How should a mid-size fashion group set up an AI centre of excellence?
A small, well-placed AI team can coordinate use cases, data, governance and training without becoming a bottleneck. This guide covers structure, roles, operating rules and a cost framework, with no invented benchmarks.
KEY TAKEAWAYS Summary by the editors
- A centre of excellence in a mid-size fashion group should be small and cross-functional, owning standards, governance, data access and training while business units own use cases and results.
- McKinsey's 2026 survey found that about 44% of respondents say AI is scaling across their enterprise, and 54% of those at organisations above $1 billion in revenue report enterprise-wide scaling against about one-third of smaller ones.
- Nearly three-quarters of AI high performers in the same survey had fundamentally redesigned workflows around AI, so a centre should be judged on workflow change, not on the number of pilots.
- No credible public benchmark exists for the cost of a fashion-sector AI centre, so budgets should be built from roles, tools and usage fees, with usage-based costs given a separate line.
- The EU AI Act's Article 4 on AI literacy, applicable since 2 February 2025, gives a centre a clear mandate to coordinate staff training.
An AI centre of excellence (CoE) for a mid-size fashion group is best built as a small, cross-functional team of three to six people who set standards, run governance, manage data access and training, and support business units, while the business units themselves own use cases and results. Anything larger risks becoming a queue; anything smaller risks being a single person with a title. The team size here is an editorial suggestion, not a published benchmark.
Why do mid-size companies need a centre at all?
Without coordination, each department buys its own tool, handles data differently and repeats the same legal and security checks. McKinsey's 2026 survey found that nearly nine in ten respondents use AI regularly in at least one function, and that 54% of respondents from organisations with more than $1 billion in revenue report enterprise-wide scaling, against about one-third of smaller organisations. Smaller companies therefore have the most to gain from shared capability, since they cannot afford a dedicated team in each function.
The same survey reports that 37% attribute some EBIT impact to AI and that AI high performers, who derive at least 5% of EBIT from AI, remain about 6% of respondents. A centre that focuses on a few well-chosen workflows is more likely to belong in the second group than one that counts pilots.
What should a centre of excellence be responsible for?
- Use-case intake and prioritisation, with a simple scoring of value, data readiness and risk.
- Governance: approved tools, data rules, risk classification and a register of AI systems in use.
- Data access and quality standards in partnership with IT and the business owners of data.
- Training and AI literacy, linked to the requirements of Article 4 of the EU AI Act.
- Supplier assessment, contract templates and exit terms.
- Benefit tracking and a regular report to the executive team.
What it should not do is own every project. In a healthy model, a business owner sponsors each use case, the centre supplies method and technical support, and results are reported by the business.
Which operating model fits a mid-size group?
| Model | How it works | Strength | Risk |
|---|---|---|---|
| Centralised | One team builds and runs most AI projects | Consistent standards, concentrated skills | Bottleneck, distance from business problems |
| Hub and spoke | A small core sets standards; named people in each function lead use cases | Balance of control and ownership | Spoke roles are part-time and can be squeezed by day jobs |
| Federated | Functions run their own AI with light shared guidance | Speed, close to the work | Duplication, inconsistent risk handling |
For most mid-size fashion groups, hub and spoke is the practical choice: a core of a few people plus champions in design, merchandising, wholesale or sales, e-commerce, supply chain and finance, each with a defined share of their time.
Who should be in the team?
Typical roles are a leader with business credibility, a data or analytics engineer, someone responsible for governance and legal liaison (often part-time from legal or compliance), and a change and training lead. Larger groups may add a product manager for AI tools. Where internal skills are scarce, external partners can fill gaps, but the centre should keep the decision-making and the knowledge of why choices were made.
McKinsey's State of Fashion 2026 emphasises upskilling, acquiring new talent, including from outside fashion, and strong change management as critical. Hiring someone with fashion operations experience who can learn AI is often easier than the reverse.
What does a centre cost?
There is no credible public benchmark for the cost of a fashion-sector AI centre, so this article does not offer a figure. Build the budget bottom-up from four lines: people (salaries and any external support), platforms and licences, usage-based fees, and training. McKinsey's survey found that about 20% of respondents say operating costs, including tokens, have constrained AI use, so usage fees deserve their own line and a monitoring routine. It also found that 28% of respondents spend more than 10% of their enterprise ICT budget on AI, which can serve as a context point for how seriously some organisations treat it, not as a target.
How do you start in the first 90 days?
- Appoint the leader and name the executive sponsor.
- Take stock of AI tools and experiments already in use across the company.
- Publish a short policy covering approved tools, data that must not be shared and who to ask.
- Choose three use cases with owners, baselines and review dates.
- Start role-based training, and record who attended, since Article 4 of the EU AI Act has applied since 2 February 2025.
- Report back to the executive team with what was learned, including what was stopped.
What are the common failure modes?
- A centre created for visibility, with no budget authority or access to executives.
- Pilots chosen for novelty rather than for a process owner who wants the result.
- Governance so heavy that teams bypass it and use unapproved tools.
- No link to finance, so benefits are claimed but never confirmed.
- Dependence on one or two individuals whose departure would end the programme.
Each of these can be spotted early in a quarterly review. A short list of questions, answered honestly by the sponsor, is usually enough: what did we stop, what did we scale, and what did it change in the numbers?
A centre of excellence is a means, not an end. The goal is that AI becomes part of normal planning, buying and selling work, at which point the centre can shrink to the governance and standards it needs.
Frequently asked questions
How big should an AI centre of excellence be in a mid-size fashion group?
An editorial rule of thumb is a core team of three to six people supported by part-time champions in each function. This is not a published benchmark and should be adjusted to the company's data maturity and number of use cases.
What is the difference between a centre of excellence and an IT department?
IT runs systems and infrastructure. A centre of excellence coordinates use cases, governance, training and supplier choices across business functions, working with IT on data and security.
How much does an AI centre of excellence cost?
No reliable public benchmark exists for fashion companies. Build the budget from people, platforms, usage-based fees and training, and track usage costs separately.
Does the EU AI Act require a centre of excellence?
No. It does not require any particular structure. Article 4 requires measures to support AI literacy of staff using AI systems, and a centre is one practical way to coordinate that work.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
