What it is and where to find it
The Business Analyst lives at "Ask your shop a question" (`/app/ai/analyst`). There's no dedicated sidebar entry for it — and no "AI" sidebar entry of any kind anymore. Reach it by opening Growth & Marketing from the sidebar (under "Run the shop"), then clicking its AI Assistant tab, which loads the AI Tools hub inline (that tab also has an "Open standalone" link straight to `/app/ai`) — there it's grouped in the "Owner intelligence" section as the "Business Analyst" card, marked with a "New" badge, alongside Demand Forecast and Today. Plain Cmd+K opens a different feature ("AI Help," a KB-grounded Q&A popup), not a page-jump search — despite what the hub page's own code comments suggest about mirroring the command palette, and the actual Cmd+Shift+K page-search palette doesn't have this tool registered yet either.
The idea is simple: you ask a plain-English question — the textarea's own placeholder shows "Why was revenue down this period?", and the five suggested prompts below the box (see "How to ask") cover similar ground — and it answers using only a fixed set of read-only, tenant-scoped aggregate metrics from your own shop. The page is explicit that this tool never changes any data; it only reads and explains it.
How to ask
The "Your question" card has a two-line textarea (500-character limit) where you type your question, plus an "Ask" button. You don't have to reach for the mouse: pressing Cmd+Enter (or Ctrl+Enter on Windows) submits the question directly from the textbox. While it's working, the button switches to a spinning "Thinking…" state and is disabled, along with the suggested-prompt buttons below it, so you can't fire off a second question mid-answer.
Below the textbox are five suggested-prompt pills: "Why was revenue down or up this period?", "Which service category brings in the most jobs?", "Who are my best customers by lifetime spend?", "How is my quote close rate looking?", and "What's my average ticket lately?" The one thing to know before you tap one: clicking a suggestion doesn't just drop the text into the box for you to edit — it submits that exact question immediately. If you wanted to tweak the wording first, type your own version instead of tapping the pill.
Reading the Live / Stub badge
In the header you'll see either a "● Live" badge (emerald) or a "● Stub mode" badge (amber), driven by whether an `ANTHROPIC_API_KEY` is configured on the backend. When it's in stub mode, a banner explains the gotcha directly: you'll still get a written answer, but it's a canned stub rather than a model-generated one — and critically, "the numbers shown are always live" regardless of which mode you're in. In other words, stub mode doesn't mean fake data; it means the sentence wrapping the data isn't AI-written yet. Once a real answer comes back, the Answer card's subtitle tells you which you got — "Live model" or "Stub output" — so you're never guessing.
Reading the answer and its sources
Before you've asked anything, the Answer card just prompts you to ask a question or tap a suggestion. Once a question is submitted, you'll see "Reading your numbers…" with a spinner, and if something goes wrong you'll get a plain "Couldn't answer that right now. Try again." message rather than an error dump. A successful answer appears as a written paragraph, followed by a "Based only on" box — this is the tool's transparency mechanism, listing the specific metrics the answer drew from (each as its own bullet). That list exists precisely so you can double-check the answer wasn't built on anything beyond your shop's real revenue, jobs, quotes, and customer figures — there's no hidden context or invented numbers behind it.
Because everything here is read-only, there's no risk in experimenting: ask a vague question, ask a pointed one, tap a suggestion, compare the "Based only on" list each time. If the metrics listed don't match what you were actually asking about, that's your signal to rephrase the question rather than trust the prose on faith.