TL;DR
- This edition's thesis: it is not technology that blocks AI scaling in luxury, it is culture. Secrecy, silos, undocumented craftsmanship are simultaneously what creates a maison's value and what prevents AI from operating within it. I call this the Vault Dilemma.
- The number that says it all: 80% of AI pilots meet their technical objectives. Only 23% produce a financial impact. Two-thirds of organizations are stuck in what Dominic Weir calls pilot purgatory (Bain, Luxury Society).
- AI works when culture does not resist. Mandarin Oriental: -36% food waste, $207K in savings across four palaces. Saks Global: Agentforce, Amazon, Salesforce on the cap table, unified data. And yet, bankruptcy (Chapter 11, January 2026). AI is not a lifeboat.
- Agentic infrastructure is being built without luxury. J.P. Morgan x Mirakl, Spreedly, Nvidia NemoClaw, OpenAI/Promptfoo: the rails, the security, the tools are there. Not a single maison among the early adopters. Meanwhile, AI plays Anna Wintour: Bernstein scores runway shows with ChatGPT and Gemini (Dior 9.1, Chanel 9.0, Gucci 7.6) and converts creation into a stock market signal. The maisons do not know it yet.
- And the customer hasn't (yet) asked for anything. 8% of consumers would pay for an AI buying agent, 74% want a human in-store (Capgemini). Triple Lock: culture resists, infrastructure is built without luxury, demand does not yet exist.
WHAT'S MOVING
Mandarin Oriental: when AI scales in a palace's kitchens
Mandarin Oriental deploys Winnow (an AI food waste reduction system, used by kitchen teams and analyzed by F&B management) across four of its palaces: Hong Kong, London, Miami, Dubai. The system identifies and categorizes food waste in real time. Result after six months: 36% reduction, $207,000 in savings, 66 tons of food saved. The deployment is expanding to 40 hotels.
At the Forbes Travel Guide Summit (late February), two numbers echoed each other: 60% of luxury travelers prioritize technology in their hotel choice, but 63% demand a "human-first" approach. Mandarin Oriental solves the equation: technology operates in the kitchen, where the guest never sees it. Food waste does not touch a palace's identity. This is exactly where deployment happens without friction.
Sources: Winnow Solutions, 2026; Forbes Travel Guide Summit, Feb. 2026
Moncler x Google Veo 3: AI in the service of commerce, not the Instagram feed
Moncler, with R/GA (creative agency) and Google Veo 3 (AI video generation model), produces animated visual content for its e-commerce. 38 countries. 9 languages. Per-market personalization. No controversy.
Meanwhile, eMarketer (March 12) reveals that Gen Z rejects AI ads (39% negative sentiment, nearly double Millennials). 70% of Gen Z favor UGC (User Generated Content). Moncler avoids the trap by reserving AI for the product catalog, not brand communications. Same reflex as Shiseido in edition #3 (VOYAGER in R&D, zero controversy): the bottle law applies to images too. You don't say "AI-generated content," you say "market adaptation."
Sources: Google, March 2026; eMarketer, March 12, 2026
J.P. Morgan x Mirakl: the rails of agentic commerce are being laid. Without luxury.
J.P. Morgan Payments partners with Mirakl (marketplace and agentic commerce platform, Nexus solution) to secure transactions initiated by AI agents (March 10). In the same week, Spreedly (payment orchestration) enables agentic commerce as a native channel for merchants, with Priceline as the first enterprise partner (March 4). Two deals in ten days structuring the rails of agent-delegated commerce.
The signal: the infrastructure is ready. Agents can trigger purchases, select suppliers, compare offers. To my knowledge, not a single luxury retailer, not a single group, not a single maison among the early adopters. The rails are being laid without them.
Sources: PYMNTS, March 10, 2026; Spreedly/PRNewswire, March 4, 2026
AI, the new Anna Wintour? Bernstein scores runway shows with ChatGPT and Gemini.
Luca Solca, luxury analyst at Bernstein (Sanford C. Bernstein & Co., an investment bank specializing in equity research), now cross-references trade press, ChatGPT, Google Gemini, Instagram comments, and proprietary analysis to assign a score to collections. Results for Fall 2026 debuts: Anderson/Dior 9.1, Blazy/Chanel 9.0, McCollough & Hernandez/Loewe 8.4, Trotter/Bottega Veneta 8.4, Demna/Gucci 7.6. For comparison: De Sarno/Gucci in 2023 scored only 6.8.
