Organizations are investing in AI without fixing the underlying data problem. Unstructured, siloed, and inconsistent data will produce unreliable AI.
Your AI Is Only as Good as Your Unstructured Data.
One Idea Worth Acting On.
Every oncology organization wants AI to synthesize insights, flag safety signals, and accelerate decisions.
Most of them are feeding it manually filtered CRM entries, fragmented system exports, and MSL meeting notes compressed into three bullet points.
The bottleneck to digital transformation in oncology is not the technology. It is the data going into it.
Organizations are investing millions into generative AI, predictive analytics, and advanced digital platforms. The assumption is that layering sophisticated AI on existing systems will produce strategic intelligence.
It will not. Not with the current data infrastructure.
AI is a powerful engine. Fragmented, manually filtered, and structurally inconsistent data is terrible fuel. Artificial intelligence does not eliminate organizational dysfunction — it amplifies whatever system it is fed. Faster fragmentation is not transformation.
As Sarah Clark and Tam Nguyen, Chief Digital and AI Officers and contributors to Voices of Oncology, describe it: the data infrastructure in most oncology organizations is deeply fragmented. Clinical Operations has an EDC. Commercial has a CRM. Safety has a pharmacovigilance system. Medical Affairs has an insights platform. None of these systems was built to talk to the others. And none of them contains the full picture of what the organization actually knows about its asset.
The most valuable signals in oncology drug development do not live in structured databases. They live in the unstructured text that MSLs write after scientific exchanges — the physician's specific concern about a competitor's safety signal, the patient population the KOL is seeing that does not match the trial population, the access barrier a community oncologist mentioned in passing.
When that intelligence is compressed into three CRM bullet points before it enters any system, the AI analyzing those systems cannot find it. It was discarded at the first handoff.
Organizations are not just feeding AI incomplete data. They are feeding it human-filtered assumptions about what matters. That is the deeper problem. The most important strategic signals are being removed by the people who captured them — before any technology ever sees them.
An oncology company launches a new AI-powered insights platform. Within six months, the volume of analyzed data points has tripled. Leadership celebrates the digital transformation milestone.
But the insights being surfaced are synthesized from the same manually filtered CRM entries the team was reviewing before the platform existed. The AI has made the existing data faster to access. It has not made the existing data better.
The MSL's three pages of field notes — the ones that were compressed into three bullet points before entering the system — are still sitting in a laptop. The AI has never seen them.
Before investing in AI, organizations need to invest in what AI will be fed. That means full MSL meeting notes — not three-bullet summaries — entering a system that can analyze unstructured text. It means cross-functional data architecture that connects the CRM, the safety database, the insights platform, and the field medical system into a queryable ecosystem.
And it means acknowledging that the data problem is a governance problem as much as a technology problem. Each function has been treating its data as a local resource for its own work. Building an AI that benefits the whole organization requires functions to share what they know — which requires governance that most organizations have never built.
AI is not the transformation. Better data is the transformation. AI is what makes better data faster.
The organizations that will lead in digital oncology are not the ones buying the most sophisticated AI. They are the ones fixing their data infrastructure first.
Add Your Voice — Where are you seeing this? What's working — and what's not?
Emmanuelle Jacqueta, Ghania Kerouani-Lafayeb, Francoise Grudeb, Sergio Goncalvesb, Annie Lorenced, Florence Turcryb, Liora Brunelb, Laetitia Belgodereb, Adrien Monardc, Gaëlle Guyaderb, Lotfi Boudalib, Nicolas Albin
iNIZIO
iNIZIO
Jayson L. Parker, Sebnem S. Kuzulugil, Kirill Pereverzev, Stephen Mac, Gilberto Lopes, Zain Shah, Ashini Weerasinghe, Daniel Rubinger, Adam Falconi, Ayse Bener, Bora Caglayan, Rohan Tangri, Nicholas Mitsakakis
Avi Cherla, MSc; Huseyin Naci, MHS, PhD; Aaron S. Kesselheim, MD, JD, MPH; Bishal Gyawali, MD, PhD; Elias Mossialos, MD, PhD
Michael Pietrack