2026 Annual Conference - Agenda

What if every insight you needed could be surfaced instantly, governed, and trusted —freeing you to focus on the work that truly moves the needle?

In this two-part workshop, Axtria will demonstrate what becomes possible when AI becomes your most powerful ally, transforming you from a report generator into a strategic intelligence leader.

In Part One, Axtria's AI and Data Team, joined by a customer speaker from Alcon, will share how AskADI, powered by Axtria InsightsMAx.ai, transformed customer experience and clinical performance across its online store and insight center. By deploying intelligent agents capable of resolving FAQs, tracking orders, and surfacing KPI-driven GenAI insights through natural language, Alcon achieved 80% faster query resolution and 24/7 autonomous service delivery, giving you the space to do more of the work that matters.

In Part Two, you'll hear a compelling real-world success story before going hands-on with Axtria Headquarters Intelligence (HIQ). HIQ doesn't just answer questions — it reasons through them, validates every response with built-in confidence scoring, and connects directly to your existing data warehouse. No ETLs. No guesswork. Just you, equipped with a tireless intelligent copilot that handles the data legwork while you drive the strategy.

Together, these sessions chart a bold new path: from reactive reporting to proactive, governed, autonomous commercial intelligence, with you at the helm. This is what the future of analytics looks like. Come experience it firsthand.

AI agents have graduated beyond POCs to reshape how Pharma commercial teams plan, act, and adapt. But isolated agents only go so far. The real breakthrough happens when agents collaborate: a targeting agent feeds an omnichannel orchestration agent, which informs a field coaching agent, all aligned to a shared commercial objective.

In this hands-on workshop, participants will work through a realistic end-to-end commercial scenario from opportunity identification through pull-through execution, mapping where agentic collaboration unlocks the most value and where it breaks down. Participants will leave with a practical framework for designing multi-agent workflows, identifying the right human-in-the-loop checkpoints, and building the organizational buy-in needed to deploy at scale.

Session is designed for Business leaders, Analytics leaders, commercial insights teams, and AI practitioners.

Despite the promise of AI, commercial teams experience responses that feel like a used car salesman – confident, persuasive, and subtly wrong. How do you generate answers you can trust to drive key business decisions? This hands on workshop explores how contextual data built for life sciences and multi-agent orchestration can work together to reduce hallucinations, improve precision, and be transparent on uncertainty. Through commercial use cases such as understanding brand performance, stress-testing scenarios, and customer insights, we’ll explore how to get trustworthy answers, make the limits of AI transparent, and keep humans in control.

Eden Wells, Chief Insights and Decision Science Officer, Novartis U.S., is the keynote speaker at this year’s PMSA annual conference. Eden is widely recognized for her visionary leadership in advancing AI and technology and systemizing data-driven decision-making within the pharmaceutical industry. Eden’s keynote will address topics such as leveraging Talent, Technology, and Process Systemization to elevate decision-making across the enterprise, breaking down silos and unlock innovation-powered business excellence.

As pharmaceutical organizations expand the use of AI across the drug development lifecycle, the demand for governed, interoperable data platforms becomes paramount. From early discovery and clinical trials to real-world evidence generation, enabling AI requires seamless integration of diverse data sources while maintaining compliance, traceability, and regulatory alignment.

This session will outline strategies for designing modern data platforms that foster innovation while upholding robust data governance. Key challenges addressed include fragmented data landscapes, evolving protocols, and the integration of multi-modal data (clinical, genomic, imaging, and real-world data) to support scalable AI applications in drug development.

Central to the discussion is the ANCHOR framework:

• A – Accessible & Interoperable: Unified, standards-driven data access (e.g., CDISC, FHIR)
• N – Normalized & Contextualized: Analysis-ready, context-rich datasets
• C – Compliant & Controlled: Built-in governance, lineage, and regulatory adherence
• H – Harmonized Integration: Cohesive multi-modal data ecosystem
• O – Operationalized AI: Scalable, reproducible AI/ML deployment
• R – Responsive Architecture: Modular, adaptable to protocol and data changes

Attendees will receive a practical blueprint for building AI-ready platforms with governance embedded by design—empowering faster, compliant, and innovation-driven drug development.

More information coming soon

Medical Affairs teams generate a wealth of insights from unstructured data but turning these into timely, consistent, and actionable intelligence remains a challenge. Manual processes are slow, resource-intensive, and often limit the ability to act when it matters most.

This session explores a pragmatic shift: an SME–AI collaboration model that blends AI-driven synthesis with expert validation. By applying LLM-based workflows to unify and analyze diverse insight sources, teams can rapidly identify emerging themes and signals while maintaining scientific rigor through expert-in-the-loop review.

Rather than replacing human expertise, this approach redefines it, freeing SMEs from manual summarization and enabling them to focus on interpretation, context, and decision-making. Early adoption shows meaningful improvements in speed, consistency, and transparency of insight generation. Know more about how to operationalize Human–AI collaboration to unlock faster, more reliable medical insights.

Pharmaceutical sales representatives generate thousands of field execution signals daily - yet most incentive compensation (IC) plans continue to reward lagging sales outcomes rather than the leading behaviors that drive them. High call volumes don't guarantee strong performance, and traditional IC metrics rarely capture execution quality.

This session introduces a conceptual framework that uses AI to bridge that gap. By identifying which rep activities, sequencing patterns, and engagement approaches correlate with downstream commercial outcomes, the framework embeds those insights directly into IC plan design.

Three integrated layers form the foundation: a behavior-performance attribution model that distinguishes high-impact activities from low-value signals; an AI-generated impact scoring methodology that quantifies the commercial contribution of specific behaviors; and a governance structure that ensures transparency, explainability, and auditability - critical for sales leadership trust and compliance.

The result is a structured approach to shifting compensation toward the behaviors that demonstrably matter: the right customer, at the right time, with the right message. Better alignment between controllable behaviors and incentive design means improved rep focus, reduced wasted activity, and more consistent commercial execution.

