From Evidence to Action: Speeding Pharma Impact with Practical AI Adoption

Meta description: Pfizer medical director Nabil Khan explains how pharma teams can use AI to turn evidence into faster, verified clinical and strategic decisions.

Tags: Pharma AI, Evidence Synthesis, Medical Affairs, AI Governance, Nabil Khan


Pfizer medical director Nabil Khan is mapping AI evidence workflows in pharma as clinical information grows faster than teams can turn it into decisions. The gap matters because pharmaceutical strategy increasingly depends on digesting evidence that arrives daily from publications, congresses, real-world studies, and field conversations.

Background: Evidence Growth Outpaces Decision Speed

In fiscal year 2023, PubMed added more than 1.5 million citations, roughly 4,300 a day, according to the National Library of Medicine. Medical affairs teams must also absorb congress data, real-world evidence, and field insights from healthcare professionals, known as HCPs. That volume makes it harder for pharma organizations to move from clinical information to action at the speed the information arrives.

The bottleneck is not usually access to evidence. Khan’s view is that synthesis, analysis, and decision-making have not scaled at the same pace as the evidence being generated. In the field, medical teams must judge which HCP conversations contain clinically relevant insight. Inside medical affairs and clinical development, those insights arrive secondhand and must be converted into strategy. On both sides, the main struggle is separating noise from signal.

Development: AI Speed Arrives With New Failure Modes

AI can shorten synthesis time, but it introduces its own risks. In 2023, researchers reported in npj Digital Medicine that large language models, or LLMs, used for medical evidence summarization could produce factually inconsistent and overly convincing summaries, with more errors on longer texts. The finding pushed regulated industries to ask a narrower question: which decision should AI improve, and how will leaders know it worked.

The U.S. Food and Drug Administration has since proposed a risk-based credibility framework for AI models used to support drug regulatory decisions. For pharma leaders, that makes the conversation less about model capability and more about evidence governance.

On the AI in Business podcast, Emerj Senior Editor Yolandi de Weerdt spoke with Khan in the episode “Accelerating Evidence to Action in Pharma with Practical AI Adoption.” Khan joined Pfizer in 2024 as Medical Director for Internal Medicine Antivirals. His expertise spans Medical Affairs, Clinical Development, Clinical Trial Operations, and Scientific Engagement. Before Pfizer, he monitored clinical trial sites and supported studies in vaccines, cardiometabolic disease, ophthalmology, and neurology in senior clinical research roles at PPD, ICON, and IQVIA. He served as Director of Clinical Operations at Impact Physician Group. A physician with more than a decade in clinical research and operations, he holds a medical degree from Xavier University School of Medicine.

Khan sees three shifts when AI enters evidence work: faster synthesis, better detection of emerging patterns, and a more consistent evidence base across medical affairs, clinical development, and commercial teams. He calls the last point a stabilizing feature. His goal is to let experts spend less time gathering information and more time acting on it.

Still, he warns that adoption without a defined problem is like owning a toolbox with no sense of which tool fits the job.

“Successful adoption of AI has to start with a clearly defined business problem, not necessarily the technology itself. So leaders should ask themselves and should ask their teams: What decision are we trying to improve on? What decision are we trying to accelerate? What decision are we trying to scale?” — Nabil Khan, Medical Director for Internal Medicine Antivirals at Pfizer

Khan’s framework asks leaders to name the decision, the evidence source, the governance point, and the downstream outcome. Trial site selection shows how far downstream the proof can sit. AI can help identify potentially relevant HCPs in a region, but the test comes only after field medical teams vet investigators and clinical development moves sites through protocol training, regulatory review, compliance, and activation. The final measure is whether those sites are producing clean trial data.

Current Stage: Verification, Training, and Defensible Use

Khan’s team uses AI heavily at medical congresses to plan limited time on site and summarize large volumes of material. The output must be checked because separate presentations can be merged in AI summaries when titles, topics, or speakers look similar. Once an incorrect synthesis moves into field reports or strategy decks, the error can affect every downstream decision.

His rule is to treat AI as a tool rather than a source. Before an AI summary moves to other teams, it should be traceable to the original publication, abstract, poster, or presentation; receive closer review for look-alike inputs; and be validated by a named human expert.

He also warns that consistency can be misleading. “Sometimes consistency in the wrong direction is not a good thing,” Khan says. If AI makes the same mistake across many summaries, the error is not just isolated; it can spread.

Training is the next bottleneck. Khan says organizations often underweight training, even though employees must learn how AI intersects with regulatory requirements, compliance obligations, copyright, and external communications. He notes that ChatGPT, Claude, Copilot, and other systems behave differently, and that using AI on a personal phone is a different activity from using it on a work computer for regulated work. When AI-generated content is shared externally, the company carries the risk as well as the individual.

Khan frames the learning curve as a funded sequence: a training quarter, about three to four months; a supervised-use quarter; and routine integration once teams can interpret outputs with confidence. He adds that the pace depends on the use case: summarizing or drafting emails requires less readiness than summarizing a clinical or journal article or building a presentation.

Large organizations are still in the trial-and-error phase of AI implementation. The current stage is moving from raw speed toward decision-anchored, source-verified, and supervised AI use.