Google’s wider Conversational Analytics rollout comes with a migration boundary

Google’s new Conversational Analytics experience in Data Studio removes one commercial gate while adding a consequential architecture constraint. The overview says the new experience is available to all Data Studio users, whereas the legacy experience required a Data Studio Pro subscription with Gemini enabled. But the same comparison says the new experience supports BigQuery data agents, not the legacy mix of CSV, Sheets, Looker and BigQuery sources.

That is not an in-place upgrade. Google says legacy conversations and data agents cannot be migrated. Teams moving to the new experience therefore need to rebuild the governed object in BigQuery and treat old conversations as history rather than portable operating context.

The control plane moves to BigQuery

In the new design, authors create, edit and publish agents in BigQuery; Data Studio becomes the consumption surface. The Data Studio agent guide says a published BigQuery agent appears automatically for users with whom it is shared, and that revocation or deletion is also handled from BigQuery.

Access is layered rather than inherited from a shared chat link. Google’s setup guide requires bigquery.jobs.create on the billing project and the BigQuery Data Viewer role on the queried project, dataset or table. It also warns that being granted access to an agent does not necessarily grant access to the underlying Google Cloud project.

For a rollout, that means testing at least three identities separately: the agent owner, an intended consumer with data access, and a user who can discover the agent but lacks one underlying permission. A successful owner demo does not prove that sharing is configured correctly.

Rebuild semantics before opening access

The BigQuery creation guide defines agents as a package of knowledge sources, metadata and query-processing instructions. It recommends structured context such as table and field descriptions, glossary terms and verified queries before relying on broad custom instructions. Verified queries can execute exactly as written when they match a question; parameterized verified queries can reuse an approved SQL template while extracting values such as product, region or date from natural language.

Teams should use the forced migration as a semantic review, not a copy-and-paste exercise. Inventory each legacy agent’s source, instructions, business definitions, example questions and access list. Recreate only the supported BigQuery sources, convert repeated high-value questions into verified or parameterized queries, then validate IAM with non-owner accounts before publishing.

The practical trade is clear: Google is widening access to the chat surface, but narrowing the new agent model around a BigQuery-owned control plane. The rollout is simpler for consumers only after platform teams do the migration and permission work upstream.