As pharma organizations push for sharper, faster, and more localized decisions, forecasting can no longer remain a top-down planning exercise. This session will explore how advanced machine learning and deep learning are reshaping subnational forecasting—moving it from static target-setting to a dynamic decision intelligence capability. We will discuss how next-generation forecasting approaches can detect local demand shifts earlier, identify which geographies, accounts, or HCPs should be prioritized, flag where sales are outperforming or falling short of expectations, and determine what execution changes are required to deliver on plan.

Beyond forecast accuracy, the session will focus on the broader value of cutting-edge ML/DL enabled forecasting in improving agility, reducing planning bias, and creating actionable alignment across field, analytics, and marketing teams. It will also examine how these approaches can better handle sparse and cold-start markets while preserving consistency with brand-level strategy.

Finally, the discussion will look ahead to the next frontier: moving from brand-by-brand model building toward reusable, adaptive forecasting systems that can learn from prior implementations and rapidly generalize to new brands, refresh cycles, and business contexts with far less retraining effort.

Commercial analytics has spent a decade optimizing the pull model. Better dashboards. Faster queries. Conversational search. Each generation made it easier to ask questions, but someone still had to ask. That assumption has become the bottleneck, for both field leaders and home office teams inundated with reports.

At Bayer, we believe the next frontier is push. Role-differentiated intelligence delivered proactively. AVPs receive area-level briefs, AGMs receive territory diagnostics, reps receive HCP-level priorities, and home office leadership receives synthesized signals instead of chasing answers across dashboards.

Getting there is an evolution, not a deployment. Our journey with Tellius started by treating AI as role-specific interns: agents trained on narrow contexts, with every output human-reviewed. As the interns proved themselves, they graduated into experts. Certified, trusted, and shared across the organization as AI buddies that extend individual expertise to broader teams.

This session walks through where Bayer is on that journey, what each phase required, and why push-based intelligence is the inflection point that will reshape how commercial teams operate in pharma.

This session explores how enterprises can move from fragmented AI experiments to scalable, production-grade AI through an “AI Factory” approach. Using a real-world case study, we demonstrate how a structured, reusable ecosystem of components - spanning knowledge layers, agent orchestration, and AgentOps, enables rapid development of agentic AI solutions from idea to MVP in weeks.

Attendees will learn how unstructured and structured data can be transformed into AI-ready knowledge using ontologies, knowledge graphs, and vector databases, forming the backbone for intelligent agents. The session also highlights how reusable assets, standardized workflows, and governance frameworks drive speed, consistency, and cost efficiency while ensuring trust, safety, and compliance.

Through practical examples across market research, medical, and commercial use cases, we will showcase how organizations can industrialize AI, improve productivity, and accelerate business decision-making.

The session is designed for leaders and practitioners looking to operationalize GenAI and agentic systems at scale, turning AI from isolated pilots into a sustainable enterprise capability.

Launch windows are short. Analytical backlogs are not.This session explores how a conversational AI layer built on structured healthcare claims data is changing the pace of launch analytics. Traditional approaches — custom SQL, manual cohort construction, static dashboards — create 4-6 week turnaround cycles at the exact moment speed matters most. We'll walk through how generative AI translates plain-language business questions into validated analytical workflows: automated cohort construction, treatment line classification using longitudinal lookback, multi-dimensional prescriber and patient stratification, and iterative hypothesis refinement — all with built-in statistical validation against published prevalence benchmarks.

Attendees will leave with a clear picture of where AI fits in the launch analytics stack, what it takes to implement responsibly with healthcare data, and how teams are using it to answer the questions that matter — while the launch is still happening.

This moderated leadership roundtable explores how Data, Talent, and AI truly converge to deliver measurable commercial impact in pharma. We will share practical lessons learned from three real world applications: an AI-powered Bot for vaccine market insights, a Return on Data Dashboard that quantifies the true business value of third-party data assets, and a three-tier Intelligent Care-Gap Signal Service that surfaces actionable opportunities from unstructured data.

We’ll share the common solution frameworks and methodologies that connect these use cases, then dive into our AI-driven data engineering approach that dramatically accelerates data completeness and velocity while reducing cost and improving quality.

The session is highly interactive, with open Q&A and discussion prompts designed to let attendees’ helicopter between strategic business outcomes and technical execution details based on their specific interests.

Why did Brand X lose 3 points of share in the Southeast, and how much of it is a formulary access problem versus a field execution problem?" That question today gets split across two teams, four data sources, and a three-week turnaround. AI can now answer it before the plates are cleared — delivering a finished board deck, a territory coaching summary, or a payer escalation brief, not a chart your team has to rebuild. And while you're eating, it keeps monitoring your business for the next shift. We'll prove it with a live demo over lunch: a real multi-source commercial question, answered end-to-end, with finished work products as the output. Then we'll open the table for discussion: what would change about how your team operates if questions like that took minutes instead of weeks?

Everyone wants you to “think outside the box.” But to truly give yourself a competitive advantage and lead in the business world, you must approach problem-solving to invent an entirely new box. Welcome David Corsaro. For the past 25 years, David has been the "go-to" guy in the business world regarding creativity and problem-solving. Additionally, David is an accomplished magician and mentalist, performing in 4 different off-Broadway magic shows in NYC and he even appeared on the hit TV show “Penn & Teller: Fool Us”....and yes, he fooled them. In David’s seminar, “Problem-Solving like a Magician”, you will not only learn how a piece of magic works but more importantly you will learn WHY it works and how you can take the principles of psychology and neuroscience and apply them to spark creativity in you and your team members.

Enterprise-wide AI adoption requires more than technical feasibility. It requires understanding people, their attitudes and behaviors, and how they work. Analytics teams are building increasingly sophisticated solutions, but adoption remains the bottleneck. When stakeholders across sales, marketing, finance, and operations respond to the same AI initiative with vastly different levels of enthusiasm, skepticism, or indifference, a one-size-fits-all rollout strategy won't cut it. Understanding these differences in needs, expectations, mindsets, concerns, and readiness across functions, roles, and fluency levels creates opportunities for targeted communication and change management strategies.

