Chat with your data
Conversational analytics becomes the front door to BI. You ask in plain language; the system answers from governed definitions, not guesses.
Why did EU margin drop last quarter, and where was it worst?
EU margin fell 2.1 points to 41.2% last quarter. The drop was steepest in EU North, where freight costs rose while average selling price held flat. EU South held roughly even.
Every answer shows the definition it used and the query it ran. An answer you cannot trace is not an answer.
For twenty years, the front door to BI was a dashboard, and a narrow one. If the view you needed didn’t exist, you joined a backlog.
The new front door is a sentence. You ask the way you’d ask a colleague, and the answer comes back grounded in definitions your business already agreed on. The dashboard becomes one thing the conversation can produce, not the only way in.
Why now
Two shifts converged:
- Language models got good enough to sit in front of real business questions, not just demos. People don’t want to learn a query language; they want to ask.
- Grounding made it trustworthy. Pointed at a raw schema, text-to-SQL is right maybe 17 to 40% of the time. Grounded in a semantic layer, accuracy climbs to roughly 85 to 95%. That gap is the whole story.
What it looks like
A finance lead asks: “why did EU margin drop last quarter?”
A naive tool guesses at the table, picks a date column, and hopes. A grounded system resolves “margin” to the governed definition, scopes “EU” to the agreed regions, compares quarters, and answers: freight costs rose in two markets while selling price held flat. It shows the query and definitions alongside, so anyone can check it in seconds.
The point isn’t the chat box. It’s that the answer is sourced: something you can act on and defend, not take on faith.
Where it’s heading
From answers to investigations. The next step is an agent that doesn’t just answer “why did EU margin drop” but pursues it: checks the drivers, rules some out, follows the thread that matters. An analytical partner, not a faster lookup.
How we think about it
An answer you can’t trace is not an answer. Every response should be grounded in the semantic layer and show its work, the definition and the query both. When there’s no trustworthy basis to answer, the system should say so rather than guess, which is as much a question of trust and governance as of clever models. A wrong answer delivered fluently is worse than none, because someone acts on it.
Chat with your data, in short.
Is this just a chatbot bolted onto a dashboard?
No. A chatbot guesses against raw tables. This answers from your governed semantic layer, so "revenue" means what your business agreed it means, and every answer can show the query behind it.
How do we know the answer is right?
It's traceable. Every answer returns the definition and the query behind the number, so an analyst can check it. An answer you can't trace isn't an answer, and the system should say so rather than guess.
Does this replace our analysts?
No, it changes what they spend time on. Routine "what's the number" questions answer themselves, freeing analysts for the harder work: defining metrics well, chasing the surprises, and curating what the system knows.
Keep exploring
Reports that design themselves
Dashboards and reports generated from intent. The "can you build me a view" backlog disappears; humans curate what the AI drafts.
Insights that surface themselves
The system narrates what changed and why, proactively, instead of waiting to be asked.
Alerts that find the right person
Metrics watched continuously; anomalies and threshold breaches reach the right person with context, not noise.
Where could this take your BI?
If this is the direction you want to head, we should talk.