BI with artificial intelligence for hospitals

Artificial intelligence in hospital BI is for one simple thing: letting any manager ask a question about the hospital's data and get a correct, explained answer with its source, without having to build filters or read pivot tables. To work in a hospital, AI needs three guarantees: it uses the BI's own numbers, it shows where they come from and it handles health data with the care Brazil's LGPD requires.

What AI does well in BI

• Answering direct questions: "what was October's revenue and how did it change against September?", "which payers have the lowest margin?".
• Explaining the dashboard: what each indicator means and how it was calculated.
• Pointing out what changed: the period's largest variances, by area, payer or cost center.
• Organizing action: turning a problem seen on the dashboard into an action plan (PDCA) or a SWOT analysis for the meeting.

Where AI needs limits

Generic language models without access to the hospital's data can produce numbers that look plausible but are made up. In management, that is worse than having no answer. That is why trustworthy BI with AI follows rules:

1. The number comes from the BI, never from the language model. The AI chooses the measure and the cut; the calculation is done on the data.
2. The answer shows its source: the measure, the period, the filters and what was left out.
3. The AI says when it does not know, or when the question belongs to another dashboard.
4. Sensitive data stays where it belongs: patient data should not be sent to third-party AI services without a legal basis and a risk assessment.

How AI works in Clic Health

Clic Health has two AI modes, both inside the system itself:

• Essential AI: each dashboard has a library of questions, with phrasing variations, and answers are calculated on the spot with the data and filters on screen. It does not depend on an external AI service. This is the default mode.
• Advanced AI: a language model that runs in the user's browser, for freer conversations. It builds on the answers calculated by the Essential AI, which reduces the risk of made-up numbers, and keeps the source in view. It requires a computer with a compatible graphics card.

In either mode, the answer states its origin ("ready answer from the page library", for example), the period and the measure used, and suggests related questions. Today the assistant works in Portuguese.

Example questions

• "What was last month's revenue and how much did it change?"
• "Which payers have the lowest contribution margin?"
• "How has average length of stay evolved over the last 12 months?"
• "Where was actual furthest above budget?"
• "How much is in accounts pending for more than 30 days?"

Tips for asking better questions are in the article how to ask AI good questions about hospital data. For the uses and limits of AI in management as a whole, see artificial intelligence in hospital management.

AI with Tasy data

For hospitals running Tasy, the AI works on the same data model as the dashboards, fed by Tasy. Details are in AI applied to Tasy data. The most common questions are in the AI FAQ.