Through mixed-methods research including in-depth qualitative interviews and structured surveys with commercial leaders across multiple biopharma organizations, we developed a behavioral segmentation framework that categorizes stakeholders by their orientation toward AI, capturing dimensions like enthusiasm, data fluency, risk tolerance, and decision-making criteria. We've identified five distinct segments, each with unique barriers and enablers that inform how to tailor engagement strategies for different audiences within the commercial organization.

In this interactive session, we'll walk through the framework, share what's worked and what hasn't in practice, and provide analytics teams with a practical playbook for building organizational buy-in. Live polling will let you benchmark your organization's adoption landscape against peers in the room.

This session shares a practical blueprint for building Enterprise AI in pharma and scaling beyond pilots. We introduce a three-layer model: (1) AI for intelligent engagement (patient/HCP and field-facing assistants), (2) AI for operational excellence (automating analytics workflows for productivity), and (3) a scalable foundation—the platform layer that enables reuse and trust through context engineering, orchestration, integration, and governance.

We cover different “where to start” paths—engagement-first, productivity-first, or foundation-first—based on an organization’s maturity and constraints, and discuss the trade-offs using common pharma scenarios. Attendees leave with clear sequencing guidance and a roadmap from MVP to scale.

GEO represents a critical new discipline for pharma, merging scientific rigor, metadata strategy, content engineering, and AI governance. Companies that adopt GEO early will shape how their science is understood in a generativeAI-first world, enhance AI-driven medical insights, and build competitive advantage across commercial and clinical domain. The session will focus on real world applications of this in the healthcare ecosystems and key ways in which brands can leverage data and insights to boost their presence within the constantly evolving ecosystem.

This session explores the transition from "Flashy Experimentation" to "Disciplined Engineering" through the implementation of Agentic AI within the theme of Advanced Machine Learning & Deep Learning. Drawing on parallels from large-scale deployments such as agentic voice ordering systems that must handle high-concurrency, real-time reasoning and the safety guardrails developed for frontier models like Gemini, we will discuss how to architect a "Cognitive Engine" for pharma.

AI in pharma is at a turning point, where pilots are no longer enough and real impact is the expectation. This panel dives into what works when scaling AI, from proving value fast to building talent that bridges science, data, and business. Panelists will share how leading organizations are creating a culture where AI empowers people and drives smarter decisions. The conversation will also tackle where true competitive advantage lies and how to strike the right balance between AI autonomy and human judgment.

Abhishek N. Singh will share how leading pharma organizations are rethinking analytics and decision‑making at the convergence of data, talent and AI. Drawing on Merck’s journey, he will explore how strong analytical talent, modern data foundations and applied AI come together to drive meaningful business impact—while keeping human judgment, accountability and outcomes at the center.

More information coming soon

Healthcare media planning still treats HCP and consumer audiences as separate worlds—despite the reality that clinical decisions and patient behavior are deeply interconnected. This session introduces a data science framework that bridges that gap by identifying and activating overlapping HCP–consumer audiences using privacy-safe modeling and human insight.

Instead of relying on fragile individual-level linkage, we focus on cohort-level alignment—where provider intent and patient behavior converge across therapeutic areas, geography, and time. By combining AI and deterministic modeling, we enable planners to prioritize high-impact overlaps and move from channel optimization to true cross-audience orchestration. We will share a framework which is dynamic and privacy-compliant, re-framing audience overlap from a reporting metric into a strategic planning lever.

We will also discuss how data science and clinical expertise jointly shape model design, ensuring outputs are not just accurate—but meaningful and actionable. Real-world results will also be shared, demonstrating how campaigns targeting high-overlap cohorts outperform traditional segmented approaches, driving stronger engagement and better outcomes.

Attendees will leave with a practical roadmap for implementation with a new lens on how to build adaptive, healthcare media systems that reflect how decisions actually happen.

This session presents a sophisticated Agentic AI architecture designed to unify structured databases and unstructured document repositories within a single, enterprise-ready intelligence framework. It demonstrates how a modular dual-agent design enables organizations to process natural-language questions through parallel pathways, delivering both real-time quantitative insights and rich qualitative or clinical context.

At the core of the approach is a semantic layer, supported by a scalable knowledge graph, which serves as the orchestration engine for interpreting queries, routing them to the appropriate data source, and generating accurate, relevant responses. The session also emphasizes the importance of enterprise-grade governance, including traceability, source attribution, security, and controlled access, to ensure that AI-generated insights remain trustworthy, verifiable, and aligned with organizational standards.

By combining semantic intelligence, modular AI agents, and governed data access, this presentation illustrates how human expertise and artificial intelligence can work together to accelerate decision-making, reduce manual effort, improve insight quality, and lower operational complexity. Attendees will gain a practical understanding of how this architecture can modernize enterprise knowledge discovery and decision support at scale.

More information coming soon

Marketing teams regularly collaborate with Commercial Insights to run budget scenarios that optimize ROI spend. Historical processes have been manual, fragmented, reliant on external support, with limited transparency, resulting in slow cycles, low reusability, and inconsistent auditability. This session explores how a successful, scalable, standardized platform-based setup is helping streamline Commercial Spend Optimization.

This presentation outlines how commercial pharmaceutical organizations can modernize complex data operations using a scalable Agentic AI–driven automation framework. It demonstrates how combining deterministic automation with AI-based decisioning enables autonomous data ingestion, quality control, anomaly detection, and insight generation across diverse commercial data sources. The session shares a practical, end‑to‑end methodology and real-world architecture that reduces reporting cycle times, improves data reliability, and accelerates time‑to‑insight—equipping attendees with a repeatable blueprint for building always‑on, AI‑enabled commercial data ecosystems.

As the pharmaceutical landscape becomes more complex, organizations need faster and more reliable ways to understand GTN risk. This session will highlight how AI/ML enabled forecasting can surface early signals of emerging risks and revenue shifts, helping teams respond more proactively to market changes. By combining advanced analytics with business understanding, the approach strengthens rebate strategy and supports more informed decision making across finance, commercial, and market access functions.

How can life science teams turn thousands of congress outputs into timely, actionable insight? Keeping pace with the impact of new clinical data poses a challenge to Commercial and Medical Affairs. This presentation introduces an AI-enabled, human-guided framework that accelerates scientific intelligence 5x faster than traditional methods while preserving analytical rigor.

