Case Study: Using AI to Triage Medical Information Requests

Medical Information (Med-Info) teams are the first people that pharmaceutical companies and healthcare professionals (HCPs) can turn to for help. But as drug portfolios grow and the number of enquiries rises, traditional triage methods, which often depend on manual work, begin to break down under pressure.

In this use case, we look at how a global pharmaceutical company worked with Newpage to set up a GenAI-powered triage system that cut response times by 65%, improved compliance, reduced stress for internal teams, and set the stage for fully automated Med-Info operations.

The Problem: Too Much, Too Complicated, and Too Late

Let us assume that the company in question had a specific process in place:
We in fact had many companies fall in a similar bucket. 

• A presence in more than 40 countries around the world.
• More than 150,000 Med-Info questions every year. 
• The company receives a variety of requests via email, web forms, and phone calls. 
• Each inquiry is manually handled by a small intake team. 

This means that someone had to read every single request, sort them by how urgent they were, tag them by therapeutic area, and send them to the right internal team—all in person.

What happened?

Questions that aren’t urgent may take 2 to 4 days to answer.

Missed escalations for complaints about product quality (PQCs) and bad events (AEs)

The burnout rate among medical affairs reviewers is on the rise.

Risk of not following the rules in audit trails and response times

The goal was clear: create an AI-powered triage engine that could work with their Salesforce Service Cloud system, speed up intake, and keep all the rules in place.

The Solution: Putting GenAI into Med-Info Workflows

Newpage worked closely with the business to add a GenAI triage layer to their current CRM system. The system was made to work like the way human intake specialists make decisions, but faster, more consistently, and on a larger scale.

Main Features:

• Natural Language Processing (NLP) helps you understand the whole situation behind a question.
• Models that classify things and give them tags like ‘therapeutic area’, ‘urgency’, and ‘safety flags’
• Smart routing to send questions to the right team without any help
• Logic for flagging AEs, PQCs, or mentions of products that aren’t on the label
• Reviewer dashboard for human approval and override
• Full audit logging to meet the requirements of 21 CFR Part 11 and internal SOP
• Governance Model: The organisation implemented a human-in-the-loop model to foster trust and ensure adherence to rules.
• For the first three months, someone had to approve all AI-generated classifications by hand.
• Weekly feedback sessions helped retrain the model and make tagging more accurate.
• Regular audits made sure that the AI kept up with compliance standards.

The Results: 6 Months to Go from Reactive to Proactive

The company underwent a complete makeover of its Medical Information operations just within six months of the GenAI-powered triage engine implementation. The time taken for responses decreased hugely, the reviewers could devote their attention to the medical content that was of a higher value, and compliance tracking was much more efficient.

The AI system was accurate all the time in identifying safety-related cases and it did so at a much faster rate than human reviewers. Human reviewers noticed a significant reduction in the triage manual workload as well. The entire workflow became more open, easily audited, and even more proactive – allowing Med-Info to make a shift from a reactive support function to a predictive and data-driven capability.

Important Things to Remember

1. Begin with Human-in-the-Loop

It was important to validate humans early on. It helped reviewers trust each other and gave them important feedback to make the model more accurate.

2. Prompt Engineering Is Important

Twelve percent more accurate results came from making small changes to how the model was prompted (language structure, formatting).

3. Don’t forget about language differences

Adding German and Spanish to NLP made it more consistent and usable worldwide.

4. Continuous validation is a must.

Monthly audits made sure that the company followed the rules and standards that were changing.

What’s Next: Getting to Full Content Automation

The company is now testing:

1. GenAI makes response drafts for common types of questions.
2. Integration of the Med-Info content library for quick access
3. Automatically flagging content that is old or doesn’t match

What is the end goal? A full-fledged AI-powered Med-Info ecosystem, where questions are sorted, written, checked, and stored in a single compliant workflow.

AI That Follows the Rules

In the pharmaceutical industry, AI needs to do more than just move quickly; it needs to move in a way that can be tracked.

This use case shows that GenAI can deliver speed, scale, and compliance without cutting corners if you have the right partner, governance, and design strategy.

We at Newpage make AI systems that obey the rules of science, the law, and the people who use them.

Trust is everything in healthcare, so AI has to earn it.