Skip to main content
Apolla AI
The China business of a multinational pharmaceutical company

Pharmaceuticals

Average medical information response time from 2 days to 10 minutes, 100% response consistency

Multinational pharma, China: a medical information Agent that makes expert responses fast and compliant

Challenge

Medical information requests from physicians, pharmacists and patients were answered by the medical affairs team by hand, strictly on the basis of the label and approved materials, with nothing off-label. Tens of thousands of requests a year meant long turnaround, wording that varied by person, and more than the compliance team could review line by line.

The company wanted standard questions answered instantly and identically, with medical experts handling only complex cases.

Approach

We built a standard response library approved by both medical and compliance, then launched a medical information Agent strictly confined to approved content.

  • The standard response library is maintained by product and question type, with each response carrying its approved version and validity period
  • The Agent answers only from the library; anything outside it is routed automatically to a medical expert and logged
  • Calls and emails are classified automatically, and any possible adverse event is identified and forwarded to pharmacovigilance
  • Monthly reports on inquiry hot spots go back to the medical and marketing teams

Architecture highlights

A retrieval plus constrained generation architecture: generated content must cite a library entry, and the Agent declines to answer when there is no citation. Every response carries a version number, so historical responses remain traceable after the library is updated.

Customer feedback

The compliance baseline is to say only what has been approved. This Agent follows the rules better than any new hire.
Head of Medical AffairsThe China business of a multinational pharmaceutical company

Want more detail on this project?

In a workshop we can share the architecture decisions, the pitfalls and the actual timeline.