At the 2025 American College of Cardiology (ACC) Annual Scientific Session, more than 4,000 abstracts and a surge of professional social media created an overwhelming volume of information. We’ll showcase how this framework cuts through the noise to rapidly identify what matters most.

You’ll learn how combining automated text analysis, large language models, and expert oversight from Medical Affairs, Commercial, and Data Science teams enables faster decision making with greater than 90% confidence in thematic and sentiment analysis than human alone. Rather than replacing expertise, AI enhances it by accelerating pattern detection and insight generation at scale. We’ll also share what is next in integrating claims analysis and the future of impact measurements.

We’ll highlight key trends generated from ACC, including rising focus on lipoprotein(a), expanding cardiovascular relevance of GLP-1 therapies, and increased attention on ATTR-CM and pulmonary hypertension innovation.

Join this session to discover a practical, scalable approach to modernizing intelligence while keeping experts firmly in control.

Join us for an inspiring Analytics Luncheon, designed by members of the PMSA Women in Analytics program team, where leaders of all genders explore how data, talent, and AI intersect to shape the future of business. This year’s theme, “Leading with Impact in the Age of AI: Visibility, Voice, and Influence,” celebrates the contributions of female leaders while engaging everyone in the conversation.

The session begins with a brief presentation by Tamara Burzinski (Novartis) to set the stage, followed by facilitated table discussions, where participants will:

• Hear success stories from leaders elevating women colleagues
• Explore strategies to cultivate influence and secure a seat at the table
• Learn how leaders are staying ahead in technology and AI
• Discover opportunities for career growth and talent development

This luncheon celebrates leadership in analytics and encourages all attendees to amplify their visibility, voice, and impact in the age of AI.

This panel will discuss an evolution from analytics to Agentic and Gen AI, areas where analytics might be enhanced or even replaced by Agentic/Gen AI, and areas where AI is still has limitations.

More information coming soon

Commercialization in COPD is becoming increasingly complex, with fragmented patient journeys, evolving treatment pathways, and shifting access dynamics. Traditional approaches often struggle to fully capture this variability. This session explores the use of Digital Twin technology to model the COPD ecosystem by creating dynamic representations of patients, physicians, payers, and market interactions. By integrating real-world data with simulation techniques, teams can examine how different factors such as access changes, treatment patterns, or competitive activity may influence outcomes.

We will discuss how this approach supports more informed planning across areas like targeting, launch sequencing, and market access, while enabling a more forward-looking view compared to static analytics. Join us to know more about how Digital Twins can support more adaptive and data-driven decision-making in COPD.

In this session we will discuss the power of integrated healthcare behavioral data with consumer elements aligned to HCPs to provide an expanded insight into how consumer characteristics (of the HCP) predict and drive prescribing behavior. This session will examine how these techniques can be applied to better understand behavior and better align targeting and segmentation efforts.

Synthetic data has an image problem. It is often dismissed as “fake data” because it fails to capture the nuanced realities of medicine. For synthetic data to be viable, it must be clinically sound. This means capturing complex dynamics, such as the fourfold prevalence of migraines in females compared with males, or the increased odds linking breast cancer and ovarian cancer when a patient is ER‑negative as opposed to ER‑positive. Without this clinical integrity, the “garbage in, garbage out” principle renders the data useless for any medical or commercial inquiry.

One more thing matters: novelty. Synthetic data must bring something new; it cannot merely be a de‑identified echo of existing real‑world data with minor perturbations. Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) are powerful for generating individual records but struggle when it comes to generating longitudinal time series. It is no secret that TimeGANs and sequential VAEs fall short. There is good news, though. Variations on operator‑based approaches, as exemplified by Genetic Algorithms, prove to be well suited to this task.

In this talk, we will motivate our foray into synthetic data and discuss our findings and lessons learned. In particular, a purely rule‑based approach is a nonstarter, while the availability of real‑world data for the drug of interest—or for a suitable surrogate—is extremely helpful. We will also describe use cases that lend themselves to synthetic data. Like anything else, synthetic data is not a panacea—it cannot do alerts. So, do not throw the baby out with the bathwater.

Pharmaceutical Drug launches frequently face unexpected market access barriers such as restrictive formulary placement, prior authorization requirements, or step edits, which delay patient access and negatively impact early brand performance. While access challenges are often analyzed post-launch, there is a critical need for a structured, data-driven approach to anticipate access risks before launch to support proactive pricing, evidence, and payer engagement strategies.

Learning Objectives

1. Understand how historical payer behavior and access outcomes from prior launches can be used to anticipate market access barriers before launch
2. Learn how predictive access risk indicators can be constructed using historical launch analogs and payer policy signals
3. Recognize how early access risk identification can inform launch sequencing, contracting, and evidence-generation strategies

Pharma commercial teams rely on BI dashboards for brand performance, targeting, and field execution decisions, yet extracting actionable insights is often slow and error‑prone. Users must navigate multiple views, interpret complex visuals, and stitch answers across pages, increasing cognitive load and delaying decisions—especially when questions extend beyond pre‑built dashboards. These inefficiencies can lead to missed opportunities, slower responses to market changes, and inconsistent interpretation across teams.

This session introduces a GenAI‑powered conversational BI platform that transforms static dashboards into interactive insight engines. Users can ask questions in natural language and receive immediate, governed answers with KPIs, trend narratives, visual summaries, and transparent assumptions. Unlike typical GenAI approaches, governance and auditability are embedded by design, ensuring every insight is traceable to executed queries and aligned with approved business logic. This delivers both speed and trust—critical requirements for pharma analytics.

A key innovation is semantic onboarding: an AI‑assisted workflow that builds a shared business vocabulary across dashboards and data sources, validated by analysts for accuracy and compliance. This enables faster dashboard onboarding and consistent interpretation of user intent. An admin panel monitors usage, reliability, and feedback for continuous improvement.

In evaluation, the assistant achieved over 90% accuracy and reduced query resolution time by ~30%, enabling faster scenario exploration and more time spent acting on insights. Attendees will learn how to operationalize governed conversational BI to accelerate insight‑to‑action cycles across commercial teams.

