What can you ask the analyst?
Anything the analytics engine can compute. Which products lost money after returns last month. How many days of cover the top twenty lines have. Which store converted footfall best on Saturday. What the realised discount was on a code. The analyst turns the question into tool calls, receives computed figures, and writes the answer around them.
The answer leads with the conclusion, then the evidence. It streams live, so a long question does not mean a blank screen, and it comes with source chips for each figure. The analyst does not speculate about numbers it cannot compute; it says the input is missing and which integration would provide it.
The trust footer is the second law made visible
Under every answer is a trust footer listing exactly which tools were called and how fresh each system's data is. If the answer cites net sales, the footer shows the net-sales tool and the time of the last store sync. If it cites footfall, it shows the people-counting source and its last run. A reader can check the answer's foundation without leaving the screen.
This is how the commerce engine proves that the AI interprets and never calculates. There is no answer without a footer, and a number without a tool behind it cannot appear, because the agent has no way to produce one.
Why is it both a workspace and a side panel?
As a full workspace, the analyst is where longer investigations happen, with projects and pinned answers so a line of questioning about a category or a store can be kept and returned to. As a side panel, it opens on any page and already knows the page, the tab, the date range and the record on screen, so a question can be as short as why did this move.
Every KPI card carries an Explain-this marker. Clicking it asks the analyst to independently recompute the figure through the same tools and explain the result, which is how an owner checks a number that looks wrong without opening a spreadsheet.
What do select, shortcuts, pins and the watchlist add?
Select any text on a page, in a brief, a report or a card, and ask about it. Slash shortcuts open the common questions in one keystroke. Answers can be pinned into a project so a week of questions about a launch stays together. And the watchlist lets you tell the analyst to watch a figure and say when it moves, which turns a question into a standing alert.
Where an answer implies an action, the follow-ups are one click: draft the email to the store manager, open the inventory list this figure came from, put the code on the discount monitor's watch. Every one of those actions that leaves the platform still goes through Review & Send.
Voice is the same analyst, spoken
Voice mode is a live spoken conversation with the same analyst, backed by the same tools and the same trust footer, with barge-in so a question can be interrupted and redirected. It is built for the owner walking the shop floor or the buyer at a supplier, not for a demo.
The rules do not change when the interface is a microphone. The analyst still cannot produce a number without a tool, still says when a source is stale, and still routes any action through Review & Send on a screen.
How does the analyst fit the Digital FTEs on ftes.ai?
The chat analyst is how a manager interrogates the work the roles have done. A proposal from the Buyer FTE can be questioned in the side panel from the Approvals tab; a figure on the Chief of Staff FTE's morning page can be recomputed with Explain this. The analyst is not a role; it is the way to talk to the engine every role works on.
The next step for a brand that wants it is to connect the store, open the side panel on the Dashboard, and ask the first question. The trust footer on that first answer is the fastest way to see how the engine keeps its promises.