Introduction

SPEAKER: 
Avinob Roy, VP & GM Commercial Analytics Product Offerings, Global Commercial Solutions, IQVIA 

Agentic AI is rapidly moving from experimentation to production in pharma. Yet despite increased investment and technical progress, most programs stall before they meaningfully change how commercial decisions are made. The problem is no longer building AI agents. It is ensuring they operate reliably, govern themselves appropriately, and are adopted by real teams inside real workflows. 

As agentic AI matures, the differentiator between success and failure is not model sophistication. It is readiness. Organizations that treat agentic AI as a technology problem struggle to scale impact. Those that treat it as an operating model change are far more likely to see durable results. 

Building an AI agent is no longer the hard part. Making it matter to commercial teams is. 

At Pharma 2026, Avinob Roy, VP and GM of Commercial Analytics Product Offerings at IQVIA, offered a blunt assessment grounded in real production experience. Building the agent is easy. Deploying it in production is harder. Operationalizing it and driving adoption is hardest of all. 

This perspective is shaped by reality, not theory. Today, many top pharma companies have agents running in live commercial environments. These are not pilots. They reflect years of evolution from narrow NLP tools and chatbots into fully agentic systems embedded in day‑to‑day decision making. 

What differentiates success from stalled programs is not the model. It is readiness. 

Experience across deployments shows that agentic AI consistently delivers value only when three prerequisites are in place: 
• Machine‑readable, context‑rich commercial data 
• Codified commercial logic that reflects how decisions are made 
• An intentional adoption architecture tied to governance and workflows 
 
Organizations that address these foundations before scaling agents are far more likely to achieve lasting impact rather than short‑lived experimentation. 

Commercial Data Was Never Built for Machines 
When agents fail, organizations often blame the AI. In practice, the root cause is structural. Pharma commercial data environments were built for humans, not for large language models. Data lakes and warehouses assume institutional knowledge that machines do not possess. 

Early production deployments made this clear. In one case, an agent delivered correct answers only part of the time and hallucinated the rest. The solution was not replacing the model. It was rethinking how data is prepared for machine consumption. 

The fix was context, not centralization. Metadata, annotation, and enrichment allowed data to remain where it lived while becoming understandable to machines. For commercial leaders, this is not a technical nuance. When agent outputs influence targeting, territory design, or spend allocation, poor grounding introduces enterprise risk rather than simple usability issues. 

Your Best Commercial Logic Lives in People’s Heads 
Even perfectly prepared data cannot compensate for undocumented decision logic. In most pharma organizations, critical commercial processes are unwritten. Segmentation rules, targeting heuristics, escalation paths, and market‑specific exceptions exist as tribal knowledge inside experienced teams. 
Agents cannot operate within guardrails that do not exist. Governance is impossible when logic is implicit. 

This is where many programs break down. Agentic AI is often framed as a headcount reduction strategy. In reality, successful deployments require deeper human involvement up front. Subject matter experts must externalize decision logic, define boundaries, and continuously validate outputs. 

The deeper trap is replication. Treating agents as faster versions of humans simply digitizes workflows that were never designed for machines. Real value emerges only when workflows are reimagined for agent participation rather than automated as‑is. 

Counting Agents Instead of Designing an Ecosystem 
As agent tooling becomes more accessible, agent proliferation is inevitable. Every function, market, and team will build them. The mistake is celebrating volume without architecture. 

Organizations that fail to define an ecosystem struggle to measure ROI and adoption. The early internet was not won by those who built the most websites, but by those who solved interoperability, discovery, and trust. Agentic AI is entering a similar phase. 

Competitive advantage will come from how agents connect, share context, protect data, and produce auditable outcomes. As agents influence commercial decisions at scale, regulators and internal stakeholders alike will expect transparency into how conclusions were reached. Retrofitting governance after agent sprawl is far harder than designing it intentionally from the start. 

The Readiness Sequence Most Teams Skip 
Data readiness, process codification, and adoption architecture are not parallel workstreams. They are sequential dependencies. 

Agents cannot be governed without documented processes. Processes cannot be tested without machine‑ready data. Adoption cannot be measured or scaled without both. Yet many organizations attempt all three at once, which helps explain why proof‑of‑concept to production conversion rates remain low. 
One global pharma organization followed this sequence intentionally and replaced hundreds of reports across markets with a single agent delivering consistent answers to executives and field teams alike. This was not an automation win. It was a transformation in decision architecture made possible only by sustained upstream readiness. 

The Real Question for Commercial Leaders 
The question is not how many agents your organization has built. It is which prerequisite you are furthest from meeting, and whether you are willing to invest there before building the next agent. 

Teams that redirect spend from agent construction to readiness will deploy fewer agents this year. They will also be the teams whose agents are still delivering measurable value years from now. 

