Hermès' digital strategy — the AI investments behind a no-AI posture.

The Dumas position: what it actually says

Dumas's public statements about AI have been more precise than the trade press's summaries have suggested. In an interview with Les Échos in March 2024, asked about the integration of AI into the company's design and craftsmanship work, he answered with a sentence the company's communications function has since used as a reference formulation: "Hermès does not employ artificial intelligence in the work of its artisans, and it will not." The sentence is exact. It addresses the work of the artisans — the cutters, the leather-stitchers, the saddlers, the silk-printers, the craftsmen across the Maison's production lines who are the visible signature of the company's product. It does not address the work of the company's commercial functions, its supply-chain operations, its client-service infrastructure, or its anti-counterfeit programme. The distinction is not accidental. It is the framing that the Maison has been deliberate about preserving across every public statement since.

Dumas restated the position in the company's first-half 2024 results presentation in September 2024, when an analyst from a Paris-based equity-research firm asked specifically about the company's investment in AI infrastructure and whether the public no-AI posture was sustainable as the technology's commercial penetration deepened across the sector. Dumas's response separated two registers: he was clear that Hermès would not pursue AI integration into its creative work because the company's commercial proposition is built on the visible authorship of human craft, and he was equally clear that the company invested in technology across its operations as any responsible enterprise must. The phrase he used — that the company invests in operational excellence wherever operational excellence is the right register — has become the internal vocabulary the company's executive committee uses to describe what is, in practice, a substantial AI investment programme. The phrase performs important work: it preserves the no-AI posture for the artisan-facing dimension of the brand while creating space for the operational-facing investment the company's commercial sustainability requires.

The communications discipline around the position has been notable. The company's investor relations function has briefed senior analysts privately on the operational AI investment programme, on the reasoning that selectively disclosed information to professional investors does not contradict the company's public posture in the way that broad public disclosure would. The press has been managed differently: trade journalists who have approached the company with questions about its AI deployments have received responses that emphasise the no-AI posture in creative work and that decline to enter into substantive discussion of the operational programme. The asymmetry is intentional. The company's strategic reading is that the no-AI posture has commercial value for its client base — the clients who pay €11,000 for a Kelly bag and €260,000 for an Hermès Birkin in exotic leather are paying, in part, for the assurance that the object's authorship is human in every element that matters to its identity — and that the value of the posture depends on its public visibility being uncomplicated by operational disclosure that would, in client perception, blur the line the company has drawn.

The strategic underpinning of the position is articulated more frankly in internal documents than in public statements. The Maison's commercial leadership treats the no-AI posture as a competitive moat that is becoming more, not less, valuable as the technology's penetration deepens elsewhere. In a sector in which fast-fashion competitors integrate generative AI at depth, in which mid-luxury houses experiment publicly with AI-assisted design, and in which even premium luxury maisons — Louis Vuitton most consequentially — build sophisticated archive-trained image models for their creative work, the maison that can credibly maintain a no-AI position in its craft becomes a category of one. The scarcity is not merely product scarcity. It is propositional scarcity. The Maison's intent, in Dumas's internal articulation, is to be the last European luxury house that can say with truth that no element of its craft is mediated by AI, and to hold that position for as long as the commercial value of holding it exceeds the operational cost of its constraints.

The operational AI portfolio: clienteling, allocation, leather, e-commerce, anti-counterfeit

The operational AI investment programme Hermès has assembled across the past three years covers five distinct domains, each of which sits clearly within the operational rather than creative register as the Maison defines the distinction. The first is clienteling. The Maison's client base concentrates in a relatively small number of high-value relationships managed across its 311 stores worldwide, with the most consequential relationships handled by client advisers operating under what the company calls the métier protocol — a long-form documentation discipline that captures the relational context, preference profile, and purchase history of each client across every interaction with the Maison. The protocol, historically maintained through structured manual notes, has been augmented since early 2024 by a clienteling-intelligence layer that operates against the protocol's accumulated documentation. The layer surfaces relational context to client advisers in real time during conversations with clients, identifies patterns across the protocol's database that individual advisers would not retrieve from memory, and produces structured allocation recommendations that the company's client advisers can incorporate into the discretion they exercise in offering scarce references to qualified clients.