Despite significant investment in AI, many pharmaceutical organizations have yet to realize its full potential in forecasting — missing opportunities for dramatic reductions in effort and cost. The barrier is often a perceived tradeoff: that adopting AI means sacrificing the granular, epidemiology-based insights that commercial teams depend on. This session challenges that assumption, showing how agentic AI and workflow automation can be embedded within proven epi-based methodologies to deliver forecasts that are faster, more reliable, and analytically richer.

Production-grade, AI-enabled forecasting solutions are transforming how organizations approach model development, scenario planning, insight generation, and stakeholder alignment — combining intuitive self-service workflows with advanced analytics into a seamless end-to-end process. The session will highlight how AI can surface and quantify the impact of external market signals, generate actionable recommendations, and maintain the transparency and explainability that established forecasting frameworks support.

Critically, we will also examine the organizational and behavioral shifts necessary for successful adoption, including change enablement strategies and the evolution of the forecaster's role in an AI-augmented environment.

Forecasting leads, commercial analytics professionals, and their business partners will gain practical insights into how pharmaceutical organizations can:

• Modernize forecasting to significantly reduce effort while improving accuracy and business impact
• Elevate the role of forecasting in strategic decision-making

Transformer-based AI is redefining how pharmaceutical teams identify and engage patients earlier in their treatment journey. While much of the industry is focused on traditional machine learning or LLM-based approaches, this session highlights a different path: purpose-built transformer models trained directly on longitudinal real-world data to predict near-term disease progression and drive actionable commercial strategy.

Using a real-world case study in uveal melanoma, we demonstrate how Prospection partnered with Immunocore to predict which patients were likely to metastasize in the next 3–6 months—enabling earlier, more precise engagement with treating physicians. Unlike traditional models that rely on predefined data features, these transformer architectures learn directly from sequences of clinical events, automatically capturing timing, context, and complex patterns across patient journeys—delivering stronger predictive performance.

The result is stronger alignment between commercial activity and patient need—demonstrating greater physician engagement, improved access to the right patients at the right time, and ultimately the potential for increased market share.

Beyond the case study, we explore how these models can be rapidly retrained across therapy areas and data sources, unlocking scalable, high-impact predictive capabilities for commercial teams.

In the sesssion we will learn how Pharma is using AI to streamline patient starts. In case example we will talk about how a rare disease product leveraged AI to identify patients needing more support to get started on therapy

Multi-Touch Attribution is increasingly seen as a critical capability for understanding omnichannel effectiveness in complex healthcare ecosystems. However, many organizations struggle to move beyond pilot-stage measurement into scalable, real-world application. This session will present a practical, end-to-end approach to designing, implementing, and operationalizing MTA using integrated data sources such as claims, digital engagement, and field interactions. Attendees will gain insights into key modeling considerations, including handling data limitations, selecting appropriate techniques, and aligning MTA outputs with broader measurement frameworks.

Beyond modeling, the session will focus on how to embed MTA into commercial workflows while enabling brand, omnichannel, and field teams to translate insights into actionable decisions. It will also highlight the importance of governance, cross-functional collaboration, and continuous feedback loops to drive sustained adoption. By combining AI-driven analytics with human expertise, this session provides a practical playbook for moving from attribution insights to measurable business impact.

More information coming soon

Weighted Statistical Methods for Novel Feature Selection in Claims-Based Predictive Models: Feature selection is the cornerstone of effective predictive modeling in Real-World Evidence studies, yet conventional approaches—mutual information and p-values—struggle at the scale and sparsity of administrative claims data. Mutual information surfaces rare, high-signal features that lack prevalence in the target population, while p-values lose discriminatory power in the massive sample sizes these datasets demand.

This presentation introduces a prevalence-aware weighting framework comprising four composite metrics—two uni-directional and two bi-directional—that embed class prevalence ratios directly into statistical evaluation. Uni-directional metrics isolate features uniquely predictive of the target class; bi-directional metrics identify features that maximally separate both classes. Together, they deliver a more statistically robust feature set than either MI or p-values alone.

Attendees will see the framework validated through a real-world case study identifying undiagnosed MASH patients, where the model achieved 89% recall and surfaced non-obvious clinical predictors—neurological comorbidities, cardiac disease, antidepressant use—that would traditionally require extensive Key Opinion Leader review.

Key Takeaways:

• Why standard MI and p-values underperform at claims-data scale
• How prevalence-weighted metrics correct these shortcomings
• Practical formulas ready for immediate implementation
• Demonstrated reduction in manual KOL dependency
• A validated, reproducible approach to accelerating feature selection in binary classification for RWE studies

This session presents an Incentive Compensation Digital Twin built as a governed analytical replica of plan rules, performance drivers, and key constraints. It is designed for self-serve use by Home Office teams to compare plan options and incentive add-ons before deployment, then calibrate plan structure against benchmark archetypes. Agentic artificial intelligence powers the framework end-to-end: it retrieves approved plan logic and industry benchmark references, runs standardized scenario tests, and drafts a concise narrative that highlights trade-offs and assumptions in plain language. The goal is consistency in how plan decisions are evaluated and explained across stakeholders, including sales operations, finance, and commercial leadership, so decisions are not influenced by individual analyst style or institutional memory.

For pharmaceutical organizations, the practical value lies in speed and consistency across design cycles and launch windows, along with reduced manual effort spent on scenario comparison. Using benchmarks from comparable industry peers, teams typically target 30% faster plan design and approval cycles. They also target a 40% reduction in manual analyst effort. By surfacing fairness, stability, and budget risks earlier, teams can reduce post-rollout rework cycles and disputes by 25%. This agentic AI–powered twin accelerates the entire insight-to-decision value chain, enabling more motivating and impactful incentive strategies, strengthening field trust, and improving responsiveness across go-to-market execution, including stronger support for launch success.

In healthcare analytics, a key limitation is not model performance but incomplete data. Relying on incomplete claims is like navigating a city with a map missing critical streets. Fragmentation between adjudicated closed claims and near real-time open claims creates blind spots in longitudinal patient journeys. These gaps can obscure important clinical events and introduce bias into downstream analytics, particularly in oncology.

