The assistant now shows its reasoning, and offers the next question

Until now an assistant answer stopped at the data: a figure, the definition behind it, and the records that add up to it. That is still there — but an answer that only hands you a number leaves you to work out what it means and what to ask next.

What the assistant reads into the number

Below the figure, the assistant now sets out its reading as a short numbered argument: what it concludes, the counting rule that could be misread, and — deliberately — the assumption you should attack first. It is kept visually apart from the data, because the numbers are what they are and the reading is an opinion you should be able to argue with.

A vendor asking about the funnel is told which stage most of a cohort stops at, and reminded that a younger cohort has had less time to convert than the one it is being compared with. A CSM asking what is blocked is told where the blockage concentrates, and that the grouping itself is the thing most likely to be misleading.

Follow-up questions, one click away

Under the reading sits a row of follow-up questions — one obvious next step and a few alternatives. Clicking one asks it: the button’s words are sent as your next message, exactly as if you had typed them. Nothing else happens, nothing is changed, and every question offered is one the assistant can actually answer.

Several of them exist to test the reading rather than confirm it. “Show me the same funnel over the last 12 months” is how you check whether a trend is real or an artefact of the window.

Status you can see at a glance

Where a table reports a status — an SLA band, a coverage state — it now renders as a coloured pill rather than a word among words, so the row that is costing you most is the one you see first. Red is used only for the problem.

A figure that comes from the metric dictionary also shows what kind of figure it is, in the dictionary’s own words: a cohort followed forward and a stock read at this moment no longer look identical.

For service delivery managers

Asking how bids split across the price bands now returns the split as bars, each carrying that band’s win rate and how much of the won work landed on time — so you can tell a pool nobody can serve from a pool nobody will bid on at that price. Where a band has not won anything yet, it says so rather than reporting nothing as nought per cent.