Negotiate the Looker license and size the data token pool in the same conversation with sales. The license sets the floor. Conversational Analytics draws on a per-instance token pool that runs free of quota limits only until Google ends the current promotion, and every answer also runs a warehouse query.
When you need this
Evaluation criteria
Edition
Which platform edition fits?
Look forStandard for under 50 users with no private connections, VPC-SC, or CMEK; Enterprise for internal BI with those controls; Embed for external analytics with signed embedding. Each includes 10 Standard and 2 Developer users and one production instance.
User licenses
Who needs which license?
Look forDeveloper users for LookML and administration, Standard users for Explores and building dashboards, and Viewer users for dashboards and scheduling only.
Data token allocation
How far does the instance's monthly pool go?
Look for60M input and 1.2M output tokens on Standard, 300M and 6M on Enterprise, and 1.2B and 24M on Embed. Non-production instances get 6M and 0.12M. All users share the pool, and it resets monthly. At one pilot team's reported 250,000 input tokens per question, Standard's pool covers about 240 questions a month.
Overage terms
What happens after the promotion?
Look forAt least 90 days' notice before billing and quota enforcement, then overage of US$3.00 per million input and US$20.00 per million output tokens, billed monthly. Ask sales whether your order form fixes different terms.
Visibility
Can you see who and what consumes tokens?
Look forThe Token usage tab of the Conversational Analytics dashboard in System Activity, which shows estimated tokens by surface and by day. It is generally available on Looker (original) 26.16 and later and a pre-GA feature an admin enables on Looker (Google Cloud core). Tracking starts only once enabled and refreshes once a day, so turn it on before the pilot.
Warehouse compute
Who pays for the queries?
Look forA warehouse budget that counts dashboard tiles, Explores, and agent queries, with the slowest derived tables persisted. One team converted its Tableau dashboards to LookML and heard from IT that they consumed too much compute.
Questions to ask vendors
- What does our order form say about data token quotas and overage?
- How many people need Standard licenses, and how many only read and schedule content?
- Which Explores send the heaviest queries, and are they cached or persisted?
- Do our non-production instances need their own token allocation for agent testing?
Red flags
- Rolling out agents company-wide before measuring token use per surface.
- Treating the promotional period as permanent.
- Buying Standard edition and then needing private connections or CMEK.
- No owner for warehouse cost generated by Looker queries.
Products in this piece
Sources
- Official documentation
- 4
- Public source
- 5
- Looker pricing
- Looker docs: Looker (Google Cloud core) features and entitlements
- Looker docs: Conversational Analytics in Looker overview
- Looker release notes
- Reddit post on dashboards converted to LookML consuming too much compute (Aug 11, 2026)
- G2 review from a BI architect on self-serve and price (Jun 27, 2025)
- Reddit comment on Looker's license cost next to Power BI (Apr 14, 2026)
- Google Developer forums: Conversational analytics / gemini token usage data (ruthc and others, Sep 14, 2026)
- Google Developer forums: Introducing Data Token Observability for Conversational Analytics in Looker (Tarunima, Aug 18, 2026)