I attended a great panel discussion at WDSC2025, moderated by Lucinda Smith with Julie Thomas, Mariette Boerstoel, and Ellen Mishalov. I took some notes and here is my summary of the discussion.
What we heard
• Foundations first: Governance and data reality matter. Know what data you have, where it lives, and its quality. Don’t wait for “perfect” data.
• Right-sized governance: Create an AI council with privacy, legal, security, and data leads. Use it to enable, not block. Start with low-risk use cases to build trust.
• Small bets, fast learning: Replace “big bang” programs with short pilots that stack into production. Accept “good enough,” then iterate.
• Redesign processes, not just systems: Don’t “AI-ify” old workflows. Simplify first, then add automation with a clear human-in-the-loop.
• Measure what matters: Track technical metrics at field level and business KPIs like time-in-motion and cost to value. Watch end-to-end flow to avoid downstream bottlenecks.
• Explainability and trust: Pair outputs with human review, learn from correction patterns, and monitor for drift to sustain gains.
• Change management is the unlock: Bring partners and junior operators to the table early. Upskill broadly. Showcase quick wins so adoption spreads.
Real examples
• Scaled language translation to speed case handling.
• Automated intake and auto-narratives in production; literature extraction and GenAI for aggregate docs in progress.
• Exploring agentic patterns and reflection loops to move from human-in-the-loop toward lighter oversight.
Closing advice from the panel
• Remember the patient and your people.
• Simplify, then experiment.
• Stay anchored to purpose, not hype.
What’s the one process you’d simplify tomorrow to make AI deliver measurable value in PV?
#WDSC2025 #Pharmacovigilance #DrugSafety #AIinHealthcare #GenerativeAI