In a webinar hosted by The Top Voices, data platform leader Maxim Zolotarev explored how data organizations evolve as startups grow. Drawing on practical lessons from Tabby, the discussion focused on why early centralized data teams eventually become bottlenecks, why full Data Mesh decentralization is often harder than expected, and why many growing companies move toward a hybrid Hub and Spokes model.
Speaker
Maxim Zolotarev is Head of Data Platform and Machine Learning at Tabby, where he leads the development of the company’s data infrastructure and analytics platform.
Centralised model in early stage startups
Most startups begin with a centralized data team responsible for analytics pipelines, infrastructure, and reporting across the company. This model provides unified standards, strong governance, and minimal duplication of tools. It works particularly well when the team is small and coordination overhead is low.
As companies grow, however, the central team often becomes a bottleneck. Multiple domains depend on the same backlog, while the data team may lack deep context about each business area. Over time, this structure struggles to scale with organizational growth.
Data Mesh and the limits of full decentralisation
Data Mesh promotes a decentralized approach where domains own their data end to end and treat it as a product with documentation, quality standards, and clear SLAs. In theory, this increases autonomy and scalability.
In practice, however, implementing Data Mesh requires a mature platform, strong data culture, and well-established governance processes. According to Tabby’s experience, moving directly from centralization to full decentralization created significant complexity and coordination overhead.
Hub and Spokes as a practical hybrid model
Many growing companies adopt a hybrid structure known as Hub and Spokes. In this model, a central data platform team maintains shared infrastructure, governance, and standards, while data specialists are embedded within individual business domains.
This structure balances autonomy with consistency. Domain teams gain deeper business context and faster iteration cycles, while the central hub ensures unified tooling and architectural standards across the organization.
Choosing the right model for company growth
The appropriate data organization model depends on company maturity. Early stage startups typically benefit from centralized teams that establish standards and infrastructure quickly. As companies expand and domains become stronger, Hub and Spokes often becomes the natural next step. Fully decentralized Data Mesh architectures tend to work best only in large organizations with mature platforms and strong data culture.
Conclusion
There is no universal data management structure that fits every company. What matters most is aligning the model with the company’s stage, platform maturity, and team capabilities. For many startups the practical path is gradual: begin with a centralized team, evolve toward a Hub and Spokes structure, and consider full decentralization only when the organization and platform are ready.