Agentic BI: give it a goal, not a query.
Traditional BI answers the question you already know to ask. An agentic approach starts from the goal — “Why did revenue dip last month?” — and works out which questions to ask. SchemaChat's Analyze mode does exactly that on your own data, and keeps every step visible.
How it works
- 1
It proposes a plan
A short list of steps — typically a few — that you can edit before anything runs.
- 2
It runs each step as a read-only query
Each step is a real query against your data, shown with its SQL.
- 3
It checks each result, then writes it up
Every step is reviewed: does it answer the sub-question, is the scope right, are the cited numbers really in the data? The summary cites its steps and ends with a “Not covered” line listing anything it couldn't answer.
Where agentic BI helps, and where it doesn't
It helps with open questions and root-cause hunts: “why did margin fall?”, “what changed since September?”, “where are returns coming from?”. It saves the back-and-forth of working out which cuts to look at.
It does not replace governed executive reporting, it doesn't know your business definitions unless you give it metrics or a glossary, and it can be wrong — which is why every step shows its query and the summary says what wasn't covered.
How it differs from a dashboard
A dashboard shows what you already decided to track. An agent can investigate what you haven't. Most teams end up using both: dashboards for the numbers you're accountable for, questions for everything else — and anything you ask three weeks running should become a dashboard or an alert.
Control and safety
- Read-only access to your data, enforced in code
- An editable plan: you decide what gets asked
- A per-step check and a “Read with care” flag when a comparison rests on very small groups
- A workspace audit log, roles, and optional required two-factor sign-in
Honest limits
- It is not unsupervised: review the plan and read the “Not covered” line.
- It can't answer what your data doesn't contain, and it will say so.
- It does not forecast the future or make decisions for you.
- There is no SSO or SOC 2 certification yet; see the security page.
We would rather tell you now. Read the security page for what we do and don't have.
Frequently asked questions
- What does “agentic” mean here?
- You give a goal rather than a single question. SchemaChat plans several steps, runs them as queries, checks each one and writes a summary that cites its steps.
- How many queries does one analysis run?
- The plan shows the steps before it runs — typically a few. You can edit or remove steps first.
- Can it change my data or take actions?
- No. It only reads your data.
- How do you reduce wrong answers?
- Each step is checked, every answer shows its SQL, and releases are tested against 115 questions with independently computed answers. None of that makes it infallible, so check the numbers that matter.