A RadianNa product
SmartQuery
Ask the question. The data answers.
SmartQuery plugs into the database you already have and lets your teams question it in plain language. An AI translates the question, picks the shape of the result, then reads it back to you — in the vocabulary of your field. No SQL to write, no migration, no change to your systems.
Your data already exists. The path to it does not.
A public body or a company that has been running for ten years is sitting on a considerable asset. The problem is almost never the data. It is the distance between the person who has the question and the person who can write it.
You have to know how to write the question
Querying the database means knowing SQL and the shape of the tables. Two skills a director should not need in order to run their business.
Every question becomes a ticket
The request goes to IT, takes its place in the queue, and comes back when it comes back. Technical teams spend their time writing reports instead of building the platform.
The answer arrives after the decision
A figure obtained three weeks later is no longer used to decide. It is used to justify what was already decided without it.
Fixed reports age
You write five hundred queries once and for all, and the five hundred and first question goes unanswered. Every new problem reopens the cycle.
One question, one complete answer
The user writes the question the way they would put it to a colleague. What comes back is not a list of rows: it is the table, the chart that fits, and the analysis of what they show.
“Compare this month's revenue with the same month last year, by branch.”
Revenue by branch
SmartQuery analysis
The North is down 8%% year on year, while the East gains 15%% and the South 26%%. Overall the month is up 7%%. The North is the only branch falling — worth checking before close.
Prepared query
Two more questions, two other shapes
The same chain, with different outputs: the list of cases to look at, and the trend over time. The tool decides the shape, from the question.
“List the operators whose declared value is more than 30% below the median of their category.”
Gaps to the category median
| Operator | Declared | Median | Gap |
|---|---|---|---|
| Operator Y | 85 | 120 | −29 % |
| Operator X | 72 | 120 | −40 % |
| Operator Z | 68 | 120 | −43 % |
SmartQuery analysis
Three operators below the median of their category. X and Z are more than 40% off: the comparison runs inside the category, not against a general average. A check is recommended before clearance.
Generated by the AI
“What is the average processing time per officer this week?”
Average processing time
SmartQuery analysis
The weekly average is 4h20. Tuesday alone falls outside the usual range at 5h30 — one day carries the whole gap, the other four sit within twenty minutes of each other.
Prepared query
“A model let loose on my database will make things up.”
That is the right objection, and it is the one SmartQuery is built to answer. A model discovering your schema at question time is guessing; it does not know that your BOX47_AMT column holds what your teams call revenue. SmartQuery does not let it guess: three layers answer in order, and the model is the last.
- 01
A dictionary of your business
Before any question, your tables and columns are described in plain words, and your teams' vocabulary is translated into calculations. “Revenue” becomes a sum over one precise column, “this month” a date boundary, every business term its filter. That dictionary is the heart of the product — far more than the model, it is what makes the answers right.
- 02
A library of vetted queries
The questions that come back every week are written, tested and frozen once and for all. They answer without going through the model, so they always answer the same way. This library is what carries day-to-day reliability.
- 03
The model as a second line
It only steps in for what the library does not cover yet. Its query is checked before it runs, and the connection to your database stays read-only: SmartQuery reads, never writes.
Two different regimes, and that is deliberate: the model is kept on a short leash when it writes SQL, and left free when it reads results. A wrong query is expensive; a reading you can check costs nothing. Every answer says which of the three layers answered, with its query open to inspection.
Three outputs, every time
The table
The result rows, formatted and readable, exportable in one click for the next meeting.
The chart, picked for you
Bars, lines, pie, treemap, scatter, heatmap, bubbles, outliers: eight families, and thirteen composed views for the frequent cases. The tool picks from the question and the shape of the data — often the choice the user would not have known to make.
The written analysis, by the AI
This is where the model really works. It has been given your field's vocabulary, so it reads the figures in your terms: it names the trend, measures the gap against what is normal for you, and goes as far as the check it recommends. It also says what you did not ask for — which is often where the value is.
Eight generic families. Thirteen business views. And yours.
Twenty-one views are already built. Eight are generic families that suit any data — here they are. The other thirteen were drawn for one specific field: a ranking that knows its categories, a heatmap crossing entities and months, a gaps view comparing each case to the median of its own family rather than to a general average.
Bars
Line
Pie
Treemap
Scatter
Heatmap
Bubbles
Gaps to the median
That is where the difference is made. A business view is not one more chart: it is your way of looking at a problem, written down once and for all. Yours are built at deployment, and the catalogue grows with every client.
What you can ask it
Four shapes of question cover most of what a management team asks. Here is how they are put, in plain language, exactly as written.
Rankings
- “Top 10 suppliers by volume since January.”
- “Who handled the most cases last week?”
- “The five sites that weigh most in the yearly total.”
Comparisons over time
- “Month-by-month trend since the start of the year.”
- “The last thirty days against the thirty before.”
- “Which weekday concentrates the most activity?”
Breakdowns
- “Breakdown of cases by control channel.”
- “What share does each category represent this quarter?”
- “Cross volumes by site and by month.”
Gaps and anomalies
- “Show me the outliers against their own category average.”
- “Which cases went past the processing deadline this quarter?”
- “The amounts falling outside the usual bounds this month.”
Time references are understood as written: this month, last month, the last seven or thirty days, the current quarter, the year — or a date range you set yourself.
How it comes in
- 01
We map your database
Your tables, your columns and your teams' vocabulary are surveyed and written into that dictionary. It is the work that takes the most care, and everything else depends on it.
- 02
We connect read-only
SmartQuery reads, never writes, and every query is checked before it runs. The model runs on your infrastructure or in the cloud, your choice: your data need never leave your premises.
- 03
We prepare your recurring questions
The requests that come back every week become immediate, consistent answers. The rest is translated on the fly by the model.
- 04
Your teams ask their questions
In their own words. No SQL training, no going through a developer, no waiting their turn in a queue.
- 05
Every answer stays checkable
The screen shows whether it came from a prepared query or from the model, and the query stays open to inspection. A figure you can trace beats a figure you have to believe.
What comes next
The product roadmap. These pieces are not shipped yet — they are stated here so you know where SmartQuery is heading before you commit to it.
- Dashboards built from the most frequently asked questions
- Proactive alerts when an anomaly appears, without having to ask
- Connectors to Power BI and Tableau, for the tools you already use
- A REST API, so your other systems can ask their questions too
See SmartQuery on your own data
The demonstration runs on a dataset from your own field, not on a generic example. Write to us and we will find a slot.