+150 XP

Build vs buy vs partner for luxury AI

Build vs partner vs buy for luxury AI

A creative director at a European leather goods maison wants visual search. A customer photographs a stranger's bag on the street, uploads it, and the app finds the closest match in the maison's current collection. The technology exists. Whether the project succeeds depends less on whether the team can build it and more on a different choice: build it, buy it, or plug into someone else's platform.

For most luxury executives this is the most consequential AI decision of the decade. It often gets made badly. Teams tend to adopt whichever option the loudest vendor or the most confident engineer pitches first.

Three paths, one recurring mistake

Build: an in-house team develops the model, owns the code and controls the roadmap. Chanel's internal data science work and LVMH's investment in proprietary AI tooling sit here. LVMH also scouts startups through its annual LVMH Innovation Award and its Maison des Startups incubator at Station F.

Buy (license): you pay a vendor for a packaged solution. Fashion-tech vendors sell subscription access to models they trained, often on cross-brand data. Examples include Heuritech (trend forecasting), Vue.ai (visual merchandising and search) and Lalaland.ai (AI-generated model imagery, now part of 3D design company Browzwear after a 2023 acquisition).

Partner: you integrate with a platform that already has the audience, the infrastructure and often the AI layer. Alibaba's Tmall Luxury Pavilion is the standard example: a maison plugs into someone else's stack instead of owning or licensing a standalone tool. Farfetch's Platform Solutions business, which ran e-commerce for several brands, was the other common example until Farfetch's near-collapse (more on that below).

The recurring mistake is to treat this as a procurement decision about cost, speed and vendor reputation. It is really a decision about strategic control, and three axes frame it.

The decision framework: three axes that matter more than price

1. Brand control

Does the AI touch anything customer-facing that carries brand voice, image or aesthetic judgment? Visual merchandising, styling recommendations, chatbot tone and generated imagery all express the brand.

Tasks with high brand sensitivity push toward build or a tightly customized buy. A virtual try-on that must render fabric drape exactly as the ateliers intend is one example. A shared vendor model trained across many brands tends to drift toward a generic aesthetic middle.

Tasks with low sensitivity don't touch the brand at all. Fraud detection on payments and warehouse inventory forecasting are examples. For these, partner or buy freely.

2. Data ownership

Who keeps the customer interaction data, the images and the purchase signals that the AI system generates?

This matters for two reasons.

The first is competitive. Many fashion-tech vendors serve several brands in the same category. If a maison's customer data trains a vendor's shared model, it may indirectly improve a competitor's results.

The second is regulatory. Under the EU's GDPR, contracts must define two roles:

  • The "data controller" decides how data is used.
  • The "data processor" handles the data on the controller's behalf.

Platform partnerships often make the platform a joint controller of shared customer data. That is a very different position from a pure licensing deal, where the maison usually keeps control.

The EU AI Act now adds a second layer. Obligations for general-purpose AI models have applied since August 2025, and further obligations are phasing in. Contracts should state who handles transparency duties, such as labelling AI-generated imagery.

Ask concretely: if we exit this vendor tomorrow, do we leave with the trained model, the raw data, both, or neither? Most vendor contracts, if nobody examines them, give you neither.

3. Reversibility

This axis asks how expensive and slow it would be to unwind the choice in two years if it underperforms or something better appears.

Build is slow to start but highly reversible, because you own everything. You can change the model without renegotiating anyone's contract.

Partner is fast to start but usually the least reversible. Farfetch shows why. In late 2023 the company ran out of cash, and Coupang rescued it in a deal completed in January 2024. Brands that had embedded their catalog, pricing and customer relationships in Farfetch's ecosystem had to plan migrations at short notice, and several moved off. Leaving meant rebuilding distribution, not swapping software.

Buy sits in the middle. Switching vendors is usually possible under the contract but disruptive in practice: you have to re-integrate systems, retrain staff and migrate data.

A simple weighted scoring exercise

Score each option from 1 to 5 on each axis, then weight the axes by what matters most for the use case. Here is a worked example for the visual search tool from the opening scene.