Anna Wintour prescribes: she decides what is beautiful and influences the market. Solca does the opposite: he measures a collection's reception and converts it into a stock recommendation. AI does not prescribe, it quantifies. And unlike Anna Wintour, no one sees it entering the room. Solca himself says: "Clearly not a home run, but still an improvement." The vocabulary is that of the P&L, not the runway. AI does not create, does not manage the CRM, does not optimize the supply chain. It decides, by inference, how much an artistic director's vision is worth. A territory the maisons have not anticipated: who, in the boardroom, steers how AI evaluates creation?
The paradox is that the same technology could also strengthen an artistic director's vision. Raphael Doan (Si Rome n'avait pas chute) observes that a filmmaker will soon be able to control their film pixel by pixel, without logistical compromises, revealing their vision more authentically. Transposed to luxury: an artistic director using AI to refine a silhouette, test a drape, adjust a palette without going through the constraints of the workshop -- would they express their vision more fully? Would AI amplify the auteur? Perhaps, unless it is the Solca score that ends up dictating choices before the first sketch is even drawn...
Source: WWD, March 2, 2026
PARADOX OF THE WEEK
71% of luxury executives declare that AI adoption "can no longer wait" (DLG, 2026). And yet, 55% are still in exploration mode. This gap is not inertia -- it is luxury working as designed. The long game is an invariant of the maisons: you do not launch a bicentennial savoir-faire into a 90-day sprint. But the technology window does not negotiate. Emeric Rouchet (Ultra Intelligence, Odile Jacob) recalls the implacable mechanics of scaling laws:
Every tenfold increase in computing power mechanically raises the intelligence of models. A model tested today will be radically outpaced in six months. Luxury's long game collides with a clock that accelerates with every cycle.
Agentic rails are being laid right now (J.P. Morgan x Mirakl, Nvidia NemoClaw, OpenAI/Promptfoo), without waiting for the maisons to be ready. To accelerate is to betray the DNA. Not to accelerate is to let others set the rules.
DEEP DIVE
Luxury's AI pilots work. It is when they leave the lab that everything stops.
Late 2025, executive committee of a luxury ready-to-wear maison. The AI pilot has delivered on its promises. The recommendation agent cross-references purchase history, wishlists, and the events calendar in real time. The purchase propensity score gains three points over the quarter. KPIs (Key Performance Indicators) in the green. Nods around the table.
Then the CRM director asks the question that kills: "If we deploy, does the algorithm access our VIC files?"
Silence.
I have heard this silence at other clients (and even in digital agencies), in other rooms, on different floors. Every time, the same cause: to deploy is to give a machine access to knowledge the maison has been protecting for two centuries. This is not indecision. It is lucidity.
Dominic Weir (Luxury Society, January 2026) has a term for it: pilot purgatory. Two-thirds of organizations remain stuck between the successful pilot and deployment. Bain & Company (March 9) puts numbers on this purgatory: 80% of generative AI use cases meet or exceed technical expectations, but only 23% of companies manage to link these initiatives to measurable gains. The cause: legacy systems are structurally incompatible with the multi-step workflows of autonomous agents. Deloitte (March 13) confirms: markets already reward retailers that announce AI investments, before P&L results even materialize. But only "Leaders" (their most advanced category) produce a credible impact.
The Vault Dilemma
The more a maison cultivates scarcity, the more it locks down its data, processes, and access. And the more it locks down, the less AI can operate. The maisons that would benefit most from AI (UHNWI client data of unmatched richness, artisanal savoir-faire to document, complex supply chains to optimize) are also those that struggle most to deploy it without questioning what makes them singular. This is the Vault Dilemma.
The problem is cultural, not technical. Fred Cavazza diagnoses it precisely: AI adoption stalls despite technological progress. The bottleneck is semantic (the "job replacement" narrative terrorizes teams) and practical (interfaces designed by engineers, not by business users, and near-zero prompting fluency on the operational side). In the data world, they call it Databesity -- the accumulation of data without governance or purpose. The maisons do not lack data. They have too much, everywhere, in formats nobody reconciles. Nobody, almost: with 81 billion euros in revenue, LVMH has taken the lead by centralizing its data across its 75 maisons (I devoted the DEEP DIVE of edition #1 to this). But LVMH is the exception. A very large exception, granted, but an exception nonetheless. For Chanel (private, opaque by doctrine), for Hermes (sanctuarized craftsmanship), for a Kering in restructuring: the Databesity diagnosis remains intact.
In luxury, technical debt is called "heritage." Data silos are called "maison culture." And that is the heart of the problem: in a luxury maison, the silo is not an organizational accident. It is a choice. The artistic director does not share moodboards with the supply chain. By design. That is how you create a show that surprises, not one that optimizes. Georg Simmel analyzed this as early as 1904 in The Philosophy of Fashion: luxury operates through distinction, meaning deliberate separation. Opacity, control, verticality are not organizational bugs. They are mechanisms of active distinction, the very ones that create perceived value.
