Finance
Waiting for analysis is what delays the decision
The data is already there. Only a handful of people can get at it, so everyone else waits days.
These are the places you can press
- 1 Three questions. Press a number to jump
- 2 Pause and play
- 3 Switch: conversation, ontology, ledger
- 4 Conversation. Hover to pause while reading
When an action is proposed, approve and hold appear; we will point them out then. Fictional data, nothing runs.
What share of last quarter’s new loans turned delinquent?
Press a number above to jump to that question · the left rail opens the ontology and the ledger
What share of last quarter’s new loans turned delinquent?
2.4% — 28 of 1,183.
| New loans | Delinquent | Rate |
|---|---|---|
| 1,183 | 28 | 2.4% |
Show generated SQL
SELECT COUNT(*) AS loans, SUM(is_delinquent) AS dq FROM loan_master l JOIN delinquency d USING (loan_id) WHERE l.opened_at BETWEEN '2026-04-01' AND '2026-06-30'
Documents and records used
loan_master2026 Q2 · 1,183 rows · IDs masked
delinquency30+ day flag
Questions and answers run on their own. Hover to pause while you read.
Example conversation with sample data. Three questions run in sequence; a person approves or holds the proposed action.
What we hear again and again here
Requests pile onto one team
Business asks, the data team answers. When questions arrive faster than answers, people stop asking the ones that are merely important.
One metric, several numbers
Each team writes its own query. Finding where the definitions diverged takes another meeting.
Evidence has to be gathered again
When a report or an audit asks where a number came from, someone reproduces the calculation by hand.
What we do
Ask internal data in plain language
People who do not write SQL can query directly. People who do get a draft and verify it.
Show the calculation next to the answer
Which table, which column, under which condition — recorded. Not just the number.
Connect read-only
Nothing is written back. Accounts get read access only, and what each person can see is scoped.
What we usually connect
We connect read-only. Nothing is written back, and nothing gets restarted.
Common questions
Does this work under network separation?
Yes. We install on-premise and run without outbound calls. How far the separation goes is checked in the pre-assessment.
Can personal data leak into an answer?
The columns to read are decided when we connect. Data outside a person’s access is never in scope, and every question is logged.
We work out where to start, together
We start by finding what people keep looking for. What to connect, and by when, follows from that.