AxisWeightBuildBuy (vendor)Partner (platform)
Brand control0.4532
Data ownership0.35531
Reversibility0.25432
Weighted score4.653.001.65

Calculation for Build: (5 × 0.4) + (5 × 0.35) + (4 × 0.25) = 2.0 + 1.75 + 1.0 = 4.65.

The scoring is illustrative, and the weights should shift by use case. Take a back-office task like demand forecasting. Brand control drops toward 0.1, cost and speed matter more, and the outcome often flips toward buy or partner.

Where the calculus flips: cost, speed, and talent

The framework deliberately leaves out upfront cost, but cost can't be ignored. Building capable computer vision or generative AI systems in-house takes three things:

  • machine learning engineers
  • annotated training data
  • ongoing compute spending

Most maisons, even large ones, lack these at the depth pure tech companies have. Before committing to build, read Google's guide to responsible AI practices. It is a free reference on what internal AI governance requires day to day.

This is why LVMH and Kering, groups with real balance sheets, still license specialized vendors for narrow tasks like trend forecasting and counterfeit detection. Build makes sense for the few capabilities that set the brand apart and will keep doing so. For everything else, buy or partner is usually the rational choice.

Knowledge check

1. According to the lesson, what is the recurring mistake luxury executives make when choosing between build, buy, and partner for AI?

2. A maison is evaluating an AI chatbot that will interact directly with customers using the brand's distinctive tone and aesthetic sensibility. Which axis of the decision framework is most directly implicated?

3. Why might licensing (buying) a vendor's AI model, such as one trained on cross-brand data, pose a strategic risk for a luxury maison specifically?

MULTIPLE CHOICE

4. Select ALL correct answers describing characteristics of the 'partner' path in the build/buy/partner framework.

Select all the correct answers.

MULTIPLE CHOICE

5. Select ALL correct answers about why this build/buy/partner decision is described as 'the single most consequential AI decision' for luxury executives.

Select all the correct answers.

Applying the framework: three quick scenarios

Scenario A: a maison wants a customer service chatbot that handles order status queries.

  • Brand sensitivity is low: the task is functional, not aesthetic.
  • Data concerns are moderate: order data is sensitive, but it won't define a proprietary model.
  • The need for reversibility is high, and chatbot vendors are commoditized, so switching later is easy.

Verdict: buy.

Scenario B: a maison wants an AI system that recommends which fabrics and cuts to produce next season. It would be trained on the house's own archive and sales history.

  • Brand control matters a great deal, since this is core creative judgment.
  • The data stakes are high, because this is the house's design IP.
  • Tolerance for mistakes is low: a bad call damages the collection, not just a feature.

Verdict: build, or at minimum license a heavily customized, exclusive model where the maison owns the outputs.

Scenario C: a maison wants presence in a major Asian marketplace's AI-curated recommendation engine to reach new customers.

  • You have little control over the algorithm itself, since you don't control Alibaba's recommendation logic.
  • The data-sharing trade-offs are real.
  • Reversibility is low once you are integrated.

Verdict: partner. Negotiate data and exit terms explicitly before signing, and treat this as a distribution decision as much as a technology one. Farfetch showed that platforms can change owners or strategy quickly.

🎬 [VIDEO: "Build vs Buy: The Innovator's Dilemma for AI" - youtube.com - search for recent talks from MIT Sloan or Harvard Business Review on AI make-or-buy decisions in retail and consumer brands, applicable directly to luxury]

Key Takeaways

  • Score every AI opportunity on brand control, data ownership and reversibility before discussing price or vendor reputation. These axes predict long-term regret better than upfront cost does.
  • Reserve build for capabilities that set the house apart and will keep doing so, such as design-related work. Functional or commoditized tasks like chatbots, fraud detection and inventory forecasting should default to buy or partner.
  • Before signing, get written answers on who owns the trained model and the raw customer data, what happens to both if the maison exits, and who carries EU AI Act duties.
  • Platform partnerships offer speed and reach but are the hardest to reverse. Farfetch's 2023 collapse and sale to Coupang showed how quickly a partner can become a risk.
  • Revisit these decisions at least yearly. Vendors get acquired (Lalaland.ai joined Browzwear), platforms change hands, and a 2024 buy decision may look different in 2026.