Emeric Rouchet speaks of an "AOHP (Appellation d'Origine Humaine Protegee -- Controlled Human Origin Label)" to designate the only territories AI will never be able to claim. Luxury is already an appellation of origin: artisanal savoir-faire, the human hand, made in France or made in Italy are certifications of human provenance. The question is whether this certification will suffice to protect value when AI can reproduce its appearance.
Velvet silos
Edgar Schein, the father of organizational culture theory (Organizational Culture and Leadership, Wiley), distinguishes three layers: artifacts (what you see), espoused values (what you say), and basic assumptions (what you never question). In luxury, the basic assumption is clear: scarcity creates value. Everything else follows. The creative director who refuses to share moodboards is not dysfunctional. They are protecting the surprise. The sales advisor or boutique director who keeps their VICs in a Moleskine notebook rather than in the CRM is not resisting change. They are cultivating intimacy. Bourdieu would call it habitus (Distinction, Minuit): dispositions internalized so deeply they are no longer up for debate.
Secrecy at Hermes, creative verticality at Chanel, collegiality at Kering, autonomy and competition between each maison at LVMH... These cultures are luxury's DNA. And they are, to varying degrees, structurally incompatible with standard large-scale AI agent deployment. As Mandarin Oriental (see WHAT'S MOVING) and Saks (see THE STORY) show, AI scales when culture permits it: in the kitchen for one, in retail for the other. Place Vendome is a different story. To function, an AI agent would need access to the purchase histories of clients who spend 500,000 euros a year. Clients whose name must never appear in a CRM accessible to a third-party vendor. Pieces whose price is not displayed, because "if you have to ask the price, you are not the client," as Loic Prigent might say.
The customer hasn't (yet) asked for anything
Meanwhile, Nvidia open-sources NemoClaw (its enterprise AI agent platform) at the GTC (GPU Technology Conference) this week. OpenAI acquires Promptfoo (an agent security startup, 350,000 developers, 25% of the Fortune 500). Anthropic launches its Claude Partner Network, $100 million and enterprise certifications. The tools are here. They have never been more accessible.
And the customer in all of this? Capgemini ("What Matters to Today's Consumer," 2026): As of now, only 8% of consumers would pay for an AI buying agent. 74% want a human in-store, up 20 points year-over-year. Culture resists. Infrastructure is being built without luxury. Demand does not exist -- or rather, not quite yet. Triple Lock.
My two cents: I am convinced luxury will not solve this equation from the top down -- through grand digital transformations proclaimed on slides. Or through five-year plans designed to please shareholders. The maisons that will scale are those that identify, process by process, what can be shared with an agent without eroding symbolic capital. The sorting will be long, unglamorous, and political. And it will happen in the silence of committee rooms, not at GTC keynotes.
That said, the triple lock observed in this edition (culture resists, infrastructure is being built without luxury, the customer hasn't -- yet -- asked for anything) is not a diagnosis of deadlock. It is a diagnosis of tempo. Luxury has always been late on technology, and it has always ended up embracing it -- even absorbing it -- but on ITS terms. Somewhere, in another executive committee, the same CRM director will ask the same question about VIC files. And this time, the answer will be ready. The question is who will have formulated it: the maison, or the platform that did not wait.
Sources: Dominic Weir, Luxury Society, Jan. 2026; Bain & Company, "Why Agentic AI Demands a New Architecture," March 9, 2026 (see READING LIST); Deloitte, "AI in Retail: ROI, Valuation, and Investment Strategy," March 13, 2026; Fred Cavazza, March 6, 2026; Eric Briones, Luxe et Digital, Dunod; Georg Simmel, The Philosophy of Fashion, 1904/Allia; Edgar Schein, Organizational Culture and Leadership, Wiley; Nvidia GTC 2026, CNBC; OpenAI/Promptfoo, TechCrunch; Anthropic Claude Partner Network; Capgemini, "What Matters to Today's Consumer," 2026
THE STORY
Saks Global: when the best AI stack in retail counts for nothing against a broken model.
A few years ago, I had the privilege, at Le Bon Marche (LVMH), of seeing from the inside what a luxury department store represents: a place of prescription, a multi-brand showcase, a space where the customer discovers what they were not looking for. Where all manner of socio-cultural and psychological phenomena are at work. But one must distinguish between two models. Le Bon Marche, like Harrods in London, and presumably a few iconic concept stores, is a unique flagship, culturally anchored in its neighborhood, carried by a loyal local clientele. Saks, on the other hand, was a chain: 39 stores, a network logic, a dependency on tourist traffic. The same distinction is observable in France, where Galeries Lafayette no longer works everywhere: only the historic flagships (Haussmann, Champs-Elysees) survive, carried by their own gravitational pull. Provincial branches are closing one after another. The "luxury department store chain" model is the one suffering most.
