Fashion enterprises in 2026 are transitioning from experimental AI pilots to robust, governed deployment frameworks. To manage risks associated with model drift, intellectual property, and data residency, large groups are adopting specific organizational structures that define how talent and technology interact. These patterns ensure that generative tools remain compliant with global regulations while maintaining the creative agility required for seasonal cycles.
Key takeaways
- Centralized AI Centers of Excellence (CoE) provide the highest level of compliance but can introduce latency in brand-specific workflows.
- Internal Platform-as-a-Service (PaaS) models allow multi-brand groups to scale infrastructure while maintaining tenant isolation for private data.
- Governance must now extend beyond text and images to include 3D garment geometry and production-ready pattern data.
- Embedded ML squads are increasingly integrated into design and supply chain units to reduce the friction between technical development and creative output.
How do fashion groups organize their AI talent?
The allocation of data science and machine learning resources determines the velocity of AI adoption. Enterprises must choose between centralizing expertise to ensure rigorous oversight or distributing it to foster localized innovation. In 2026, the complexity of fashion-specific data—ranging from high-resolution fabric textures to complex .DXF pattern files—requires a governance layer that understands both technical constraints and aesthetic requirements.
1. The Centralized AI Center of Excellence (CoE)
This model consolidates all AI engineering, legal, and data science resources into a single corporate unit that services the entire organization. This structure is designed to enforce a unified technology stack and ensure that every model deployment adheres to a strict security protocol. By centralizing the "brain" of the company, fashion groups can negotiate enterprise-level agreements with providers like Azure OpenAI, which is now operated by Microsoft and provides cloud-hosted access to foundation models with enterprise-grade security.
Why it matters: It prevents the proliferation of "shadow AI" where individual teams use unvetted tools that might leak proprietary design data. For large groups like H&M, a centralized approach allows for standardized auditing of model outputs across different global markets.
What is still unclear: Whether a centralized unit can react fast enough to the hyper-local trend shifts that define fast-fashion segments.
2. The Internal AI Platform-as-a-Service (PaaS)
Large conglomerates are building internal platforms that provide pre-approved AI tools to their various brand subsidiaries. This allows the parent company to maintain control over the infrastructure while giving individual brands the freedom to build custom applications on top of it. Gap Inc., which operates Old Navy, Gap, Banana Republic, and Athleta, focuses on building a high-performing house of brands where infrastructure can be shared to shape culture and drive community impact efficiently.
Why it matters: This pattern is effective for diversified groups like Otto Group, which operates across retail, financial services, and logistics. By deploying AI-commerce capabilities through a unified platform, they can ensure that a tool developed for their retail arm, such as otto.de, can be safely adapted for their logistics or financial services subsidiaries.
What is still unclear: The long-term maintenance cost of building a proprietary internal platform versus relying on evolving third-party enterprise suites.
3. The Governance-First Hub
In this structure, the primary focus is on risk mitigation and legal compliance rather than direct development. The hub acts as a clearinghouse for all AI initiatives, requiring a rigorous review of data privacy, bias, and IP rights before any tool is greenlit. According to research from McKinsey State of Fashion, the 2026 edition of their report—released in November 2025—identified AI adoption and geographic diversification as defining forces that require such structured oversight.
Why it matters: As brands use technology to create better-selling designs and reduce marketing costs, as noted by McKinsey, the governance hub ensures these efficiencies don't come at the cost of legal exposure or brand dilution. It provides a safety net for hyperpersonalized customer communications.
What is still unclear: How to balance the "veto power" of a governance hub with the need for rapid creative experimentation during peak design seasons.
4. Embedded ML Squads in Design and Supply Chain
Instead of a separate department, data scientists and ML engineers are embedded directly into creative and production teams. This ensures that AI tools are built with a deep understanding of the fashion workflow, such as the nuances of fabric simulation and 3D garment fitting. On February 5, 2026 (source), industry reporting highlighted how leading brands are increasingly using 3D fashion technology and digital garments to streamline their development pipelines.
Why it matters: Embedded squads can solve specific technical hurdles, such as integrating AI-assisted design with existing CAD systems. This is particularly relevant for technical brands like Arc'teryx, which is owned by Amer Sports and focuses on high-performance outdoor apparel where precision in 3D simulation is critical for product integrity.