This session introduces neural linkage, a practical approach for reconstructing the invisible patient journey when key clinical events are missing. Neural linkage learns patterns from high‑fidelity closed claims and applies those patterns to fragmented open claims. Patient journeys are modeled as ordered sequences of clinical events, allowing sequence-based models to infer missing events based on surrounding clinical context and cohort-level patterns.

The presentation will describe how masked sequence models are trained to simulate real-world data fragmentation and how inferred events are evaluated against held-out closed claims data. A working prototype will demonstrate reconstructed patient timelines along with confidence signals associated with each inferred event. Attendees will gain a clear understanding of how neural linkage methods can be applied to improve the completeness, interpretability, and reliability of patient-level analytics when claims data is incomplete.

Closing Novo Nordisk's Monday Morning Gap: How We Reduced Field Analytics Latency from 10 to 2 Days.

Forecasting teams today support an expanding breadth of assets—often with multiple indications—and increasingly rapid planning cycles. Yet forecast model design remains largely manual, inconsistent, and time intensive. Reliance on legacy templates and individual judgment leads to long build times, variability across assets, and limited scalability, creating a material bottleneck as organizations move from brand level forecasts toward portfolio and enterprise wide decision support.

This session introduces a smartly automated and AI driven forecasting model builder that automates the creation of fit for purpose models tailored to a product’s scientific profile, therapeutic context, market dynamics, and lifecycle stage—while preserving the familiarity and flexibility of spreadsheet. The approach replaces one size fits all templates with intelligent, context aware model architecture and enables secure reuse of common patient and market structures across similar indications.

By combining automated model generation, pre filled assumptions from curated sources, and human in the loop expert refinement, the solution transforms model design from a manual craft into a scalable, consistent capability. In practice, build times shrink from weeks to days, governance is strengthened, and teams can focus on what matters most: assumptions, scenarios, and the insights that drive strategic decisions.

This panel brings together medical leaders to explore the role of Medical Affairs within a modern engagement ecosystem. The discussion examines how omnichannel engagement is being used to close clinical care gaps, accelerate evidence dissemination, and enable high‑value scientific exchange, while shaping the pre‑launch scientific landscape across field medical and digital channels. Panelists will highlight where misalignment between medical and commercial engagement surfaces for healthcare professionals, and what it takes to design experiences that work seamlessly across both contexts. The session also traces the evolution of medical engagement and personalization, from early models to today’s state of practice, and looks ahead to how emerging agentic capabilities are poised to redefine the future of Medical Affairs.

Predictive modeling in clinical and real‐world settings is often challenged by small sample sizes and severe class imbalance, particularly in rare disease research. These constraints can substantially degrade model performance, limiting the practical utility of machine learning approaches. Synthetic data (SD) are artificially generated datasets designed to preserve key statistical properties and multivariate relationships present in real‐world data (RWD). In healthcare applications, SD can augment patient‐level records, mitigate class imbalance, and expand representation of underrepresented populations. Generative adversarial networks (GANs) are deep learning architectures composed of a generator and a discriminator trained in an adversarial framework. This study evaluated the effectiveness of GAN‐based synthetic data augmentation for improving predictive performance in rare and ultra‐rare disease modeling scenarios characterized by low prevalence and limited sample sizes. Across three case studies, GAN‐based synthetic data augmentation consistently enhanced minority class detection and outperformed traditional class imbalance mitigation techniques. These findings demonstrate that generative models can meaningfully strengthen predictive modeling in rare disease contexts and extend the applicability of advanced analytics in data‐constrained healthcare environments.

Background: For patients living with chronic, progressive conditions such as Spinal Muscular Atrophy (SMA), sustained treatment adherence is essential to maintaining therapeutic benefit, slowing disease progression, and preserving quality of life. Over time, adherence may be compromised by evolving challenges, including treatment fatigue, complex dosing regimens, caregiver burden, access barriers, and changing personal or social circumstances.

Traditional adherence monitoring approaches are largely retrospective, relying on lagging indicators such as historical proportion of days covered (PDC) or static rule-based thresholds. These methods often detect non-adherence only after it has occurred, limiting opportunities for proactive intervention.

Analyzing longitudinal, high-dimensional real-world data enables earlier identification of subtle behavioral changes that precede clinically meaningful adherence decline, which are often invisible to rule-based systems but consistently precede real-world adherence deterioration.

Adherence is strongly influenced by temporal dynamics, where longitudinal patterns such as refill regularity and gradual behavioral drift often signal clinically meaningful decline, as well as by patient- and disease-level characteristics, including demographics, disease severity, and comorbidities, which also drive heterogeneity in adherence trajectories.

The complex interaction between these factors makes simple analytical approaches insufficient in real-world settings. Stratifying patients across all these factors becomes essential in deriving meaningful and accurate insights. Equally important is ensuring that model outputs are relevant for clinicians, patient support teams, and field partners – requiring explainability methods capable of translating complex temporal patterns into actions.

We propose an AI-driven adherence risk prediction framework that models whether a patient’s adherence is likely to fall below a threshold within a predefined future time horizon. This will help us to understand key drivers of Adherence Risk and predict the Adherence Risk in real time.

This will help in:

1. Integrate into commercial workflows for field-based interventions strategies as part of the Next Best Action portfolio, enabling personalization, proactive outreach, and improved HCP coordination.
2. Integrate outputs into patient support programs and care management to improve patient outcomes.
3. Support continuous learning to refine adherence programs and improve long-term patient outcomes.

In this poster, we highlight a focused case study illustrating the application of the proposed approach in patients treated for SMA (Spinal Muscular Atrophy)

This poster presents a scalable framework for detecting Fraud, Waste, and Abuse (FWA) in longitudinal healthcare claims using unsupervised outlier detection. Given limited labeled data and high provider variability, traditional rule-based approaches often miss emerging or subtle anomalies.

We apply and compare multiple methods, including ECOD, Isolation Forest, and Local Outlier Factor, on HCP-level features derived from multi-year claims. Features capture prescribing intensity, patient mix, channel utilization, and temporal patterns. Longitudinal feature engineering enables detection of behavioral drift and sudden deviations over time.

