What it does

Analyst is a data-reasoning agent that turns raw data — spreadsheets, exports, tables you give it — into worked analysis. It reasons in steps and can run Python to model the data, showing its working.

Key facts

• Built on a reasoning model tuned for data work; it plans an analysis rather than answering blind.
• Runs Python behind the scenes to do the actual number-crunching, and you can inspect the code.
• Aimed at questions like "what's driving this trend" across data you supply.
• Part of the same agent line-up as Researcher, launched through early access first.

When to use / skip

Use it when you have data but not the time or the analyst to interrogate it properly. Skip it for a simple sum or chart — Copilot in Excel is quicker and lighter for that.

Configuration decisions

• Whether the Python execution path is permitted, and the data-egress implications.
• Access scope during any early-access phase.

Gotchas

• It'll happily analyse dirty data and give you a confident, wrong conclusion. Garbage in, garbage out still holds.
• The Python is inspectable but most business users won't read it — so the "show your working" reassurance is partly theoretical.
• Big or awkwardly-shaped datasets can trip it up; check the assumptions it made about your columns.

Consultant notes

• This lands best with analysts who can sanity-check the method, not as a replacement for them. Pitch accordingly.
• The Python execution is a real governance point — know where the data goes before you enable it for regulated teams.
• Like Researcher, availability has shifted quickly. Confirm what's live in the tenant rather than trusting the launch blog.

---

Worth revisiting after the next release wave.