/ Solutions / Conversational analytics
Ask anything. Know why.
Your data stays where it is. Ask in plain English, in ContextBase, Claude, or ChatGPT. Every answer arrives with its definition, its rows, and its checks.
Why did paid PMPM rise in March?
Paid PMPM rose $6.40 in March, to $472.60. Most of the rise came from outpatient surgery that moved from surgery centers to hospital outpatient departments.
- Outpatient surgery+$3.90
- Specialty drugs+$1.70
- Everything else+$0.80
The same procedures cost about twice as much there.
Evidence · 4 sources6 checks passed
Answer ready: 7 queries across 6 sources, with 2 approved entries applied.
The difference
AI on your data, before and after
Today
- AI answers pasted into files, then forwarded anywhere
- The copy loses the question, the query, and the context
- Each chat decides what PMPM means
- A number moves, and nobody can say why
With ContextBase
- Ask in ContextBase, Claude, or ChatGPT, and the data stays where it is
- Share the whole chat, with its history and artifacts
- Every metric read from one approved definition
- Every answer traced to its rows, its rule, and its cause
Know why
A number, and the reason behind it.
Follow any number back to its rows, its rule, and its cause.
Paid PMPM · MarchDemo data
$472.60+$6.40 vs February
Outpatient surgery claims, March212 of 1,480 moved to a hospital department
paid_pmpm.md → claim_line → place_of_service 22 → +$3.90
Stay live
Answers that stay true.
Save an answer as a dashboard. It refreshes on your schedule and spends no tokens.
The same answer, a week laterDemo data
/ Asked on MonPaid PMPM by week?
Chat exportpmpm.html
$405.20Asked today
Your dashboardin ContextBase
$405.20Refreshed 06:00 · Mon
Getting started
Three steps, on the warehouse you already run
No new BI tool and no migration project. The same warehouse, the same questions, answered in a form that lasts.
01 / Connect
Read what is there
Point ContextBase at the warehouse your teams already query. It reads the schema, the joins, the stored procedures, and the metrics people keep asking about, read-only and in place.
Nothing copied
02 / Define
Agree the numbers
The metrics everyone asks for are defined once, approved by the people who own them, and versioned in the knowledge graph.
One definition each
03 / Ask
Answers that last
People ask in plain English, in ContextBase or the AI assistant they already use. What comes back says why, and is logged.
Live and governed
Use cases
Where it starts
01Claude and ChatGPT, governed
Keep the assistants your teams already use. Over MCP they answer from the same approved definitions.
02Why a number moved
Ask why paid PMPM rose or cost share fell, and get the cause with the rows, the rule, and the checks behind it.
03Retire the pasted charts
Save the answers people reuse as dashboards that refresh without spending tokens, and let the one‑offs go.
04Board-ready numbers
Every figure traced to its query and its definition, so the number in a chat is the one finance signs off.
05Scheduled answers
The question asked every Monday, answered every Monday by a bot that runs on a schedule and logs every run.
06Shared chats
Send a teammate the whole conversation. They check out a copy and keep going.
Questions
What teams ask before they start
We already connected an AI assistant to our warehouse. Why add ContextBase?
Keep it. The gap is around the answer. Which definition did it use? Why is the number what it is? Over MCP, your assistant answers from the definitions your team approved, and every query is logged.
Do people have to stop using their AI assistant?
No. Claude, ChatGPT, and your own agents read ContextBase over MCP. People keep asking where they already work.
How does it find the cause?
Every answer arrives with its queries, its rows, and checks that say where it is weak. The knowledge graph knows what each column means, where it is set, and what reads it, so an answer can follow a number back to the rule and the table behind it.
How does it handle sensitive data?
ContextBase reads your systems where they sit, through the grants you set. PHI protections are configured per deployment, we sign a BAA, and nothing you send trains a model. See security
Is this a new BI tool?
No. ContextBase works alongside the BI you run, and answers saved as dashboards refresh without spending tokens. If you want fewer BI tools, that is its own project. BI consolidation
See it on your data.
Bring the questions your teams already ask AI. We connect the warehouse behind them, define the metrics they lean on, and show you every answer with the reason behind it.
- 01A call to pick the questions and the warehouse behind them
- 02We connect it read-only and define the metrics behind those questions
- 03Your team asks away, and every answer says why