How Zalando Built an Internal AI Platform Without Buying a Startup
· Last updated:Zalando has opted to build its own internal AI platform to maintain absolute control over its unique customer data and ensure that machine learning models are deeply integrated into its existing logistics and customer experience stacks. By partnering with infrastructure providers like Databricks rather than acquiring niche AI startups, the company avoids the high cost of integration while building a proprietary moat around its size, fit, and personalization capabilities.
Key takeaways
- Zalando prioritizes "data sovereignty," ensuring all customer and inventory data remains within a unified, internally managed environment.
- The company uses a "Lakehouse" architecture to bridge the gap between raw data storage and active machine learning deployment.
- In-house development allows for specialized solutions, such as their proprietary size-recommendation engine, which reduces return rates.
- The decision to build rather than buy signals a higher barrier to entry for startups looking to sell generic AI tools to major European platforms.
- Strategic partnerships with cloud infrastructure leaders provide the scale needed without the baggage of external startup culture.
Why did Zalando choose an in-house AI strategy?
For a platform serving over 50 million active customers, the primary challenge is not just having AI, but having AI that understands the specific nuances of European fashion retail. When you look at the Zalando ecosystem, the complexity of managing thousands of brands across different languages and sizing standards is immense.
In our experience, large fashion entities often find that third-party AI startups offer "black box" solutions. These tools might perform well in a silo, but they struggle when required to communicate with a bespoke warehouse management system or a proprietary customer loyalty program. By building internally, the company ensures that every algorithm is a native citizen of their technical environment. This approach allows them to iterate faster on specific business goals, such as reducing the environmental and financial cost of returns, rather than waiting for a vendor's roadmap to align with their needs.
How does the Databricks partnership support this internal platform?
Instead of buying a company that makes AI, Zalando bought into the infrastructure that makes AI possible at scale. The partnership with Databricks is central to this. By utilizing a Lakehouse architecture, the engineering teams can store vast amounts of unstructured data—like product images and customer clickstreams—and structured data—like purchase history—in one place.
This unified layer is critical because it removes the friction between data engineers and data scientists. In many organizations, these two groups work in different worlds. At Zalando, the platform provides a shared workspace where models can be trained, tested, and deployed into production within hours. This infrastructure-first approach means they don't need to acquire a startup for its "talent" or its "tech" because they have already built the factory where that tech is produced.
What specific fashion problems is the platform solving?
The internal platform isn't just a theoretical exercise; it is focused on high-impact retail problems that directly affect the bottom line.
Solving the "Size and Fit" dilemma
One of the most significant costs in fashion e-commerce is the return rate, often driven by inconsistent sizing across brands. Zalando’s internal teams have developed models that predict whether a garment will run small or large based on historical return data and brand-specific measurements. Because this model is built on their own data, it is more accurate than any generic tool a startup could offer. They aren't just guessing; they are using millions of data points from actual customer behavior to refine the fit experience.
Hyper-personalization of the digital storefront
Every time a user opens the app, the AI platform is working to curate a "personal shop." This goes beyond simple "you might also like" carousels. The system analyzes real-time intent, seasonal trends, and even local weather patterns to surface products. By building this internally, they can ensure that the personalization engine respects strict European privacy standards (GDPR) while still delivering a highly relevant experience.
Build vs. Buy: A comparison for fashion enterprises
| Strategy | Best For | Limits |
|---|---|---|
| Internal Platform Build | Large enterprises with massive proprietary datasets and high engineering capacity. | High upfront cost and continuous maintenance requirements. |
| Startup Acquisition | Rapidly entering a new, specialized market or acquiring specific patents. | Difficulties in data integration and cultural alignment. |
| SaaS Subscription | Small to mid-sized brands needing standard industry solutions. | Limited customization and no ownership of the underlying IP. |
What does this mean for the fashion-tech startup ecosystem?
The "build-first" mentality of giants like Zalando creates a challenging environment for startups. If a major platform can build a functional version of your product in-house, your value proposition must shift from "we have AI" to "we have a unique dataset or a specialized patent you cannot replicate."
We see a trend where the most successful startups are those that focus on very narrow, highly technical niches—such as circularity tracking or advanced textile analysis—rather than broad personalization or recommendation engines. As the next phase of the startup ecosystem evolves, as noted by observers of the Silicon Valley and LatAm corridors, the focus is shifting toward deep research and sustainable AI that solves problems the big platforms haven't prioritized yet.
What are the risks of the "build" approach?
While the benefits are clear, building an internal AI platform is not without its dangers. The most prominent risk is the "talent war." Maintaining a world-class AI team requires constant investment and a culture that rivals top-tier tech firms. If key engineers leave, the internal platform can quickly become a legacy burden that is difficult to update.
There is also the risk of insular thinking. By relying solely on internal development, a company might miss out on breakthrough innovations happening in the wider startup world. To mitigate this, many large players maintain open innovation programs, similar to how other brands manage their tech stacks, ensuring they stay connected to the broader research community without becoming dependent on it.
Conclusion
Zalando’s strategy proves that for the largest players in fashion e-commerce, the most valuable asset isn't the AI itself, but the data and the infrastructure that powers it. By choosing to build a unified platform rather than a patchwork of acquired startups, they have created a scalable, flexible system that is deeply integrated into their business DNA. For startups, the message is clear: to win in this environment, you must offer something that a 1,000-person engineering team cannot simply build on a Tuesday afternoon.
For more on how companies navigate these choices, read about 6 Fashion-Tech Startups That Rebuilt Around a Niche and Survived.
FAQ
Why doesn't Zalando just buy AI startups for their talent? While "acqui-hiring" is common in Silicon Valley, Zalando has found that integrating external teams into their specific culture and technical stack often takes longer than training internal talent. By providing a world-class platform like Databricks, they attract top engineers who want to work on large-scale problems without the friction of legacy integration.
How does an internal AI platform help with sustainability? The primary sustainability lever is reducing returns. By using AI to ensure customers get the right size the first time, the company significantly lowers the carbon footprint associated with shipping and processing returned goods. Internal platforms allow these sustainability metrics to be baked directly into the model's success criteria.
Is building an internal platform cheaper than buying a startup? In the short term, no. It requires significant capital expenditure and a high payroll. However, in the long term, it is often more cost-effective because the company avoids recurring licensing fees and the massive "integration tax" required to make third-party software work with internal systems.
Can smaller brands replicate Zalando's build strategy? Most small to mid-sized brands lack the data volume and engineering budget to build a full-scale AI platform. For them, a hybrid approach—using specialized SaaS tools for non-core functions while focusing internal efforts on their most unique data—is usually the most viable path.
What role does generative AI play in this internal platform? Generative AI is treated as another toolset within the platform. Instead of buying a separate "GenAI" startup, the company integrates large language models into their existing data lakehouse. This allows them to create features like AI fashion assistants that have access to real-time inventory and customer preferences.
How does Zalando ensure data privacy on an internal platform? By keeping the data in-house, the company has a single point of control for privacy compliance. They can implement strict access controls and auditing across the entire platform, which is much harder to do when customer data is being sent to multiple third-party AI vendors.
Further reading * Zalando Corporate News * Databricks Industry Solutions