Give the CEO a reviewable answer in minutes.

"Why did ad revenue growth slow in March?"
What the agent must prove
Metric definition comes from dbt docs.
SQL runs locally against DuckDB in read-only mode.
Every claim cites a source query and caveat.

First, Boardroom Analyst checks whether the datamart is safe to reason over.

dbt projectDiscovery martvisual_discovery_boardroom_mart loaded
model grainmonth × surfacedocumented in metadata
warnings0ready for executive analysis

The agent plans claims against query IDs before writing the narrative.

q001_surface_revenue

select month, surface, net_revenue_millions
from ad_revenue_by_surface
order by month, surface
q002_revenue_growth_waterfall

with totals as (...)
select month, total_revenue_millions,
       revenue_change_millions
from changes
order by month

The evidence shows shopping and visual search grew while brand video dragged.

Jan
Feb
Mar

The output is executive-readable and audit-ready.

Total ad revenue increased from $925M to $973M, but monthly growth slowed from +$69M to +$48M. source q002_revenue_growth_waterfall
Brand Video Ads declined by $14M while Shopping and Visual Search added $62M combined. source q001_surface_revenue
Includes SQL appendix, result hashes, chart CSVs, and caveats.

Follow-up questions reuse the same governed context.

"Which surface should I ask the ads team about first?"
Brand Video Ads: -$14M in March, while shopping-intent surfaces kept compounding.

The answer stays tied to `q001_surface_revenue`, the documented grain, and the same caveats.