The second domain is inventory-allocation intelligence. Hermès operates a structurally constrained production model: the Birkin and Kelly bags, the silk scarves from the Lyon ateliers, the saddles from the saddlery in Pantin, the exotic leather pieces from the small number of artisans certified in those skins — these are produced at volumes the Maison has decided rather than at volumes the market would absorb. The allocation problem this creates is acute: the company has more demand than it has product across most of its category leaders, and the discipline of how the available product is allocated across stores, across clients, and across seasonal moments is one of the Maison's most commercially sensitive decisions. The inventory-allocation intelligence layer, which the company has developed in partnership with a Paris-based data engineering firm whose work for Hermès operates under strict confidentiality, produces allocation recommendations that incorporate store-level demand patterns, client-level relational context, and category-level production realities into a coherent recommendation set that the Maison's commercial leadership reviews and ratifies on a structured cadence.

The third domain is leather-supply-chain forecasting. The Maison's leather supply is itself constrained by upstream realities the company has limited control over: the production cycles of the tanneries it has acquired across the past decade, the availability of exotic skins that pass the company's quality bar, the climatic and agricultural patterns that affect the availability of premium calfskins, alligator skins, and the small specialised hides that go into the Maison's most exclusive pieces. The forecasting layer the company has built operates against ten years of supply data, weather and agricultural data drawn from public and contracted sources, and the company's own production planning data, to produce supply forecasts that extend the planning horizon for the Maison's leather-goods division by approximately fourteen months relative to the manual forecasting practice the division operated under previously. The improvement is not glamorous. It is the kind of operational gain that protects the company against a leather shortage in a particular skin type with two years of lead time, where the previous practice would have produced the same warning with six to eight months.

The fourth domain is e-commerce personalisation. The Maison's digital channel — hermes.com — is, by the company's strategic definition, a complementary channel rather than the primary commercial channel, but its share of the company's revenue has grown steadily and it is now a structurally significant contributor to the company's commercial performance. The personalisation infrastructure that operates on hermes.com is, by international e-commerce standards, restrained: the site does not deploy aggressive recommendation surfaces, does not employ predictive merchandising in the form most retailers would recognise, and maintains a visual register that prioritises editorial presentation over conversion optimisation. Beneath this restraint, however, the company has built a personalisation engine that operates on a privacy-conservative architecture: client identifiers are managed within the Maison's own infrastructure rather than handed to third-party advertising platforms, the inferences the engine produces are constrained to a small set of editorial decisions about which product narratives a returning client is shown, and the architecture has been built specifically to preserve the editorial discipline of the digital experience while extracting commercial value from the data the channel produces. The engine is, in the company's reading, the operational layer that allows hermes.com to perform commercially without becoming an e-commerce platform in the conventional sense.

The fifth domain is anti-counterfeit. The Maison's exotic-leather and silk products are among the most counterfeited objects in the global luxury market, and the company's protection programme is one of the most sophisticated in the industry. The programme operates through multiple layers: the legal function that pursues counterfeit producers and distributors across multiple jurisdictions, the customs function that intercepts shipments at borders, the physical authentication mechanisms embedded in the products themselves (stamping, stitching patterns, particular leather treatments that experienced authenticators can read), and the digital authentication function that operates against the online marketplaces and social commerce platforms where counterfeits circulate. The digital authentication function has been the most active site of AI investment within the protection programme. The Maison has built computer-vision models that operate against the listings on resale platforms — Vestiaire Collective, The RealReal, the secondary platforms in Asia and the United States — and that flag listings whose product imagery exhibits the visual signatures of counterfeit production. The models do not produce binary authentication outcomes; they produce probability scores that the Maison's authentication specialists use as a triage instrument against listing volumes that would be unmanageable through manual review. The scaling effect has been material: the number of counterfeit listings the Maison's protection programme can act against in a given calendar quarter has approximately quadrupled since the models entered production in 2024.

