Platform · Data analysis
Text-to-SQL is easy to demo and hard to trust. The gap is everything around the query: what the columns mean, whose rows the asker is allowed to see, and whether the number can be reproduced next quarter.
Top 3 lanes by margin loss this quarter
select lane, sum(margin_delta) as loss
from fct_margin_daily
where quarter = '2026Q3'
and region in (/* your data role */)
group by lane order by loss limit 3The highlighted clause is added by the platform, not the model. Row rules follow the person asking.
| lane | loss | share |
|---|---|---|
| CTU-RTM | -2.1 pts | 62% |
| SHA-LAX | -0.8 pts | 24% |
| SZX-HAM | -0.5 pts | 14% |
The clause you didn't write
Asking a model to respect permissions is not a control. In DataFact the scoping clause is added by the platform after the query is generated, from the data role of whoever is asking.
That is the difference between a demo you can show and a system you can roll out to four hundred people who each see a different slice.
Capabilities
Speed is table stakes. These are the parts that decide whether the number survives contact with the people who own it.
Semantics
A column called amt_net_3 means nothing to anyone. Table and column descriptions, metric definitions and the rules your analysts apply by hand are held in an analysis model that every question is resolved against.
Queries
Generated SQL is visible next to the result. Execution is read-only with a per-request timeout and cancellation, so an expensive question cannot sit on your warehouse indefinitely.
Permissions
Access is not a prompt instruction. Table-level and row-level rules bound to data roles are applied server-side when the query runs, so the same question returns the correct, different scope to a regional manager and to a group controller.
Versions
Every build of the analysis model freezes a snapshot. If a metric definition changed in March, you can see what it was, compare the two, and roll back — instead of arguing about which report was right.
Beyond a single number
The most valuable questions are not answerable from the warehouse alone. DataFact can hold a figure from a report and a row from a table in the same reasoning step.
Margin moved — reconcile the warehouse trend against what the quarterly review PDF attributed it to.
Supplier performance by lane, checked against the penalty terms written into the signed contract.
Which transactions fall outside the threshold defined in the current policy document, not last year's.
Give us one schema and the questions your analysts get asked every week. We will show you the queries it generates and where the row rules land.
SaaS subscription or private deployment · No data migration either way