Decentralize Ownership, Centralize Control: The Case for Federated Data Architecture

For most of the past decade, capital markets firms tried to fix their data the same way: pull everything into one place. A central data lake, owned by a central team, would finally give the enterprise a single source of truth. Most firms now know how that tends to end. Without governance, a data lake quietly turns into a data swamp, an inaccessible store where data quality and lineage can no longer be trusted. The industry named the failure mode precisely because it became so common. The central team turned into a bottleneck for data it did not fully understand, while the business domains that did understand it waited in the queue.

Data mesh is the industry’s answer, and by now a mature one. The idea is straightforward: treat each domain’s data as a product and give ownership of it to the domain that produces it, across equities, fixed income, FX, risk, and treasury, rather than to a single central team. Domains move faster because they no longer wait on a central pipeline.

What is driving the shift

Three forces are accelerating data mesh adoption in finance.

1. Platform proliferation. Data now lives across multiple data platforms, cloud warehouses, vendor terminals, and proprietary systems at once, and no single environment holds the true picture.
2. The limits of centralized teams. A central data team cannot scale to the needs of every domain at once, which is the bottleneck data mesh sets out to remove.
3. Regulatory pressure. Rules such as BCBS 239 and DORA keep expectations on data governance and oversight high, and those expectations apply to the firm as a whole, however its data is organized.

What data mesh gets right

On its own terms, data mesh gets a great deal right. Domain ownership puts data quality in the hands of the people who actually understand the data. Autonomy shortens iteration cycles. Treating data as a product raises the standard of documentation and discoverability, so a data product arrives with context rather than as a raw drop into a lake.

What breaks without guardrails

What data mesh does not do on its own is hold the firm together. Decentralizing ownership is not the same as decentralizing accountability, and regulators still address the firm, not the domain. Left without a center, a federated estate fragments in predictable ways. Access policies drift apart as each domain sets its own. The same data product is licensed and rebuilt in three places, with no one positioned to see the duplication. Compliance gaps open between domains that each assumed the other was covering the requirement.

A data lake without governance quietly turns into a data swamp. A data mesh without a center does the same thing faster, across a dozen domains at once.

The four capabilities best kept central

This is why decentralization in financial services needs a center. Not the old central team that owned every pipeline, but a lighter core that holds the connective tissue while domains keep ownership of their data products. Four capabilities belong at that center.

Stays centralThe connective tissue	Stays with the domainDomain autonomy
Entitlements: who can access which data product, enforced consistently across the estate rather than reinvented per domain	Ownership of the data product and its data quality
Licensing governance: vendor terms and AI usage rights tracked across the estate, so the same license is not duplicated or breached	Domain semantics and business context
Cross-platform discovery: one way to find data products wherever they physically live, without knowing the platform	Pipelines and transformations inside the domain
Usage analytics: an enterprise view of who uses what, how often, and at what cost, which no single domain can see alone	Roadmap and iteration speed for its own consumers

This division is the core of federated governance, the fourth principle of data mesh that most implementations underinvest in. Global rules for security, privacy, licensing, and regulatory standards are defined once at the center and enforced automatically and locally, so governance scales without a committee approving every change. Domains keep their autonomy. The firm keeps enterprise-wide control.

The business data catalog as connective tissue

That connective tissue has a name: the business data catalog. It sits above the domains and the platforms as the one governed layer where data products are discovered, entitlements are enforced, licenses are tracked, and usage is measured. It is the practical difference between a federated architecture and a fragmented one.

How DataHex Data Library delivers this

DataHex Data Library is built to be that catalog for capital markets. It operates as an AI-native metadata layer over a firm’s existing platforms, with no data migration, so domains keep ownership of their data products while entitlements, licensing, cross-platform discovery, and usage analytics are managed in one place. Access policies are enforced programmatically and inherited by both analysts and the AI agents working alongside them. Licensing and AI usage rights are verified before access. Demand intelligence from failed searches surfaces what domains are missing. The result is the institutional data knowledge a federated firm needs to move quickly without losing control, which is where research velocity and controlled agility meet.