The talent architecture: who Hermès has hired

The talent acquired to build the operational portfolio has been recruited under a hiring discipline consistent with the Maison's posture. The senior figure responsible for the programme is Isabelle Vernet, who joined Hermès in late 2023 from a senior client-data role at the LVMH group. Vernet's background is operational rather than technical — fifteen years at LVMH covering client data architecture, e-commerce infrastructure, and the group's anti-counterfeit programme. Her appointment was deliberately understated within the company: she was hired into a role described internally as Director of Client and Commercial Operations, not into a chief data officer or chief intelligence officer title, and her position in the Maison's organisational chart sits within the commercial function rather than within a horizontal technology function. The placement is a signal. Hermès has chosen not to create a chief intelligence officer role of the kind that Audemars Piguet, Vacheron Constantin, Patek, and Brunello Cucinelli have created. The work that those titles describe is being done. The title that would make it visible has been declined.

Beneath Vernet, the programme has assembled a team of approximately forty by mid-2025, drawn from a mixture of luxury-sector backgrounds and from outside hires that the Maison's talent function has executed under terms that emphasise alignment with the company's discretion culture. The most consequential outside hires have been a small group of senior data engineers and machine-learning specialists drawn from Paris-based fintech operations, from the Paris office of a major Silicon Valley platform company, and from European consulting firms whose luxury-sector practices have produced operators with both technical depth and cultural fluency in the industry's discretion norms. The hires have been individually significant — several of them have brought architectural experience that the Maison would have had difficulty building organically — and they have been collectively invisible. None of these hires have been publicly announced. None of them have public profiles that would surface their roles at the Maison to outside observers. The cultural discipline of the Maison around its talent practice extends to its AI programme without exception.

The technical infrastructure underlying the programme has been built with a preference for European cloud providers and European-resident compute, partly for regulatory reasons — the privacy posture the Maison maintains is materially stricter than industry norms — and partly for the cultural alignment that the Maison's executive committee has prioritised in its vendor selection. The infrastructure is hosted across a mixture of providers, none of which the Maison has identified publicly. The vendor contracts include provisions around data residency, weights persistence in cases where the Maison's vendors operate on AI models that are not the Maison's own property, and clean-termination terms that the Maison has insisted on as the baseline for any operational AI vendor relationship. The contracts are, by the description of two people who have negotiated them, more restrictive than the Maison's commercial scale would normally command, but the Maison has been prepared to absorb the friction of negotiating against the standard vendor terms because the friction is itself an expression of the discipline the company applies to the programme.

The programme's relationship with the artisan-facing dimension of the Maison is the most carefully managed interface. The artisans in Pantin, in the Saint-Louis crystal works, in the Bobinet silk-printing facilities outside Lyon, and in the smaller specialised ateliers across France that produce the Maison's product, have been deliberately insulated from the operational AI programme by an architectural choice the company's leadership made early. The systems the programme operates do not touch the production-floor environments where the artisans work; they operate in commercial, supply-chain, and digital domains whose connection to the artisan floor is mediated through human decision-makers whose role is to translate operational intelligence into operational instructions without exposing the artisans themselves to the system that produced the intelligence. The choice is partly cultural — the Maison's reading of artisan culture is that the introduction of AI-mediated systems into the artisan environment would corrode the relationship between artisan and craft on which the Maison's product rests — and partly strategic, because the visible absence of AI in the artisan environment is the commercial proposition the no-AI posture defends.

The internal taxonomy: operational versus creative AI

The distinction between operational AI and creative AI is, in the Maison's internal vocabulary, the architectural principle that organises every investment decision in the programme. The vocabulary has been articulated in internal documents, in working sessions with the company's executive committee, and in the briefing materials the Maison's investor relations function uses when engaging with analysts who probe the company's technology investment. The principle is straightforward in form and demanding in application. Operational AI is technology that is used in the management of the company's operations: in client relationships managed by humans whose work the technology supports, in supply chains whose physical realities the technology forecasts against, in commercial decisions whose ratification remains with human leadership, in anti-counterfeit work whose objective is the protection of human craftsmanship against industrial imitation. Creative AI is technology that participates in the design or making of the Maison's product: in the work of the artisans, in the editorial decisions about what the Maison produces and how it is presented, in the visual and material identity that the Maison's clients pay for when they pay for an Hermès object.

