Reporting is a safe first place to try AI in your portal: it represents your data without touching it.
Enthusiasm needs framing: an AI's answer about your data still has to be checked, along with the parameters behind it.
Ask an AI how old your deals are and it can answer four years if nothing excludes closed deals; say what you mean, or put the rule where it can read it.
Before you build a report, know which object and which properties it draws on; ask your admin, or Breeze, if you don't.
Define an aging deal as time in a stage beyond what a winning deal takes, not how long it has been open.
Write your rules down, including which custom property replaces a native one, as context Breeze and Claude can both read.
Breeze can build a simple report; check anything custom, and expect trouble with data sets and internal field names as column headers.
A pattern library documents the reports and dashboards you rebuild in every portal, so they can be deployed instead of rebuilt by hand.
Context belongs in a documentation home the whole team can read, not in one assistant's memory.