This model is going through deep mutations that Saks failed to absorb. Over twenty years, luxury maisons have reclaimed control of their distribution through three simultaneous movements. The massive deployment of DOS (Directly Operated Stores): when Louis Vuitton or Hermes opens on Madison Avenue, the customer no longer needs to go through Saks. The rise of brand flagships as cultural destinations: Dior's 30 Montaigne, Hermes's 24 Faubourg control every centimeter of emotion. And DTC (Direct-to-Consumer) e-commerce, where maisons master data, pricing, and storytelling end to end. The department store retains a discovery role for emerging brands. But for the major maisons, it has become a costly intermediary they are disengaging from season after season.
Why Saks acquired Neiman Marcus. In July 2024, Saks Fifth Avenue acquires Neiman Marcus for $2.7 billion. The thesis: respond to this erosion through consolidation. Create a unified American luxury department store giant (Saks Fifth Avenue, Neiman Marcus, Bergdorf Goodman, Saks Off 5th) to gain leverage against the maisons and reconquer a negotiating power that each banner, in isolation, was losing a little more each season. On the cap table of the new entity Saks Global: Amazon (logistics) and Salesforce Ventures (technology). Geoffroy van Raemdonck, then CEO of Neiman Marcus, spoke of a "proactive choice" in the face of retail mutations. $600 million in announced synergies over five years. On paper, the dream team.
What AI was supposed to solve. In September 2024, Saks deploys Agentforce (Salesforce): Data Cloud to unify customer data across the four banners, Commerce Cloud for e-commerce, and autonomous AI agents for customer service and clienteling. Sophie, the AI assistant, handled routine requests (order tracking, returns, in-store appointment booking) and provided sales advisors with personalized recommendations based on the client's complete purchase history. Marc Metrick, CEO: "It's not going to replace, it's going to augment." AI was targeting the right spot: the customer experience, the lifeline of the department store. So far, nothing to fault.
Why the model collapsed. Let me be clear: it is not AI that sank Saks. It is the debt. To finance the acquisition, the group of Richard Baker (a real estate man, not a retailer: he had already liquidated Lord & Taylor in 2019 and Hudson's Bay Canada in 2025 after 355 years of existence) borrowed $2.7 billion. The debt burden suffocated cash flow. To plug the gap, Saks unilaterally altered its vendor terms, capturing a larger share of sales. Result: vendors cut shipments. Shelves emptied. Sales dropped 13%. As an analyst cited by Fortune summarized: "It's very hard to sell merchandise you don't have." A $100 million payment default in December 2025. Chapter 11 on January 13, 2026: $4.7 billion in debt. Closure of 57 Saks Off 5th, 8 Saks Fifth Avenue, 1 Neiman Marcus stores. 1,200 jobs cut.
But the debt is only the detonator. When I was at Saint Laurent (Kering), the relationship with department stores was already under tension: the YSL corner in a department store meant visibility and revenue, but also a loss of control over pricing, merchandising, the customer experience, and final sell-out figures. A brand in a department store is a brand on concession: it depends on the traffic and choices of a third party. In DOS, it controls everything. Twenty years later, Hermes does not sell at Saks. Nor does Louis Vuitton.
What AI could have addressed, and what it could not solve. Saks's AI targeted real problems: unifying post-acquisition customer data, giving sales advisors real-time cross-brand recommendations in a multi-brand universe. This is precisely the terrain where the department store has an advantage over DOS: curation, discovery. AI could have reinforced this advantage. But it could not put a bag back on the shelf when the vendor stopped shipping, nor repay $4.7 billion in debt. AI was treating a symptom (the customer experience) while the disease (the business model) worsened. This is the most uncomfortable lesson for luxury CDOs: a brilliantly executed AI investment can be perfectly beside the point.
