How to ask AI good questions about hospital data
A question with a clear period, measure and scope gets a precise answer. A vague question gets a vague, or wrong, answer.
Good questions to AI about hospital data have three elements: the period, the measure and the scope. "How is the hospital doing?" is too vague. "What was private payer revenue in October and how did it change against September?" has everything the AI needs to find the right number. This guide shows how to phrase questions, examples by topic and how to check the answer.
The three elements of a good question
| Element | What it defines | Examples |
|---|---|---|
| Period | When the data is from | "in October", "last closed month", "over the last 12 months", "year to date" |
| Measure | Which number | revenue, contribution margin, denial rate, average length of stay, unbilled accounts |
| Scope | Of whom or where | by payer, in the ICU, in SUS, in orthopedics |
Add a comparison when you want the variance: "against last month", "against last year", "against budget".
Examples by topic
- Revenue: "What was last month's revenue and how did it change against the same month last year?"
- Payers: "Which payers have the lowest contribution margin over the last 12 months?"
- Billing: "How much is in accounts pending for more than 30 days?"
- Controllership: "Which cost centers were furthest above budget in October?"
- Volume: "How has the number of surgeries evolved against last year?"
- Length of stay: "Which department has the longest average length of stay this year?"
Ask on the right dashboard
In BI with AI, the question is answered using the open dashboard's data and the filters on screen. A question about margin by payer asked on the volume dashboard may not have the right measure. In Clic Health, when a question belongs to another dashboard, the AI says so and offers to open the right one. Today the Clic Health assistant works in Portuguese.
How to check the answer
A trustworthy answer shows:
- The measure used (for example, "patient revenue, accounting accrual").
- The period and the comparison period.
- The filters applied and what was left out.
- The origin of the answer.
If the answer does not show this information, be suspicious. In the first few weeks, compare some answers with the dashboard: it is the fastest way to build trust, or to find a problem.
Common mistakes
- Asking without a period: the AI will pick one, maybe not the one you wanted.
- Mixing two questions: "what are revenue and denials by payer and department this year?" Better to split.
- Using internal nicknames the AI does not know. Use the dashboard's measure names.
- Asking for the cause as if it were data: "why did margin fall?" The AI can point to where it fell; the cause requires someone who knows the operation.
- Accepting a number without a source.
After the answer
Use the answer as a starting point: open the dashboard on the indicated scope, drill down to detail and, once the problem is clear, ask the AI for an action plan (PDCA) to take to the meeting. See AI's uses and limits in artificial intelligence in hospital management, and how it works in Clic Health in BI with artificial intelligence and AI with Tasy data.
Related questions
Do I need to phrase the question in a specific way?
No. The AI understands phrasing variations. Including period, measure and scope just increases precision.
Does the AI answer about patient data?
In Clic Health, no: patients are identified only by a code in the data model, and the AI works with management KPIs.
Can I ask the AI to build a report?
Yes. From the dashboard, the AI builds a PDCA action plan or a SWOT analysis. For detailed tables, use the dashboard's Excel export.
Written by the team behind Clic Health: accountants, controllers and developers with more than 20 years of work in hospital management.