An ensemble approach improves robustness across methods, while explainability techniques provide transparency into key drivers of anomalies. Results show that different models capture complementary patterns and combining them enhances signal quality while reducing false positives.

This framework supports real-world deployment by integrating with analytics platforms and enabling human-in-the-loop validation.

Business Problem: Pharmaceutical Drug launches frequently face unexpected market access barriers such as restrictive formulary placement, prior authorization requirements, or step edits, which delay patient access and negatively impact early brand performance. While access challenges are often analyzed post-launch, there is a critical need for a structured, data-driven approach to anticipate access risks before launch to support proactive pricing, evidence, and payer engagement strategies.

Learning Objectives:

• Understand how historical payer behavior and access outcomes from prior launches can be used to anticipate market access barriers before launch.
• Learn how predictive access risk indicators can be constructed using historical launch analogs and payer policy signals.
• Recognize how early access risk identification can inform launch sequencing, contracting, and evidence-generation strategies.

Feature Accelerator is a top-down feature engineering framework designed to transform complete claims histories—diagnoses, procedures, pharmacy dispensing, and care patterns—into rich, temporally aware representations of patient journeys. Traditional, hypothesis-driven methods focus on a small set of predefined clinical signals and require weeks to produce fewer than 100 hand-crafted features. By contrast, Feature Accelerator systematically encodes event presence, timing, frequency, recency, first occurrences, temporal change, and statistical summaries, generating expansive feature sets (typically 8,000–10,000 features per cohort) within days. Because it does not rely on prespecified clinical filtering, the framework captures broader constellations of longitudinal events, enabling state-of-the-art machine learning models to detect informative, non-obvious signals across disease areas and clinical contexts. This holistic approach increases the informational yield of claims data and supports diverse predictive modeling tasks—patient segmentation, market assessment, identification of rare-disease unmet needs, and next-best-action recommendations—thereby improving the data foundation for commercial pharmaceutical decision-making.

This poster introduces IRIS, an agentic AI framework that moves beyond simple anomaly detection to deliver contextual, explainable market intelligence.

Rather than just flagging deviations, IRIS contextualizes them by combining quantitative trends with qualitative, correlative factors - such as payer dynamics, channel shifts, prescriber behavior, and geographic patterns. It explains not only what changed, but why it changed, by uncovering the key drivers behind each deviation.

The analysis follows a top-down approach, starting at the national level and progressively drilling into channels, PBMs, and payers. It also groups related signals and dimensions into a unified view, enabling faster and more structured exploration.

IRIS simplifies the analyst’s workflow by providing a guided walkthrough of the most critical signals to focus on as soon as data is refreshed. It further enhances exploration through an AI Assist, allowing users to deep dive into ad hoc questions and uncover insights on demand.

Through curated deep dives and guided storytelling, IRIS connects signals across dimensions to build a clear narrative - from what happened to why it happened and what to do next - driving faster, more informed decision-making.

Life sciences organizations are increasingly using predictive and prescriptive AI to anticipate market trends, optimize commercial engagement, and improve patient access. However, the success of these models depends on strong data foundations, including harmonized master data, analytics-ready architectures, and transparent delivery. Without these, even advanced AI struggles to gain trust and drive impact.

This presentation highlights how modern commercial data platforms enable scalable AI by integrating master data management and analytics-ready data marts. Two use cases demonstrate practical applications. In retail markets, a pre-call planning solution uses longitudinal prescribing data and behavioral signals to classify prescribers into segments like New Adopters, Growers, and Decliners. Embedded within CRM workflows, these insights help field teams tailor engagement strategies while maintaining alignment with enterprise reporting and compliance.

In rare disease settings, machine learning models predict providers most likely to complete enrollment forms using unified provider, patient, and access data. Ranked outputs with clear drivers allow teams to prioritize outreach, accelerating time to therapy and improving patient outcomes.

Overall, the session emphasizes that impactful AI requires not just advanced algorithms, but also strong data governance, human-centered design, and seamless integration into workflows to deliver actionable, trusted insights.

Radar Room is an AI-enabled Business Positioning System designed to help pharmaceutical commercial teams proactively detect territory performance risks and act earlier. Traditional quota-based tracking relies on retrospective KPIs and often misses emerging demand shifts, competitive dynamics, and access barriers influencing subnational outcomes.

Radar Room integrates hundreds of engineered features spanning patient eligibility and severity signals, HCP engagement behavior, promotional reach, sampling, competitive trends, and market access conditions. A Temporal Fusion Transformer (TFT) generates multi-horizon NBRx forecasts at national and territory levels, establishing forward-looking performance benchmarks relative to quota. Territories are dynamically classified into four segments: Champion, At Risk, High Risk, and Declining.

To translate predictions into action, LightGBM driver models combined with SHAP analysis surface territory-specific performance drivers, such as declining face-to-face reach, competitive preference shifts, sampling gaps, or access barriers. Monthly refresh cycles enable continuous learning and drift monitoring, improving signal timeliness and accuracy.

By combining advanced forecasting, granular territory modeling, and interpretable AI, Radar Room shifts organizations from reactive reporting to proactive commercial orchestration—supporting field prioritization, resource allocation, and cross-functional alignment.

A deep understanding of the patient journey is central to effective commercial strategy in pharmaceutical organizations. Insights into treatment pathways, decision drivers, switching behaviors, and real-world barriers directly help targeting, messaging strategy, forecasting, opportunity assessment, and cross functional alignment. However, traditional patient journey development requires extensive manual effort across primary market research and RWD (e.g., claims and EHR based) data engineering resulting in long cycle times and fragmented insights. In addition, there is higher analytical barrier-to-entry due to the need for foundational SQL and data science knowledge. This session introduces an AI enabled Patient Journey Analytics Framework and modular RWD-analytics engine that accelerates and strengthens journey development by unifying real world data (RWD), qualitative inputs, and machine learning driven behavioral analysis. The approach begins with building a foundational understanding informed by guidelines, epidemiology, and clinical literature, establishing hypotheses around disease dynamics and decision inflection points. Proprietary modular claims engine then rapidly processes large scale medical and pharmacy claims to map baseline journeys, identify treatment patterns, and surface therapy transitions and adherence behaviors. AI/ML based analytics then deepen insights into disease area specific questions, including barriers and drivers of treatment decisions, opportunity overlap, and predictive indicators of patient movement. These outputs are enriched through qualitative inputs such as patient charts, EMR reviews, and in-depth interviews captured efficiently through agent assisted research workflows.

