Ask your book anything.
Ask in the words you would use with a colleague. Riska reads your own objects and relationships, writes the query and runs it. The rows it read come back with the answer.
How an answer gets made. You ask once. Everything between the question and the number happens where you can watch it.
It reads your model first.
Your objects, attributes and relationships, under the labels your own team gave them. It matches your wording to a real field, never to one that merely sounds close.
It writes the query.
Policy to coverage to location, hop by hop. The filter, the sort and the top ten all come out of the sentence you typed.
Totals are computed, not counted.
Total exposure, average premium on bound policies, how many renew this month. Every figure is calculated across the whole filtered set, and it tells you how many records that was.
The rows come with it.
The result lands inside the answer, placed at the point in the explanation where it belongs. One record renders as a card, several as a table whose rows link straight into the record.
“It answers portfolio questions in a minute that used to mean an export, a pivot table and half a morning.”
The questions you actually have. One box, four different shapes of question, and no report to build first.
Book-wide numbers
Total exposure, average premium on bound policies, how many policies renew this month. Computed server side over every matching record, with the population each number covers quoted beside it.
One account, in depth
Pull the full detail of the records already in front of you. Every field, several records in one pass, returned in the order you asked for them.
Your own queue
Open tasks assigned to you, soonest due first. Or every task sitting on the policies you just pulled up, grouped under the record each one belongs to, with status and priority editable where they sit.
Find the thing
Records, people, companies, tasks, notes and files, from a name or a fragment of one. The top matches across everything you keep here, each one tagged with what kind of thing it is.
Grounded, and accountable. It answers from your records under your access, and it leaves a trail of everything it touched.
The trace is part of the answer
Each lookup names itself as it runs and stays in the thread afterwards, so you can see which query produced a figure before you act on it.
Your access, not its own
The assistant runs as you. It reaches the records your account reaches, and nothing your permissions do not.
Every change is attributed
One row per tool call, and a record the assistant edited shows in the audit log named to the assistant and to the chat it came from.
It can make the change. The follow-up to a question is usually a piece of work. Ask for that too, and approve it before it lands.
It holds on to what you just looked at.
The records a conversation has touched stay addressable, so “those policies” and “the second one” resolve to the rows you were shown rather than to a fresh search that quietly returns a different set.
See the change before it lands.
A record edit arrives as a preview: every field it would set, and anything that failed validation. Approve or reject. Nothing is written until you do, and the card never hides behind a disclosure.
Then the follow-up work.
Open a task on the policy with a due date and an assignee, or move one you already have to done. Each write is reported back from the row as it now stands, not from what was intended.
Everything it can reach.
An assistant is only as good as the model underneath it. Here is the rest of what Riska keeps live for it to read.
Ask it about your own book.
Bring the report you rebuild every Monday. We will ask Riska for it on the call, and show you what it read to get there.
Usually 30 minutes, with an underwriter on the call.