There was a time, not so long ago, when the B2B buyers from prestigious department stores like Saks or Neiman Marcus arrived at a maison's showroom for their post-show buying session, the red carpet was rolled out and the Champagne was served. After all, these buyers not only decided what would be worn in New York, Dallas, and Beverly Hills, but were also responsible for a significant share of the maisons' revenue (the famous wholesale). These buyers still exist, and the best among them remain irreplaceable prescribers. But their power has eroded as maisons reclaimed their own distribution. Tomorrow, a new interlocutor will knock on the showroom door: the customers' own AI agents, mandated to search, compare, and recommend. The red carpet will not disappear. It will change recipients. And the maisons will have saved the Champagne ;-)
The compass: 3 questions to ask yourself
1. Does your business model hold without AI? If yes, AI amplifies it: concentrate your investments on high-leverage use cases (clienteling, dynamic pricing, supply chain). If no, stop everything: AI conceals the crack, it does not seal it. First priority: solidify the underlying model. Saks invested in AI without asking this question. We know what happened next.
2. Who in your executive committee explicitly owns the viability of the model in the face of AI transformation? If the answer is "no one" -- and in most cases, it is "no one" -- the CTO talks stack, the CDO data, the CFO ROI, but none of them asks the structural question: "Is our revenue model compatible with what AI will change in distribution?" Next step: identify or create this responsibility. Estee Lauder has begun to fill this gap (see WHAT'S MOVING, edition #3).
3. If your main third-party distributor goes bankrupt tomorrow, can you sell direct within 90 days? If yes, you have a safety net. If no, this is the priority project for any maison at 20-30% wholesale revenue. The minimum: a DTC (Direct-to-Consumer) capability that can be activated quickly, AI sales agents included. A third-party partner's bankruptcy is no longer a theoretical scenario -- it is a precedent. Not to mention the COVID episode, which had the virtue of putting data sovereignty and customer knowledge on markets that were sometimes heavily intermediated, and by extension direct sales, at the heart of the maisons' concerns.
4. A question no one is asking yet: if your AI agents replace your juniors' tasks (press dossier compilation, initial product recommendations, sell-through analyses), who will train your future creative directors, your future CDOs, CMOs, your future product merchandising directors? The training paradox is this: to steer AI effectively, you need general culture, judgment, field experience. Exactly what you acquire by doing the tasks AI is replacing...
Sources: CNBC -- Saks Global files for bankruptcy, Jan. 2026; CNBC -- Saks acquisition of Neiman Marcus led to bankruptcy, Jan. 2026; Fortune, Jan. 2026; Retail Dive, Feb. 2026; Salesforce, Sept. 2024
MY INDISCREET QUESTION
The last time your executive committee discussed AI, how much time elapsed between "this is convincing" and "but not now, let's think about it a little more..."?
ON MY READING LIST
Bain & Company, "Why Agentic AI Demands a New Architecture" (March 9, 2026)
The reading that fueled this edition's DEEP DIVE. Pilots work, scaling does not. Bain names the problem: it is the company's architecture that must evolve, not the AI.
The Fashion Law, "Agentic AI Is Reshaping Commerce. Is the Law Ready?" (March 13, 2026)
15 to 25% of US e-commerce sales via AI agents by 2030 ($300-500 billion, Bain projection). No jurisdiction has legislated on agentic commerce. If an agent recommends a fake Hermes, who pays?
Jing Daily, "Why Balenciaga Sticks to a Human-Centered Approach in an AI-Driven World" (March 10, 2026)
Pierpaolo Piccioli asserts that clothing must remain a vehicle for humanity. In an edition devoted to culture as a barrier to AI scaling, this stance from a creative director illustrates that resistance is not only structural: it is also doctrinal. Luxury that says no to AI is not always dysfunctional. Sometimes, it is protecting something.
Fred Cavazza, "Agentization marks a new stage of digital maturity" (March 6, 2026)
The best French-language synthesis of the AI adoption paradox: everyone talks about it, nobody deploys. Cavazza identifies the bottleneck that English-language reports sidestep: prompting and interfaces, designed by engineers, not by business users.
Deloitte, "The Future of Luxury During the Rise of AI" (January 8, 2026)
A luxury-dedicated report (not generic retail). Four imperatives: aligning AI with brand values, operational optimization without compromising craftsmanship, data governance, and upskilling teams. The full report is available as a PDF download.
COMING NEXT
The Nvidia GTC is in full swing in San Jose this week: the L'Oreal x Nvidia session ("Beauty Reinvented: AI-Enabled Formulation Optimization") could be the first case of AI scaling in luxury beauty beyond the Shiseido pilot. Tomorrow, the Journal du Luxe Intelligence Live convenes sector decision-makers around Eric Briones: I will be watching what is said, and also what is not. LVMH and Kering annual results are upcoming: I will finally learn whether the "AI for All" doctrine has produced numbers, and whether the De Meo plan has begun to execute. In any other industry, resisting technology would be an admission of weakness. In luxury, it is sometimes the sign that you still know what you are protecting. Luxe oblige.
Mickael.