Approve what a question will compute, before it computes anything
Rover AI first shows, in plain language, what it understood: the tables it will read, the period it will cover and the filters it will apply. You approve it, or correct it in your own words, and only then does the query run.
A plain-language account of what will be computed
Before a query runs, the card states what the answer will measure, which tables it draws on, the period covered and every filter applied. It is written for the person who asked the question, not for the person who maintains the database.
Approved once, wherever the card is opened
A card raised on the web may be approved from the mobile app, and the reverse. It runs once, however many devices have it open. When a card is waiting and no one is watching, a notification says so.
- 1One approval, one execution, across every device
- 2A notification when a card awaits a decision
- 3The same card, in the same words, on desktop and mobile
From question to approved answer
Four steps, with the decision to run resting with a person.
Ask
The question is put in plain Thai or English, as it would be put to a colleague.
Read the card
The card states what will be measured, from which tables, over which period and under which filters, together with any concern found in advance.
Approve or correct
Approve the card as it stands, or describe what should differ and receive a new card.
Receive the answer
The approved query runs, read-only, and the answer arrives with its chart, its analysis and a mark showing it was reviewed.
What the card does when the matter is not simple
The cases in which a quiet adjustment would be easiest are the cases in which the platform asks again.
A new card when the meaning changes
If an approved query must be rewritten in a way that alters what it measures, it is not run. A new card is shown for approval, with the difference stated.
Broad questions brought to a readable level
When a question would return more detail than can be read, the result is summarized at a coarser level and the card says that it has done so, and how to obtain the detail.
Several parts, separated
A question that asks for two different things is separated into its parts and offered back as individual questions, rather than merged into one misleading chart.
Empty columns noted in advance
When a filter rests on a field that holds no data, the card says so before the query is approved, not after an empty answer has been returned.
Conditions in words
Common date and filter conditions are phrased as a reader would say them, such as "from the start of last year", with the exact condition one click away.
Query visibility by role
Administrators decide which roles may read the underlying query. Those who may not still approve by the plain-language card.
Questions about this feature
Does review slow every question down?
Only where the organization chooses it. Review may be left to each user, required for all, or switched off. Many organizations require it for a period after launch and relax it once confidence is established.
Must the reviewer understand SQL?
No. The card describes the tables, the period and the filters in plain language. The query is available to those who wish to read it, and administrators may hide it from roles that have no need of it.
What happens if the approved query fails?
If it can be repaired without changing what it measures, the repaired query runs and the answer says that it was adjusted. If the repair would change the meaning, a new card is shown for approval instead.
Can a required review be bypassed?
When the organization requires review, the requirement is enforced by the service for every request, whichever application or device sends it. It does not depend on the screen in front of the user.
Can a card be approved from a phone?
Yes. A card raised on the web may be approved in the mobile app and the reverse, and it will run once only. A notification is sent when a card is waiting and the user is not present.