Questions across your data

Everything below is something you say to your AI assistant once it's connected to mente. They're grouped by shape rather than by topic — because once you've seen one question of a shape answered, every other question of that shape works the same way.

The examples assume a book of business: a policies table linked to a clients table, with notes, tasks and meetings attached to both. Substitute your own nouns; nothing here is insurance-specific.

Before you start

Ask your assistant to "list what's in this workspace" first. It reads your collections, their columns, and the links between them, so the questions below land on your real field names instead of guessed ones.

Filter one table

Which policies renew in the next 30 days?

How many active policies do we have per carrier?

The second one is a count per group — you'll get one row per carrier, not one row per policy. mente counts the groups, so the total beside the answer is the number of carriers.

Reach across a link

These are the ones that used to mean two exports and a spreadsheet.

Which policies belong to a client in the enterprise segment?

Segment lives on the client, not the policy. mente follows the link and filters there.

Show me every policy with its client's name and account manager.

How many policies do we have per client segment?

Which clients hold more than five active policies?

Rank our clients by total premium.

The last two filter and sort on a number computed across the link. You never see the intermediate list; you get the ranked answer.

Ask about what isn't there

Which clients haven't had a meeting in the last 60 days?

Which policies have no account manager assigned?

That second one has a subtlety worth knowing about: mente distinguishes "the account-manager field was never filled in" from "it's filled in with nothing." If you care about the difference, say which one you mean and you'll get it.

Follow the trail across primitives

Which tasks are blocking a policy that renews next month?

For our enterprise policies, what tasks came out of meetings about them?

What's the status of every task created from last week's meetings?

Show me every note mentioning Acme Corp, plus the tasks attached to those notes.

Links read from either end. "The tasks from this meeting" and "the meeting this task came from" are the same link asked in two directions, and both work.

Mix relevance with structure

Find notes about renewals for Acme Corp that were updated this quarter.

Which clients have notes complaining about pricing?

The first ranks by how well each note matches "renewals" while still filtering to one client and one date range. The second uses relevance as a test — clients where such a note exists — and groups the answer by client.

Find anything — a note, a record, a meeting, a task — about the Alvarez cancellation.

That last one is a search rather than a query: it looks across everything indexed and ranks by relevance. Ask it when you don't know where the answer lives.

Two dimensions at once

Total premium by carrier by month, for the last 12 months.

You'll get one row per carrier, each carrying its own month-by-month breakdown — and each carrier's subtotal is the sum of its own months, because both numbers come from the same pass over the same rows. They cannot disagree.

Where to stop

Three shapes are worth knowing you can't get in one pass, so you don't spend an afternoon rephrasing:

• Arithmetic between two answers. "What share of the book renewed on time?" → ask for renewals due and renewals on time, at the same grain, and divide. mente will not invent a computed column, because a computed column is where a wrong number hides.
• A record versus its own past. "Which policies had their premium rise more than 20% at renewal?" → mente stores what a policy is, not a version history. Keep the prior premium as its own column and it becomes an ordinary filter.
• Walking into a connected app. "Which of our clients also exist in the CRM?" → connectors are readable, but they are the end of a trail rather than a step along one. Pull the CRM list in first, then compare.

Related

Asking questions for the query shape itself, structured data for the tables, and build a book of business for the workspace these examples assume.