Name the authoritative tables
Revenue, expense, and pipeline data often live in three overlapping systems. We pick the table the CFO will defend in a board meeting and label everything else as reference material.
Each table gets an owner, refresh cadence, and validation script that runs before the AI agent can read from it.
Expose assumptions to the business
Every prompt we use for forecasting calls the assumptions directly: currency, time horizon, segment definitions, and any exclusions.
Those assumptions render in the UI so finance can challenge them before the forecast lands in the Monday executive thread.
- →Use the same naming conventions across Snowflake, Notion briefs, and Looker dashboards.
- →Attach data-quality badges (green, yellow, red) so stakeholders know when to treat the output as directional only.
Keep the contract alive
Contracts fail quietly when teams archive them. We calendar quarterly reviews that include finance, RevOps, and the builders maintaining the agents.
When a schema change ships, the contract template forces us to list downstream tools and alerts the AI platform to re-ingest documentation.