Insights are Cortex analyses that appear on records, reports, dashboards and admin surfaces. They follow one firm rule: an insight surface never auto-runs. Nothing is computed, and no credits are spent, until you type a request and press Execute.
Why insights never auto-run
- Predictable cost: opening a page never triggers model usage.
- Intent: the analysis answers the question you asked, not a generic guess.
- Privacy: data is only sent to the model when you choose to ask.
Run an insight
- 1Open the record, report or dashboard that has an insight panel.
- 2Type your question in the panel, for example: Which deals stalled after the proposal stage this quarter, and why?
- 3Optionally adjust effort. Higher effort thinks longer and costs more within your budget.
- 4Press Execute. The panel shows Cortex working, then the result.
- 5Ask a follow-up in the same panel to refine, or clear it to start over.
Managing results
- Minimize a result to keep it out of the way without losing it.
- Results auto-collapse after about 15 seconds so long analyses do not crowd the page.
- Maximize to read a long result in a larger view.
Effort and budget
Cortex runs on one unified engine across the header chat, the admin console and content generation. Each request combines an effort level (how deeply the model reasons) with a budget (the ceiling on what the request may consume). Low effort suits quick lookups; high effort suits multi-step analysis.
Insights only reason over data your role can already see, including field-level restrictions. If a number looks incomplete, check your permissions before assuming the analysis is wrong.