DoorDash brings merchant analytics to thousands of users in weeks with dbt and ThoughtSpot

Governance and speed, after several acquisitions

In the conversation, ThoughtSpot's Sonny Rivera interviewed Harsha Reddy, a senior engineering leader at DoorDash who leads Data End-User Platforms within the company's data platform organization. Reddy described DoorDash's core data strategy as a balance between two goals that often pull against each other: central governance and user velocity.

The acquisitions of Deliveroo (UK) and Wolt (Finland, and much of the EU) made that balance harder. Each company arrived with its own semantic layers, often built in Looker, which then sat on top of DoorDash's own in-house semantic tooling. According to the source, consolidating them into one governed layer without slowing teams down became the real problem to solve.

Separating the semantic layer from consumption

DoorDash's answer is to keep the semantic layer apart from the tools people use to consume data. Data sits in a data lake, exposed as Iceberg tables, and in Snowflake. dbt defines the semantic and metric layer on top of both. ThoughtSpot is one consumption layer among others and connects directly to dbt's metric definitions, so it does not need a parallel set of its own.

Reddy contrasted this with an all-in-one tool like Looker, where "there's just one tool where it's got everything encoded, and it's not so much useful outside for anyone else."

The setup serves two different use cases:

• Internal finance and revenue reporting: Reddy said "the margin for error is zero." The governed semantic layer, combined with ThoughtSpot's built-in compliance and audit reporting, replaced work DoorDash used to do by hand to meet GDPR, HIPAA and other regulatory requirements.
• Merchant-facing analytics: DoorDash embeds ThoughtSpot, fully white-labeled with no ThoughtSpot branding visible, directly into merchant tools. Role-level access control ensures one merchant never sees another's data.

Reddy declined to give an exact user figure, but said the team reached "thousands of users within a few weeks" from a new use case.

Advice on semantic modeling and AI agents

Asked what he would tell other teams, Reddy said semantic modeling is hard, but worth the effort in a world where generating an answer is cheap and trusting it is what carries value. He recommended treating the semantic layer as code, meaning version-controlled and reviewed. He also said the governed path should be at least as easy as writing an ad hoc query, or people will route around it.

On AI, DoorDash has moved from a set of domain-specific agents behind a router to a single global agent. That agent is grounded in the same dbt-defined semantic layer that feeds the company's BI tools. According to the source, the model now covers roughly 70 to 80 percent of DoorDash's data domains.

What it means for data teams

The account is a company-reported case study published by dbt Labs, and it comes without detailed metrics beyond the user-growth claim. Still, it shows one pattern for organizations that have inherited multiple semantic layers: define metrics once, govern them as code, and let several tools, including BI interfaces and AI agents, draw on the same definitions. For teams weighing similar projects, Reddy's central point is that trust in the answer, not the cost of producing it, is the scarce resource.