Arjan van den Heuvel of Xebia applies the principles of the Team Topologies framework to the organizational design of Data & AI teams through concrete case studies. The article explores how different organizational structures impact the effectiveness of AI initiatives depending on company size and AI maturity.
Theoretical foundations
The Team Topologies framework defines four fundamental topologies (stream-aligned, platform, enabling, complicated subsystem teams) and three interaction modes (collaboration, X-as-a-service, facilitating). Conway's Law states that system architecture reflects the communication structure of the organization that produces it. These principles make it possible to analyze and design more effective Data & AI organizational structures.
Case 1: Mid-sized company, basic AI experience
The article examines three scenarios for a company that started its AI initiatives a few years ago. Scenario 1.1 (decentralized experts) leads to point-to-point solutions without coordination, creating a technological patchwork. Scenario 1.2 (centralized Data & AI product team) generates excessive communication load for the Product Owner and limits the team's autonomy in the face of multiple stakeholders.
Scenario 1.3 (Data & AI expert pool) emerges as the optimal solution: data experts are temporarily allocated to business/product teams as needed, spending 10-20% of their time in their "home base" for platform development and knowledge-building. Analytics Translators act as facilitators, managing resource allocation and increasing organizational data literacy.
Adaptive topologies
A key concept is the dynamic adaptation of the topology: a data scientist may start in a stream-aligned team (close collaboration), evolve toward a complicated subsystem team (reduced communication), then toward a platform team (as-a-service) as the AI solution progresses through its life cycle. This adaptability allows cognitive load to be managed and communication to be optimized.
Case 2: Large company, advanced AI experience
For mature organizations, the structure evolves toward permanent data experts within product teams, supported by an ML engineering enabling team (training, code reviews, best practices) and a data engineering platform team (pipeline templates, cloud workspaces as-a-service). Communities of Practice replace physical teams for knowledge development.
Governance and evolution
The article emphasizes the importance of data governance maturity: transferring ownership of data preparation pipelines to the product teams that generate the data reduces the complexity and operational load of the central platform team, while simplifying the system architecture in line with Conway's Law.
Practical conclusion
Van den Heuvel stresses that no single design fits every case. Each organization must evaluate its own scenarios using these principles to derive the optimal Data & AI structure for its specific context, size, and AI maturity.