In financial services, a data error doesn't just mean a wrong dashboard. It means an incorrect regulatory filing, a misstated client bill, or a risk team acting on stale numbers. The traceability of these failures makes the cost of bad data infrastructure unusually clear. That context is what makes NASDAQ’s recent platform evolution worth studying.
NASDAQ’s Eclipse Intelligence platform, built on dbt and Databricks, started as an internal solution for their own markets. After a decade of iteration, it’s now offered to external financial market infrastructure (FMI) customers—exchanges, clearinghouses, and depositories. The architecture is a governed data layer covering ingestion, transformation, validation, and business applications like dashboards and billing.
**Scale and consistency**
NASDAQ ingests between 750 billion and a trillion messages daily from 5,000 to 7,000 sources. To maintain consistency, they use a dbt Mesh structure: a family of projects with a foundational layer defining contracts for raw data. Every order chain message looks the same, regardless of origin. This lets them define metrics once and apply them everywhere. The team’s stance on errors is deliberate: “We’d rather break on a test and be late than send wrong data.”
**Removing engineering bottlenecks**
Putting governed SQL tools directly into business teams’ hands reduced time-to-market for new data products by 30-40%. One example: NASDAQ’s options sales team built client-specific visuals without any engineering support, using the same dbt models owned by the business. The key enabler is governance. As Databricks’ Andrea DeSosa put it: “You can only safely put those tools in non-engineering hands when you have a governance layer enforcing the boundaries underneath.”
**Regulatory governance**
Financial services has specific compliance obligations, like Rule 17A (write-once, read-many) and SOC 2 for billing. NASDAQ’s platform provides automatic lineage through Unity Catalog and dbt, so when an auditor asks where a number came from, the answer is in the code, not a manual document. “dbt makes that part pretty effective and pretty easy,” said Michael Weiss, NASDAQ’s AVP of Product.
**AI readiness**
A confident but wrong AI answer in financial markets is a risk event, not an inconvenience. NASDAQ uses dbt’s semantic layer as a “context layer” for AI agents. The difference in text-to-SQL accuracy is stark: near 100% with a semantic layer, versus confident wrong answers without it. NASDAQ is now building an agentic surface layer, aiming for a marketplace of governed, verifiable workflows.
The thread connecting these themes is that the work satisfying a regulator today is the same work making AI trustworthy tomorrow. For NASDAQ’s customers, the option now exists to skip the multi-year build and deploy in six months.
Source: dbt Labs Blog