To get you highlights of Pharma 2026 faster, we are using generative AI technology to summarise the transcripts of the sessions. If you have any feedback about the summary, please contact lucy.fisher@thomsonreuters.com. 

Discover more on this topic at Pharma Commercial Data & Tech Europe 2026 (4-5 November, London) Europe’s collaborative home for data and tech pioneers. Visit the website here.

"QUOTE"

 Text

HCP Fatigue Redefines the Value Proposition 
The consumerization trend Jeppe described, where patients arriving at physician appointments with LLM-generated treatment preferences, prepared to seek another opinion if rebuffed, is reconfiguring who pharma needs to engage and what that engagement must deliver.  

"If your company can bring some value to your customer beyond just an email campaign, then you are going to counterbalance the consumerization. But if your company, your tactics consist only to send 10 emails per month, then yes, you will be replaced," said Florent. The examples he cited are instructive: supporting an HCP navigating drug access for an asylum seeker, enabling a physician to organize community upskilling sessions for regional colleagues. Neither is a digital engagement play. Both are high-touch interventions that no AI-mediated channel can replicate, and both generate reciprocal loyalty that insulates a company from being disintermediated by Medscape, ChatGPT, or a competitor's next campaign. 

Sameer extended the stakeholder map to capture the structural shift underneath the consumerization trend. Patients increasingly arrive at pharmacies before they go to HCPs, self-medicating or seeking first-line OTC alternatives informed by LLM queries. Engagement strategy now requires pharmacy-level touchpoints at the awareness stage, HCP coaching on handling patient-initiated objections, and patient-facing content early enough in the journey to shape the conversation before it reaches the clinic. Most pharma commercial models have not fully operationalized any of these three. Sameer noted the social dimension bluntly: in markets with weak prescription controls, LLM-driven self-medication is not merely a commercial challenge but a public health one, and the industry's engagement needs to be matched by policy intervention. 

Florent surfaced a finding from Roche's real-time HCP feedback capture that reframes the competitive risk. "There is a digital fatigue from the HCPs. After COVID we have been doing a lot of webinars, a lot of remote meetings. They are bombarded with email campaigns and they just don't want to receive 50 emails per week. They just want sometimes to talk to a human." Engagement volume, measured as a proxy for engagement quality, is producing measurable deterioration in the customer relationship it is designed to strengthen. 

The Structural Fix Is Organizational 
The three prescriptions offered across the session, enterprise data governance, decision system architecture, product owner accountability, look like alternatives. They are not. Each addresses a different failure mode in the same organizational system. 

Phoebos was direct about the sequencing: "A lot of the time we think that we fix the technology problem and then automatically we will improve the way we do precision engagement. But actually, in my opinion, it's more about people, process and technology, and in that right order of importance." Technology scales effective organizational design. It does not substitute for it. 

Jeppe's closing provocation sharpened the boundary condition: are there barriers between R&D and commercial that need to be maintained? Florent's answer was instructive. At Roche, data domain lakes enforce access by organizational role, with workflow-governed permissions for cross-domain requests. The architecture does not eliminate boundaries; it makes them precise and navigable rather than blunt and impenetrable. Medical inquiry content stays within medical. Sales pricing data stays within its domain. But insights that can legitimately cross domains, clinical patient journey work, for instance, have a governed pathway to do so. Sameer's partial defense of silos acknowledged the CFO logic that created them: "The cost of a breach will overwhelm any additional cost of maintaining those silos." The problem is not that the logic was wrong. The problem is that it was applied without precision, producing blanket fragmentation where selective permeability would have served better.

Florent described a production-deployed pipeline at Roche that illustrates what the organizational architecture enables. HCP master data records are linked to public social media accounts via fuzzy matching. Real-time posting activity feeds into next best action content recommendations through natural language processing. If an HCP is publicly posting about a specific mechanism or molecule, the system surfaces relevant content for the next interaction. That capability exists because Roche built the data domain architecture, the product owner accountability model, and the cross-functional governance that makes it operational. The pipeline is the output of the organizational design. 

Companies still treating precision engagement as a technology procurement question are solving for the wrong constraint. The question worth asking before the next planning cycle is not which AI platform to invest in, but whether the organizational architecture exists to make any prediction actionable before the clinical moment it was designed to address has already passed. For most organizations, that answer is currently no, and no model update will change it. 

To get you highlights of Pharma 2026 faster, we are using generative AI technology to summarise the transcripts of the sessions. If you have any feedback about the summary, please contact lucy.fisher@thomsonreuters.com. 

Discover more on this topic at Pharma Commercial Data & Tech Europe 2026 (4-5 November, London) Europe’s collaborative home for data and tech pioneers. Visit the website here.