How Data Assistant is built.
Ask the AI Analytics Dashboard a question. The Data Assistant knows which tab you’re on, answers from the dashboard’s own figures, and never makes up a number.
Data Assistant — a streamed tool loop over the dashboard’s data
The assistant lives in a panel on every tab of the AI Analytics Dashboard. Each question goes out with a snapshot of what’s on screen and the monthly figures the page has already loaded. The model answers straight from those when it can. When it can’t, it calls one of six narrow tools over the dashboard’s own data files, and the answer streams back as it’s written.
Hover or tap an agent above to see its pipeline and what it returns for this bid.
◆Data Assistant— Page-aware analyst
A Claude Sonnet 5 tool-use loop running in an Azure Function, with secrets held in Key Vault and read through a managed identity. It knows who its users are (MillerKnoll’s leadership team) and the org it reports on: budget owners, the Technology Leadership Team, the fiscal year that starts in June, and the rule that M365 Copilot is only ever grouped by department. Every figure it reports has to come from the page, the loaded data, or a tool call. It can do arithmetic on those figures and show its inputs, but it cannot estimate.
- Source[Page on screen] · every questionKPI tiles, tables and notes from the tab you’re looking at
- Source[Dashboard data] · every questionmonthly totals already loaded, including the dashboard’s own forecast
- ToolSix data tools · only when the page doesn’t cover itspend · budgets · people · initiatives · adoption · data catalog
- KnowledgeOrg rules · alwaysbudget owners, TLT, fiscal year, no invented numbers
Why it matters · Leaders get an answer in the terms of the page they’re already looking at, and every number can be traced back to that page.