The Maison is explicit, internally and in the briefings that go to its commercial partners, that the boundary between the two categories is not always clear and that the work of policing the boundary is itself a discipline. The clienteling intelligence that operates against the métier protocol is operational because it supports the human client adviser whose work is the relationship; the personalisation engine on hermes.com is operational because the inferences it produces inform editorial decisions ratified by human curators rather than algorithmic ones; the anti-counterfeit models are operational because their outputs feed into human authentication processes rather than into automated takedown actions. The boundary cases are managed by review processes that the Maison's leadership treats as architectural in importance. A proposal to extend the personalisation engine to surface generated product imagery would, by the taxonomy, cross from operational into creative and would be rejected. A proposal to extend the clienteling intelligence to generate suggested conversational language for client advisers would, by the same taxonomy, cross the boundary and would be rejected. The lines are held.

The taxonomy has, in practice, produced a distinct competitive posture relative to LVMH and Kering. Both of those groups have made public investments in creative AI — LVMH at the depth Intelar has reported on in the Louis Vuitton design pipeline, Kering through a series of more cautious experiments at Saint Laurent, Gucci, and Balenciaga that have been publicly disclosed and selectively integrated into the maisons' creative work. Hermès has made comparable operational investments and no creative investments, by the standard the taxonomy applies. The competitive consequence is that Hermès has, by deliberate choice, foregone the productivity gains and the design-velocity gains that LVMH and Kering have started to capture, and has instead absorbed the operational cost of the foregone gains in exchange for the propositional value of the no-AI posture in creative work. The trade-off is explicit in the Maison's strategic planning. It is not an oversight or a competitive deficit. It is a deliberate market position.

The boundary discipline has implications for the talent the Maison hires. The senior figures recruited into the operational programme have been hired explicitly on the understanding that the work they will do does not extend into creative domains, and that the career growth available within Hermès is constrained relative to the comparable roles available at LVMH or Kering by the structural absence of a creative-AI track. Several of the senior hires Vernet has made have been publicly understood within the Paris luxury-technology community as career-defining moves into a maison whose discipline they wanted to operate within, accepting the constraint on creative-domain work as an explicit part of the package. The recruitment posture has produced, in aggregate, a programme whose internal culture reflects the same discipline the Maison applies to its artisan culture. The technologists are not being asked to participate in the Maison's creative work. They are being asked to build the operational substrate that allows the creative work to remain entirely human.

The contrast with LVMH and Kering: three readings of one technology

The three large European luxury groups have arrived at three distinct postures on AI, and the divergence has become a structural feature of the competitive landscape rather than a transient difference in the pace of adoption. LVMH has built the most aggressive creative-AI integration in the sector, with the Louis Vuitton archive-trained image model the most visible and most sophisticated expression of a doctrine that extends across Dior, Celine, Fendi, Loewe, and the textile bureaux at Loro Piana. The doctrine — that creative AI is acceptable within an architecture that preserves designer authorship, archive sovereignty, and brand-protection guardrails — is one that LVMH considers the only path to maintaining luxury's design productivity against fast-fashion competitors who have started to use AI at depths that compress design cycles to levels luxury cannot match without proportional investment. The doctrine is a competitive response, and it is also a strategic bet that controlled creative AI is compatible with luxury craft.

Kering's posture has been more cautious in its public expression and broader in its experimental scope. The group has supported AI experimentation at its individual maisons under terms that have varied by house — Saint Laurent has integrated AI into its e-commerce and merchandising operations with the same depth Hermès has within hermes.com, Gucci experimented with AI-assisted design under former creative director Sabato De Sarno before his departure in 2025, and Balenciaga has been the group's most exploratory house in its public creative work — without converging on a group-level doctrine that would parallel LVMH's. The lack of a unified posture has been criticised by analysts who have argued that Kering is failing to commit to either of the two coherent positions in the sector, but the group's leadership has defended the variation as appropriate to a portfolio whose houses operate under more diverse creative cultures than LVMH's. The competitive consequence is that Kering occupies a middle ground between LVMH's controlled creative integration and Hermès's no-AI posture, with neither the productivity gains of the LVMH approach nor the propositional purity of the Hermès approach.

