AI in Medical Affairs: Where to Start, What Actually Works
About the webinar
We want to answer the question most medical teams are sitting with right now: where does AI actually fit in our work, and where do we start? Philip Vyt and Pierre Metrailler have spent the past year working with medical teams on solving this exact question.
What was discussed
The maturity gap nobody’s naming
Philip opened with a self-assessment framework his teams use to place themselves on the maturity curve: basic (one channel, one tactic, like a single webinar or detail aid), multi-channel (one message pushed consistently across a few channels), and omnichannel. Most teams, he said plainly, are still at basic. Not a controversial claim, but one worth stating out loud, since the industry has been talking about better HCP engagement for years without much movement.
Medical affairs sits even earlier on that curve than commercial, and Philip pointed to why: for a long time, the prevailing attitude was “consent capture, that’s commercial’s job, not mine.” That mindset, more than any tooling gap, is what’s kept medical behind.
But he was careful not to be purely critical. He’s seen real movement: five to seven years ago, an omnichannel-minded medical affairs team was almost unheard of. Today, 70% of his engagements are with exactly these teams. The shift is real, it’s just uneven, and it depends heavily on a handful of strong internal believers pushing it forward.
He also named the opportunity directly. Medical holds something commercial doesn’t: physicians who actually want to hear from them. Commercial has bigger budgets and more tech licenses, but HCPs are tired of product pushes. Medical has the pre-launch access, the KOL relationships, and a “vast library” of scientific content that’s underused. His line: no touchpoints, no data. No data, and AI is an engine with nothing to run on.
The poll results backed him up. Most attendees said their congress data gets captured but scattered across systems. Only a small minority said their data is structured and CRM-ready. Philip’s read: medical isn’t behind because the opportunity isn’t there. It’s behind because the industry has historically treated consent capture and structured follow-up as commercial’s job, not medical’s.
Why congress is the highest-leverage place to start
Pierre made the scale of the opportunity concrete. A typical oncologist attends three to four congresses a year, generating roughly 150 touchpoints per event: badge scans, Wi-Fi logins, and all the pedestrian stuff. Out of that noise, a well-instrumented program can capture 10 to 40 meaningful touchpoints per HCP: booth conversations, KOL meeting transcripts, symposium questions, voice memos, and more.
Philip’s framing: without those touchpoints, there’s no data, and without data, “AI” is an engine with nothing to run on. Congress works as a starting point precisely because it’s contained. Three days, a clear start and end, and a concentration of your most relevant HCPs in one place.
The real cost of delay
Pierre and Philip returned to something HCPs are far less forgiving about than most teams realize: the Ebbinghaus curve, or forgetting curve. Data from ZS shows 81% of HCPs give you roughly two weeks post-congress before they mentally move on. Philip’s experience with teams suggests the real number is closer to seven days, sometimes less.
A live poll asked attendees what actually stops follow-up from happening in that window. The top answer, by a wide margin: no time, already onto the next congress. Philip’s response: teams plan meticulously for the congress itself and consistently underestimate the work that comes after. He called it the “soufflé effect”: all that energy and attention collapses the moment the event ends, unless a team has already planned for what happens next.
From mess to a working agent
Pierre closed with a live build: pulling structured congress data directly into Claude through an MCP connector, no CRM access or data lake required, and generating a personalized follow-up email for a specific HCP in seconds, grounded in her actual booth visit, symposium questions, and KOL meeting notes.
Philip’s takeaway: this is what happens when data capture and AI actually connect. The barrier was never the tooling. It was believing the workflow could be this direct.
Speakers
- Philip Vyt - Founder, Shyft
- Pierre Metrailler - CEO, Onomi