Organising fashion companies for AI

When a new technology enters a company, the first reaction is often to assign it to a function: IT for systems, marketing for content, HR for training. This division helps organise expertise, but it is not enough to govern a transformation that spans processes and customer relationships.

The LVMH case is interesting for a concrete reason. In presenting its collaboration with Google Cloud, the group described a data and AI platform designed to address the distinct needs of the maisons and preserve personal customer relationships. Google presents this as an approach of “quiet tech”: technology integrated into work without becoming the centre of the experience. The case documents the group’s stated approach and announced use cases; it does not by itself prove economic impact.

The organisational lesson is not “centralise AI”. It is to build a shared foundation that reduces duplication and risk while leaving each maison responsible for the choices that shape its product, service and relationships. A common platform can provide security, data access, tools and expertise; it should not decide what the brand means.

In a multi-brand group, the tension between scale and autonomy is real. If every maison negotiates infrastructure, security and tools separately, costs, risks and duplicated expertise grow. If every process is imposed centrally, the platform can become a constraint, drift away from daily work and encourage uniform solutions.

Table 1. One possible division between group platform and maison autonomy.

This table is a proposed model, not a verified account of every internal LVMH practice. Its purpose is to make decisions explicit: the centre enables, protects and makes capabilities interoperable; the brand defines the problem, the desired experience and the judgement that cannot be delegated.

This is a federated model in practical terms. It does not necessarily require a Chief AI Officer for every brand or a complex organisation chart. It does require a person with authority over the portfolio and shared conditions, alongside a business owner accountable for the outcome of each initiative. A project without an operational owner risks remaining a technical demonstration.

Wealth management faces a partly comparable problem: how to prepare a professional for a complex personal relationship involving substantial information, without turning that professional into an executor of automated suggestions. Morgan Stanley has introduced AI tools for its advisors, including an assistant that helps retrieve internal knowledge and a feature that summarises meetings and follow-up tasks.

The transferable principle for luxury is limited but useful: AI can prepare a meeting, retrieve authorised information and reduce administrative work, while the person remains responsible for listening, interpretation and the relationship. In a boutique, this could mean better preparation using a client’s authorised preferences and service history, always respecting permissions, confidentiality and the advisor’s judgement.

The difference matters. A financial advisor works in a regulated field and may make recommendations with consequences for a client’s assets; a maison advisor creates an experience tied to taste, culture, objects and rituals. Financial recommendation or profiling logic should not simply be transferred to luxury. What can be adapted, cautiously, is the design of support: relevant information before the meeting, useful notes afterwards, and continuity without replacing the relationship.

AI should also know when to stay in the background. If an assistant distracts the team, exposes unnecessary data or prompts stereotyped recommendations, the process needs redesign. The measure is not the number of suggestions produced, but service quality and time returned to human work.

A global group can invest in platforms, governance, expertise, data access and experimentation across multiple maisons. An independent house has more limited resources, data and capacity to absorb change; replicating a group’s infrastructure would often be disproportionate. This is a structural difference, not a gap to close by buying more technology.

An independent house can apply the same principle on a smaller scale: a lean steering group bringing together leadership, IT or a technical partner, marketing/creative, retail and legal; a limited set of approved tools; clear data boundaries; and a few use cases with an owner and a success criterion. It can outsource infrastructure, but not the definition of the problem, judgement over content or accountability to the customer.

For example, a house might begin with archival and materials research, appointment preparation or consistency across multilingual product descriptions. The choice depends on real bottlenecks. Before starting, ask: what activity will change? Who checks quality? Which data are essential? Which error would be unacceptable? How can the system be stopped when it is unreliable?

The CEO should sponsor the transformation when it spans functions and changes priorities, investment or accountability. They do not need to approve every prompt or model. They should set the ambition, select important processes and ask for evidence of value and risk.

A central steering function — whether a dedicated role or a small decision-making committee — governs shared standards: security, access, architecture, vendor selection and monitoring, foundational training, and criteria for starting or stopping projects. The centre does not own use cases: it removes obstacles and prevents every team from rebuilding the same foundations.

The process owner — for example in retail, merchandising, the archive or customer service — defines the problem and shares accountability for outcomes with technical teams. Legal, privacy and HR should be involved early when a use case touches personal data, employment decisions or customer-facing content. Creative and brand leadership need an explicit role whenever outputs might change voice, aesthetics or positioning.

A credible transformation starts with a small scope and an explicit decision about what not to automate. The following cycle is an editorial proposal, not a roadmap validated at a specific maison.

Table 2. Suggested initial cycle; thresholds should be defined for each use case.

A common risk is measuring activity — accounts created, prompts run, content produced — instead of outcomes. For clienteling, examine preparation and perceived service quality; for an archive, time to find sources and answer accuracy; for merchandising, the quality and traceability of assumptions. No single indicator proves causality: document the context and compare before and after cautiously.

The question is not whether to centralise or decentralise all AI. It is where sharing increases safety and capability, and where autonomy protects the quality of the maison. LVMH offers a reference for scale and intent: shared infrastructure, reported broad adoption and a plurality of brands. It is not a model that can be copied in every detail, nor automatic proof of economic return.

For an independent house, the practical path is lighter: share what does not differentiate the brand — rules, protection and foundational skills — while retaining control of the problem, data, experience and judgement. For a group, the challenge is to prevent a shared platform from becoming an imposed shared language. In both cases, value emerges when technology fits real work and gives more room to expertise, relationships and human decisions.