How it works
From a question to a number you can defend
Four steps sit between a question and the answer on screen. Every one of them exists so the number can be defended in the room, not just delivered to it.
Ask a question naturally
There is no query language, no report builder and no request form. Someone types the question they would otherwise have put in an email to the data team, in their own words, using the names their business actually uses for things.
Follow-ups keep their context, so narrowing to one region or one quarter is another sentence rather than starting again. When a question could reasonably mean two things, it asks which you meant instead of picking one.
It resolves scope, access and a plan
The question is interpreted against the definitions held in Roots — what your metrics mean, which entities relate to which, and what your teams call things — so it is answering your version of the question rather than a generic one.
Access is applied here, before any planning. The scope is reduced to precisely what that person is cleared for, and everything else is invisible rather than merely excluded. A question cannot plan against data the asker could not see.
Only then is a retrieval plan produced: which sources are needed, how they join, and what has to be computed.
It runs a real, checked query
The plan becomes SQL, and that SQL is validated before it is allowed to run: read-only, inside the approved scope, and structurally sound. There is no configuration in which a question becomes a change to your data.
Execution is deterministic. The same question against the same data returns the same rows, because the figures are retrieved rather than generated. This is the step that separates the answer from the model's opinion of the answer.
The query, the source, the row count and the timestamp are all recorded as it runs — not reconstructed afterwards.
It consolidates the result into an answer
The returned rows are turned into something a person can act on: the finding stated plainly, the arithmetic behind anything derived, and what it points to. The interface assembles around it — a trend becomes a chart, a ranking becomes a sortable table.
Before it is shown, a second model compares every figure in the draft against the actual results. Values that reconcile are marked verified; anything that does not is flagged rather than published.
If the data cannot support a conclusion, you get a plain unknown and an explanation of what is missing. That is a better outcome than a fluent answer that happens to be wrong.
What working with us involves
The platform is finished. What remains is connecting your data and describing it correctly, and that is done with you rather than handed to you.
Operational on day one
Everything that normally makes an analytics rollout a project is already built: the governed query path, the permissions model, the verification pass, the interface, the audit trail and the admin tooling. What remains is configuration, and configuration is measured in days.
The practical test of turnkey is whether value arrives before the invoice does. Here it does, because nothing about the first question depends on work your team has to finish first.
A dedicated account team behind you
You are assigned people, not a ticket queue. Your account team handles onboarding, gets your data connected and described correctly, trains your users, and stays with you afterwards as your questions change and your business does too.
When something needs attention you are talking to someone who already knows your setup. Larger customers eventually want that control in-house, which is what Studio is for.
A new domain is an onboarding, not a rebuild
The same platform serves a finance team, a logistics operator and a hospital network because domain knowledge lives in configuration rather than in code. Adding a vertical does not wait on a product release.
That is also true inside your own company: the team that adopts it second does not repeat the work of the team that adopted it first.
The fastest way to judge this is to watch it.
Bring a question your team cannot answer today. That is a better demo than anything we could script.