Hermès's posture is the most distinct of the three. The no-AI position in creative work, combined with the substantial operational AI investment in commercial domains, produces a maison whose visible identity is constructed around the absence of a technology that other houses are using openly. The market consequence has been that the maison's positioning has differentiated more sharply through the period of broader AI adoption than it had previously. Hermès's revenue growth has continued at the pace that has characterised the maison for the past decade — high single digits to low double digits on a constant-currency basis — and the maison's pricing power has, by the analyst consensus, strengthened during a period in which competitors have lost pricing leverage to consumer scepticism about whether AI-mediated luxury commands the same premium as human-mediated luxury. The data points are consistent with the strategic reading the maison's leadership has held: that the propositional value of the no-AI posture is increasing as AI adoption deepens elsewhere, and that the operational cost of holding the position is more than offset by the commercial benefit of being the maison that can credibly claim the position.

The three postures have, in combination, created a competitive landscape in which the question of AI's place in luxury has become a positioning decision rather than a technical decision. A maison that has invested heavily in operational AI but maintains a no-AI position in creative work is not the same competitor as a maison that has integrated creative AI under controlled architecture, and neither is the same as a maison that has experimented diffusely without a coherent group doctrine. The three competitors are pursuing three different commercial propositions, and the technology investment is one of the variables that defines the proposition rather than a horizontal capability that all of them are racing to acquire. The structure of the competition has, in this sense, become more interesting than the trade press's framing of AI adoption rates has acknowledged. Hermès is not behind. It is operating from a position that is, by deliberate construction, not comparable to its peers' on the dimension the peers measure. The question of which position proves most durable is the question the next five years of the sector will answer.

What to watch

Hermès's posture is the most strategically distinctive in the European luxury portfolio, and its sustainability depends on conditions that the next three years will test. The operational programme is mature enough to sustain itself; the public posture's robustness against changing market expectations is less certain.

• Whether the no-AI posture withstands competitive pressure if LVMH's controlled creative-AI doctrine produces visible improvements in design velocity, product range, or merchandising responsiveness that translate into commercial outperformance; the posture's value depends on its propositional purity, and competitive outperformance by an alternative posture would test whether the value remains net positive against the operational cost of the constraint.
• Whether the boundary between operational and creative AI holds as the operational portfolio's outputs become increasingly sophisticated; the clienteling intelligence layer that surfaces relational context to client advisers could, plausibly, evolve toward surfacing suggested conversational language, and the boundary that the Maison currently maintains between operational support and creative substitution will face increasing internal pressure as the technology improves.
• Whether the talent the Maison has assembled remains within the constrained career architecture the no-AI posture creates, or whether the senior figures who have built the operational programme begin to leave for groups whose AI portfolios offer broader creative-domain work; the Maison's hiring posture has, so far, attracted technologists who value the constraint, but the talent market for senior AI operators has tightened materially since 2023 and the comparative attractiveness of Hermès's package is, by the description of two recruiters active in the segment, more sensitive to comparable offers than it was eighteen months ago.
• Whether the disclosure asymmetry between investor briefings and public communication holds; the Maison's communication function has managed the asymmetry carefully through 2024 and 2025, but the increasing investor-community attention to AI investment at luxury houses is creating disclosure pressure that the Maison's no-AI public posture will find harder to absorb if a particular investor inquiry or trade press investigation makes the asymmetry visible to retail clients.
• Whether the operational programme's outputs become a visible competitive advantage that the Maison's leadership decides to disclose more openly; the anti-counterfeit programme's scaling effect — quadrupling the listing volume the Maison can act against — is the operational gain most aligned with values the Maison's clients explicitly endorse, and the strategic question of whether to make that gain visible in the way LVMH has made its archive-sovereign training data visible is a disclosure decision that the next eighteen months will likely force.

Frequently asked

What does the Maison's internal taxonomy of operational versus creative AI cover, and how is the boundary policed?

