Intelligent Retail Architecture: Balancing composability and integration in the Age of AI

MARKET VIEW

Interview with

Vice President of Technology for ZEOS Tech and Services at Zalando

About Luis Afonso Ribeiro

Luis Afonso Ribeiro serves as Vice President of Technology for ZEOS Tech and Services at Zalando, leading the technology strategy for this rapidly growing B2B logistics business unit. With over 21 years of experience in software engineering and technology leadership, he has built his career around creating digital experiences and scalable tech solutions across diverse industries including e-commerce, retail, luxury, manufacturing, and defence. Prior to Zalando, he spent nine years at Farfetch, where he played a pivotal role in the company's platformisation strategy. Luis holds an International Executive MBA from IE Business School and leads a 130-person engineering team focused on building innovative B2B solutions that help merchants and partners thrive in the European fashion logistics ecosystem.

How are you balancing the needs for composable architecture with the complexities of systems integration in your retail technology stack?

The industry has moved towards composability, but extreme composability creates significant integration challenges. The complexity required demands technical expertise that isn't available to all companies or merchant sizes.

I'm looking for pragmatic composability—clearly understanding and separating the main problems through domains like catalogue management, inventory, and order management. You define clear boundaries, then achieve composability by combining multiple domains, sometimes introducing vendor technology, other times complementing with custom-built software.

Instead of pursuing everything composable, we focus on where composability adds business value and flexibility. Some areas benefit more from full-stack solutions supporting wider processes. It's not about composability as an end goal, but understanding where it helps achieve our needs and solve pain points.

What role do you see AI playing in reshaping retail architecture, particularly in terms of real-time decision-making and personalisation?

AI has been in retail for years through personalised recommendations and customer segmentation. We're seeing a new trend with technologies like vector databases bringing new capabilities and algorithms, reinforcing a revolution in personalisation.

The cadence of innovation is becoming much faster. Previously, you'd train models for long periods, such as every week or several days. Now we're seeing updates in hours, probably even instant updates. This means better customer experience through real-time insights, driving better conversion.

We're seeing sophisticated AI support for decision-making. Businesses are moving to data-driven approaches, with AI democratising decision-making support. This is particularly evident in retail planning functions and supply chain management, where predictive analytics, trading insights, returns autopilots, and advanced analytics for optimising returns and order allocation are being powered by AI. Gen AI will drive new tools and algorithms that will be applied in business moving forward.

How are you approaching the integration of legacy systems with modern cloud-native solutions?

We've been cloud-native for many years, designing applications for cloud first. When modernising monolithic parts, we apply the Strangler Fig pattern—creating boundaries between legacy systems through APIs and well-defined events for data capture, then slowly phasing them out.

You reduce usage of legacy components by replacing them with new functionality whilst keeping systems up until fully decommissioned. They're still used for reads, but you shift write operations to the new system. Meanwhile, you maintain data migration and synchronisation between systems until the legacy system is reduced to practically nothing.

It's an iterative approach with well-defined boundaries on how systems are used by multiple actors. It starts by identifying legacy systems—the users, processes they support, and typical use cases—so you can map out the phasing strategy and prioritise your cloud-native solution implementation.

What challenges have you encountered in maintaining system reliability whilst pursuing architectural innovation?

The challenges of introducing new technology aren't usually technical—it's the human aspect. When doing analysis and replacement, use cases that weren't correctly captured can break operational processes when introducing new technology.

We avoid this by ensuring detail and analysis are communicated upfront, involving different stakeholders early, and clearly identifying the cutover plan before implementation. It's mostly about change management because I'm more concerned about process failures than system bugs.

For technical elements, we rely on system observability to understand real-time impacts and limit the blast radius. Big bang approaches for innovation projects are risky. We put capabilities behind feature flags and use graduated rollouts like canary deployments. We can release new capabilities in phased rollouts, understand how processes evolve from real usage, then gradually expand—whether increasing users or moving to different regions. This minimises impact on reliability when introducing new technology.

How do you evaluate the trade-offs between custom development and packaged solutions in your tech stack?

Being a tech geek, I prefer custom development whenever possible, but there are clear rules. Custom development happens when we're investing in building something unique where we can make a difference—managing capabilities we believe are unique to us.

We select packaged solutions for commoditised areas like ERPs, finance software, HR, and CRM platforms like Salesforce. There's no point reinventing the wheel when we can benefit from existing processes. Custom development is for differentiating factors.

We use a scorecard for major technical decisions, evaluating total cost of ownership, flexibility to maintain and evolve over time, whether it deals with critical data, and potential lock-in. We make decisions transparent within the organisation—ensuring business stakeholders understand the rationale and internal teams have proper documentation on why we chose particular technologies. Making the decision-making process transparent is important.

What approaches have you found effective for scaling your architecture to handle peak retail periods?

Considering the tremendous usage of our technology, we run mostly with auto-scaling across most domains. We design services as microservices that are as stateless as possible. When stateful, we ensure it's self-contained in smaller services we can independently scale, leveraging synchronisation through events across our infrastructure.

This design allows us to benefit significantly from auto-scaling elasticity. It requires upfront design and investment, but that investment has paid off with our ability to absorb peaks and treat peak trade as normal periods.

For special events like Black Friday and Christmas, we employ proper observability and capacity planning based on past projections. We set scaling limits correctly and remove impediments. We also do scenario planning and recurrent load testing to constantly evaluate and identify future bottlenecks. Auto-scaling design with proactive capacity planning has been crucial for handling retail peak periods effectively.

about

Zalando SE

ZEOS is part of Zalando SE, leveraging the company’s infrastructure, technology, and expertise to build an operating system for fashion & lifestyle. Established in 2023, ZEOS is developing tools that allow brands and retailers to manage their multi-channel business in one place. ZEOS helps fashion & lifestyle industry players unlock Europe’s full potential by enabling seamless e-commerce across the continent: from new market entry and expansion to optimisation and business steering. Our journey starts with logistics. The first ZEOS product is called ZEOS Fulfilment. It gives brands and retailers the ability to fulfill their multi-channel sales using one stock pool, one integration, and one user interface. ZEOS Fulfilment takes advantage of an extensive infrastructure spanning 23 European markets, 12 fulfilment and 20 return centres, over 40 carrier partnerships, and 165 localised delivery and returns services. ZEOS is an independently operating unit following strict data compliance rules. Around 350 people work for ZEOS in Berlin and in Stockholm.

Connect with us and let's shape the media's future together.

Globant is a digitally native company that helps organizations reinvent themselves and unleash their potential. They bring innovation, design and engineering together at scale to create impactful solutions. Globant specializes in digital strategy, design, and development, leveraging cutting-edge technologies and trends. With their agile pods methodology and commitment to innovation, Globant is a trusted partner for top brands looking to lead their industries in the digital landscape. They create digital transformations using disruptive technologies like AI, blockchain, and cloud computing. Major clients include Google, EA, and Disney. Globant bridges the gap between design and engineering to develop innovative software products. Overall, Globant helps global organizations reinvent themselves digitally.

Globant-UK@globant.com