From AI Hype to Field Reality
Much of the current discussion around AI and field readiness has focused on tactical applications, such as AI-simulated HCP conversations for upskilling. Less attention has been paid to how AI-driven drug development reshapes the scientific narrative for field teams and the questions they will face as a result.
For many in the field, AI may feel like someone else's domain, associated more with data scientists and developers than with MSLs or commercial teams. But field teams don’t need to be experts. They need fluency: sufficient AI literacy to anchor scientific credibility when AI-driven evidence is central to the product story, whether their own or a competitor’s.
AI is increasingly embedded in the evidence architecture of drug development, shaping how data is generated, structured, and interpreted. This expanding role is already reflected in its increasing presence across the drug development and regulatory landscape.
Regulators Are Already There
An analysis of FDA AI submissions between 2016 and 2021 conducted by the Center for Drug Evaluation and Research (CDER) showed a sharp (reported as near tenfold) increase in AI/machine learning (ML)-related submissions between 2020 and 2021, spanning drug discovery, clinical development, and postmarketing activities.
Applications include AI-facilitated patient enrichment in trial design; prognostic risk stratification; synthetic control arms in cases where traditional comparators are impractical (and have been accepted by regulators); dose optimization based on patient characteristics; and AI-enabled endpoint and biomarker assessment.
These AI/ML applications increasingly extend beyond clinical trials into real-world data sources used to support and contextualize clinical evidence. As adoption expands, regulatory frameworks have evolved in parallel to govern AI use across the development lifecycle.
As AI/ML applications have expanded, regulators around the world have released guidance and discussion papers on AI across the medicines lifecycle. These documents are informed in part by earlier governance models developed for Software as a Medical Device (SaMD) and reflect an evolution toward Good Machine Learning Practices (GMLP), particularly around lifecycle management, transparency, and model governance.
Across these guidance documents, regulators are converging on a consistent, risk-based principle in which expectations for validation and oversight depend on how AI is used within the development process. Importantly, the level of scrutiny applied to AI-enabled approaches depends on their context of use and their potential impact on regulatory decision-making. Applications that directly influence clinical trial design, patient selection, or data interpretation may carry higher regulatory risk and require more rigorous validation.
Regulators emphasize that AI-enabled approaches must be transparent, ethical, and built on data that is fit for purpose (appropriate in quality and scope for the intended application). Sponsors must demonstrate that model outputs are reliable, generalizable, appropriately characterized, and mitigated for bias to support applicability to the intended patient population.
When AI Becomes Part of the Story
For field teams, the implications of this regulatory evolution are indirect but important. In many cases, AI functions as a backend tool operating within the evidence architecture without being visible in clinical conversations, so field engagement remains focused on clinical outcomes, not technology. In some cases, however, AI plays a load-bearing role in the evidence underpinning the product. Building AI literacy is about being ready to address AI when it is material to the evidence, while maintaining the discipline to keep the clinical message front and center when it is not. The applications that carry the highest regulatory scrutiny — those that directly influence trial design, patient selection, or data interpretation — are the same ones most likely to surface as questions in the field.
How that discipline plays out in practice depends on the stakeholder. For formulary decision-makers, AI-derived evidence enters a framework centered on population-level outcomes and cost, a dimension that deserves dedicated treatment beyond the scope of this piece. For clinicians and clinical investigators, the obligation is to the individual patient, which means AI-driven results must be grounded in clinical logic and mechanistic transparency before they are credible.
For MSLs: When AI Shapes the Evidence
The weight AI carries in the development story determines the questions clinicians will ask:
- If AI was used to enrich a study population by selecting patients with a specific phenotype or risk profile, clinicians will ask whether those patients reflect their practice, and whether the results are applicable beyond the study cohort.
- If AI generated a synthetic control arm because enrolling a traditional comparator was impractical (as in some rare disease settings) the clinical dialogue will center on how that control was constructed, what real-world data informed it, and whether the evidence supports confidence in the validity of findings for clinical decision-making.
- If AI drove dose selection or optimization, prescribers will want to understand what patient characteristics shaped the model and whether the dose holds across populations not well-represented in the training data.
- If AI was used to define or validate a primary endpoint, such as an AI-enabled digital biomarker (Wearables) or AI-driven image analysis, clinicians who question the endpoint will effectively be questioning the entire evidence base.
Navigating these inquiries requires the fluency to maintain credible, peer-level scientific exchange — particularly when addressing the “black box” problem. Success is not about opening the box to show the code; it is about articulating the clinically relevant factors that drove the model’s output and how they align with established disease biology.
For Commercial Teams: When AI Enters the Conversation
Commercial teams face a different but related set of challenges, driven more by narrative positioning than technical scientific depth. They must proactively anchor the development story rather than react to AI-related objections as they arise. This requires establishing the validity of AI-driven data early in the conversation to maintain narrative continuity and enable appropriate escalation to medical colleagues for deeper scientific discussion.
Beyond scientific engagement, AI introduces a new layer of competitive interpretation in the marketplace. Differentiation may increasingly depend on how AI is applied during product development and how that application is communicated. For example, when AI is used to identify and enrich for the most responsive patient subgroups during trial design, the result is a shift from population averages toward a more targeted efficacy narrative — one that commercial teams need to be equipped to position, not just reference.
That preparation extends beyond narrative positioning. Certain AI applications will inevitably surface as direct objections. While not an exhaustive list, examples like AI-generated synthetic control arms or AI-enabled safety monitoring illustrate scenarios where commercial teams should be prepared to address methodology at a high level.
Addressing these AI-related objections at a high level in field discussions does not require deep technical explanation, but rather the ability to articulate why the trial design aligns with regulatory guidance and how AI contributes to strengthening the evidentiary framework, for example by improving signal detection across large datasets. Establishing this legitimacy early helps keep the discussion moving rather than stalling on technical specifics.
The New Baseline
The question is no longer whether AI will become part of clinical dialogue, but how to recognize when it is already shaping the interpretation of clinical evidence, and when it remains part of the underlying analytical infrastructure. That distinction — between AI as a tool that generates evidence and AI as the foundation that gives it meaning — is what determines when and how field teams need to bring AI into the story. So, is AI in your narrative enabling the evidence, or defining it?
