. It needs structured BOMs, digital tech packs, real-time inventory feeds, and supplier performance data in machine-readable formats. PLM adoption is the unglamorous prerequisite that determines whether a brand can participate in the agentic wave or not.
The Data Quality Problem
Ninety percent of AI initiatives fail to scale. This is not a technology failure. It is a data failure.
Demand forecasting — needs 12-24 months of clean, SKU-level sales history.
Quality control — needs 10,000+ labelled images per defect type per fabric.
Personalisation — needs structured product attributes plus 10,000+ monthly sessions.
Dynamic pricing — needs transaction data by SKU, competitor feeds, and real-time inventory levels.
Most brands have this data scattered across 4-8 systems in inconsistent formats. Product taxonomies vary between PLM and e-commerce. Supplier data sits in email threads and WhatsApp messages.
The organisations achieving the highest ROI invested 31% of their implementation budgets in data preparation rather than algorithm development. The algorithm is the commodity. The data is the differentiator. And the system that structures the data, the PLM, is the prerequisite.
Vendor Landscape
The fashion AI vendor market is consolidating rapidly. Twelve M&A transactions in fashion design and production software occurred in 2024-2025. The Big 4 (Lectra, Centric, Browzwear, CLO) control approximately 55% of fashion design and production software revenue.
Acquired aifora (AI pricing, Sep 2023), Contentserv (PIM/DAM, Mar 2025), Fashionboard (AI PLM, Mar 2025). Dassault-backed.
Acquired Lalaland.ai for AI-generated model imagery. Leading 3D simulation platform.
$36M Series D (Dec 2024). Acquired Swatchbook (digital fabric database).
$145M total funding. Chinese challenger undercutting Western incumbents by 30-40% on price.
$24M Series A (a16z, Jan 2025). Clients include Under Armour, MCM, Li & Fung.
EUR 10M+ Series B (May 2025). Profitable. Expanding to North America.
$11.6M total. 30-day ROI on demand planning.
$550M valuation. $85M raised. Nearly doubled revenue FY24.
$27.6M total. H&M strategic investor. 1M kg waste prevented.
Launched Fashion & Apparel with AI agents (Feb 2026) on Dynamics 365.
Platform consolidation
Salesforce spent $10B+ on AI M&A in six months. ServiceNow spent $11.6B. In fashion, PLM vendors are adding pricing, compliance, and AI features through acquisition. Standalone AI tools face absorption or marginalisation pressure. Style3D, headquartered in Hangzhou with $145M in funding, is undercutting Western incumbents by 30-40% on price, offering mid-market brands a credible alternative.
The Mid-Market AI Gap
Enterprise brands have dedicated AI teams, clean data, and budgets for 6-18 month implementations. Micro brands use Midjourney and Shopify widgets. Both segments are served, if imperfectly.
Brands with 10-250 people managing 25-200 styles per season are caught in between. They process enough styles to benefit from AI, but not enough to justify $200K+ enterprise deployments. Their tech stacks average 3-7 disconnected tools. Sixty percent of mid-sized companies report difficulty connecting AI solutions with existing PLM and ERP systems.
The gap is not awareness. It is access.
What to Invest in Now vs. Wait
Invest now
AI product recommendations and search — if you sell online. Proven ROI, short time-to-value, affordable entry points.
AI size recommendation — if returns exceed 15%. Reduces fit-related returns by 24-50%.
AI demand forecasting — if you manage 50+ SKUs. 20-50% forecast error reduction. Start with your e-commerce stack.
PLM adoption — the prerequisite for every downstream AI application. If your product data lives in spreadsheets, no AI tool will deliver on its promise.
Pilot cautiously
AI dynamic pricing — if you manage markdowns across 10,000+ SKUs.
AI quality control — if you own or manage factory production.
AI-powered supply chain mapping — if you sell into the EU and face DPP compliance.
Watch closely, do not overcommit
AI-assisted design beyond mood boards — maturing but not yet proven at mid-market scale.
Agentic AI workflow automation — 171% ROI at enterprise; mid-market viability expected 2027-2028.
AI patternmaking and sketch-to-3D — the technology will be materially better in 18 months.
Prepare for
Digital Product Passport compliance — mandatory for textiles by 2027-2028.
ESPR ban on destroying unsold apparel — takes effect July 2026.
Agentic AI reaching mid-market viability — expected 2027-2028. Start data preparation now.
AI needs structured data. Kobo provides it.
Kobo gives growing fashion brands the structured product data layer that AI depends on, with implementation in weeks and pricing that fits mid-market budgets.
Book a Discovery Call