Operational AI is technology used in the management of the company's operations — clienteling intelligence that supports human client advisers, supply-chain forecasting that informs human production planning, anti-counterfeit models whose outputs feed into human authentication review, e-commerce personalisation that informs editorial decisions ratified by human curators. Creative AI is technology that participates in the design or making of the Maison's product or in the editorial identity it presents to clients. The boundary is policed by structured review processes within the company's executive committee, which has rejected proposals that would extend operational systems into creative-domain functions — including, in known cases, proposals to extend the personalisation engine to generate product imagery and proposals to extend the clienteling intelligence to suggest conversational language to client advisers.

What has Axel Dumas said publicly about the Maison's posture, and how should the trade press's summaries be calibrated against his statements?

Dumas's reference formulation — that "Hermès does not employ artificial intelligence in the work of its artisans, and it will not" — has been restated consistently across his public statements since the Les Échos interview in March 2024. The formulation is precise: it addresses the work of the artisans and not the work of the company's commercial, supply-chain, client-service, or anti-counterfeit functions. Trade press summaries that describe the Maison as having a categorical no-AI position misread Dumas's framing. The Maison's position is that AI has no place in creative work and that operational technology investment is governed by the standard expectations of responsible enterprise management. The position is genuine in both registers.

Who is Isabelle Vernet and what does her role architecture say about the Maison's posture?

Isabelle Vernet joined Hermès in late 2023 from a senior client-data role at LVMH after fifteen years across client data architecture, e-commerce infrastructure, and anti-counterfeit programmes within the group. She is responsible for the operational AI programme at Hermès. Her role is titled Director of Client and Commercial Operations rather than a chief data officer or chief intelligence officer title, and she reports through the commercial function rather than through a horizontal technology function. The architecture is deliberate: the Maison has chosen not to create a visible chief intelligence officer role of the kind Audemars Piguet, Vacheron Constantin, Patek Philippe, and Brunello Cucinelli have created, because the title would make visible a programme whose architecture relies on remaining structurally subordinate to the Maison's commercial leadership.

How does the operational programme contrast with LVMH's and Kering's postures, and where does Hermès sit in the competitive landscape?

LVMH has built the sector's most aggressive controlled creative-AI integration, with the Louis Vuitton archive-trained image model extending across Dior, Celine, Fendi, Loewe, and Loro Piana on a shared doctrine that creative AI is acceptable within an architecture that preserves designer authorship, archive sovereignty, and brand-protection guardrails. Kering has experimented more diffusely across its maisons without a unified group doctrine. Hermès has refused creative AI entirely and invested operationally at depth. The three postures represent three distinct commercial propositions rather than three speeds of adoption along a single axis. The Maison's analyst-consensus pricing power and revenue trajectory through 2024 and 2025 are consistent with the strategic reading that propositional purity in creative work is increasing in value as broader sectoral adoption deepens.

Why are the artisan-facing dimensions of the company specifically insulated from the operational programme?

The artisans in Pantin, in the Saint-Louis crystal works, in the Bobinet silk-printing facilities outside Lyon, and across the smaller specialised ateliers, are insulated from the operational AI programme by an architectural choice the Maison's leadership made early. The systems do not touch the production-floor environments where artisans work; they operate in commercial, supply-chain, and digital domains whose connection to the artisan floor is mediated through human decision-makers. The insulation is partly cultural — the Maison's reading of artisan culture is that the introduction of AI-mediated systems into the artisan environment would corrode the relationship between artisan and craft — and partly strategic, because the visible absence of AI in the artisan environment is the commercial proposition the no-AI posture defends.

What is the practical scale of the anti-counterfeit programme, and what has the AI investment produced in measurable terms?

The Maison's anti-counterfeit programme operates against the resale platforms and social commerce channels where counterfeits of its products circulate. The computer-vision models it has built since 2024 produce probability scores against listing imagery that the Maison's authentication specialists use as a triage instrument. The scaling effect has been the most concretely measurable output of the operational programme: the number of counterfeit listings the protection programme can act against in a given calendar quarter has approximately quadrupled since the models entered production. The volume gain is the operational result most aligned with values the Maison's clients explicitly endorse, and it is the gain the company has discussed most openly with senior investors during private briefings.