What is still unclear: The risk of creating "data silos" where different departments develop incompatible AI workflows that cannot be unified at the group level.
5. PLM-Integrated Governance
For many brands, the most logical place for AI governance is within their existing Product Lifecycle Management (PLM) system. This model relies on the PLM vendor to provide the necessary AI tools and the governance framework to manage them. Centric Software, owned by Dassault Systèmes, ships an AI-powered PLM platform that manages everything from product development to market intelligence, effectively baking governance into the standard design-to-shelf workflow.
Why it matters: It simplifies the tech stack by keeping AI tools within the environment where designers and developers already work. For brands scaling their physical and digital presence, such as Mark Gong who opened a first stand-alone store in Shanghai on August 17, 2026 (source), having integrated systems helps maintain consistency across global operations.
What is still unclear: The degree of vendor lock-in that occurs when a brand's entire AI strategy is tied to a single PLM provider's roadmap.
6. The Federated Data Governance Model
This decentralized approach allows individual brands or departments to manage their own AI initiatives while adhering to a shared set of group-wide data standards. It is often used by groups that have grown through acquisition and have disparate legacy systems. Governance is achieved through "data contracts" that define how information can be shared and used across the federation.
Why it matters: It respects the autonomy of individual brand cultures while providing enough structure to satisfy group-level reporting and compliance requirements. It is a pragmatic choice for groups that cannot easily consolidate their technical infrastructure into a single CoE or PaaS.
What is still unclear: The complexity of managing multiple, overlapping data contracts as the number of AI agents and models within the organization grows.
Comparison of AI Governance Models
| Governance Model | Best for | Limits |
|---|---|---|
| Centralized CoE | Global compliance & cost control | Can slow down brand-level innovation |
| Internal PaaS | Multi-brand groups with high tech maturity | Requires significant internal engineering |
| Governance-First Hub | Risk-averse brands & luxury groups | May stifle rapid creative experimentation |
| Embedded ML Squads | High-performance & technical apparel | Risk of fragmented data and tool silos |
| PLM-Integrated | Mid-to-large brands seeking simplicity | High dependency on the vendor's AI roadmap |
| Federated Model | Groups with diverse legacy systems | High administrative overhead for data contracts |
Which model fits your enterprise needs?
The choice of governance structure depends on your organization's technical maturity and risk appetite. Groups with a strong engineering culture and multiple distinct brands often find the Internal PaaS model most effective for scaling. Conversely, brands focused on high-performance technical gear benefit from Embedded ML Squads that can fine-tune models for specific garment geometries. For most enterprises, the 2026 priority is moving away from unmanaged AI usage toward a protocol that secures the brand's most valuable asset: its proprietary design data.
FAQ
What is a Centralized AI Center of Excellence in fashion?
It is a dedicated corporate unit that manages all AI talent, tool procurement, and compliance protocols for a fashion group. It ensures that every brand under the corporate umbrella uses the same vetted technology stack, minimizing the risk of data leaks and ensuring consistent model performance across the organization.
How does an Internal AI PaaS work for multi-brand groups?
The parent company builds a private cloud infrastructure that offers pre-configured AI models and development tools to its subsidiary brands. This allows brands like those under Gap Inc. to build custom, brand-specific applications while the parent company maintains control over data security, costs, and foundational infrastructure.
Why is governance important for AI-generated 3D fashion assets?
3D assets contain proprietary geometry and construction data. Without governance, these assets could be generated using biased models or stored in insecure environments, leading to IP theft or production errors. Proper governance ensures that digital garments are simulated accurately and remain the exclusive property of the brand.
How do PLM systems integrate AI governance?
Modern PLM platforms, such as those from Centric Software, incorporate AI tools directly into the product development workflow. Governance is managed through the PLM's existing permission structures and audit trails, ensuring that AI-generated tech packs and designs are reviewed and approved within the standard production pipeline.
What are the risks of a decentralized AI model?
The primary risk is fragmentation. Without a central protocol, different teams may use different models that don't communicate with each other, leading to inconsistent product data and making it difficult for the group to audit its overall AI usage for legal or ethical compliance.
Further reading
- Generative AI: Unlocking the future of fashion - McKinsey
- The State of Fashion 2026 - Business of Fashion