Applications of AI in optimizing launch strategies, market access, and omnichannel engagement

This walk-through is a geo-level pseudo-experimental ROI analysis, measuring the performance of a machine learning triggering system, but the method can just as easily be applied to a geographically targeted advertising campaign that was executed without a pre-specified control group of geographies. I will focus on the aspects of the analysis separate from the choice of matching algorithm, which is a topic that has been adequately covered elsewhere, specifically (a) removing outliers (b) how to designate test and control groups of geographies eligible for matching without violating the Stable Unit Treatment Values Assumptions in causal Inference and (c) what measurement formulae to use when calculating an unbiased estimate of the average treatment effect.

Patient journey analytics serves as a critical foundation for pharmaceutical decision-making-informing strategies across forecasting, market access, and field execution and various other commercial as well as medical business use cases. Yet, constructing a medically accurate journey from real-world data (RWD) remains one of the most resource-intensive analytical challenges. Traditional approaches rely on manual code curation, iterative SQL development, and multi-stage expert review, often stretching timelines to 8-12 weeks. As oncology landscapes evolve through biomarker segmentation, combination therapies, and shifting standards of care, these linear and manual methods struggle to keep pace or scale effectively.

This session introduces an Agentic GenAI–driven framework that reimagines the patient journey creation process from the ground up. The Agentic Patient Journey framework exemplifies how human intelligence and artificial intelligence work symbiotically to elevate analytics in pharmaceutical decision-making. Instead of treating patient journey analytics as a single monolithic task, the framework decomposes it into a network of specialized GenAI agents - each responsible for a discrete domain such as market definition, cohort generation, therapy construction, line-of-therapy (LoT) logic, or insight creation. A central orchestrator agent governs execution, ensures traceability, and enables real-time collaboration between users and agents. The result: a first-pass patient journey delivered in 5-10 hours instead of multiple weeks - without compromising analytical depth or medical fidelity.

Reliable healthcare AI requires shifting from basic conversational responses toward an agentic architecture capable of navigating complex data schemas and applying healthcare-specific logic. In pharmaceutical analytics, the greatest barrier to adoption is a “Context Gap”, the disconnect between general model intelligence and specialized business rules. While Large Language Models are powerful engines, their performance in isolation remains insufficient for high-stakes, multi-step commercial workflows.

This study demonstrates that the solution lies in agentic performance, grounding AI in domain expertise via a specialized architecture. We evaluate Marmot™, an agentic AI platform developed by Komodo Health, using a benchmarking methodology designed to test end-to-end execution across use cases including market share, HCP/O targeting, and survival analysis.

Our results show that Marmot’s architectural layers yield significant improvements in accuracy and reliability over standalone foundational models. By closing the context gap, Marmot provides a blueprint for Commercial Excellence, offering the precision needed to optimize launch strategies, market access, omnichannel engagement, and more. This framework transforms AI into a dependable, scalable engine for competitive advantage in the life sciences.

Pharmaceutical commercial teams increasingly use omnichannel orchestration platforms, such as Next Best Engagement (NBE) engines, to personalize interactions with healthcare providers across channels, content, and timing. However, most current solutions are heavily manual, reliant on static rules, sequential workflows, and fragmented measurement. These limitations lead to long iteration cycles, limited adaptability, high operational burden, and slow response to changing business needs. Common challenges include manual A/B testing, poor traceability of historical executions, limited explainability, and delayed refreshes when data or objectives change.

This work introduces an agentic AI–driven redesign of omnichannel orchestration that shifts analytics from manual coordination to autonomous, continuously learning systems. Instead of incremental rule-based improvements, orchestration is re-architected as a network of intelligent agents that collaborate across the full analytics lifecycle with built-in transparency and governance. The framework consists of 18 core functional agents, supported by domain-specific sub-agents, spanning data preparation, feature engineering, modeling, execution, measurement, MLOps, deployment, and validation. Large Language Models integrated through Model Context Protocol enable advanced reasoning, automated debugging, configuration management, and natural-language interaction.

The system delivers three key user experiences: real-time dashboards, a conversational interface for non-technical users, and a governance layer for compliance and auditability. Results demonstrate substantial operational gains, including major reductions in debugging time, effort, and configuration cycles, while improving recommendation quality, measurement transparency, and scalability.

GLP-1–based therapies have transformed obesity treatment but face persistent challenges in access, affordability, and long-term adherence. Patient journeys are often fragmented, marked by delayed initiation and early discontinuation driven by clinical factors, cost burden, and social determinants of health (SDoH). Digital patient support programs (PSPs) — offering education, coaching, and financial assistance — aim to address these barriers, yet evidence on their real-world effectiveness has been limited. This study applied advanced analytics to integrated, tokenized program engagement and claims data to evaluate whether digital PSP engagement improved onboarding, treatment initiation, and therapy persistence among individuals prescribed anti-obesity medication. The analysis examined behavioral, clinical, and contextual drivers of milestone progression and characterized patient segments based on engagement patterns, barriers, and SDoH factors. Results revealed substantial heterogeneity in how patients access and persist on therapy, varying by clinical history, support timing, and contextual barriers — indicating that uniform support approaches are insufficient. PSP engagement was associated with an 8–12% improvement in 14-day prescription fulfillment and an extension of time on therapy by 6–22 days depending on services utilized. Identification of high-impact patient segments enabled formulation of targeted, segment-specific interventions and data-driven recommendations for optimizing program reach, engagement, and sustained treatment continuity.

Unified Patient Intelligence is a platform that solves a common commercial pharma problem: patient insights are scattered across social media, published literature, physician notes, and health websites, requiring weeks of manual research to synthesize. The platform lets commercial teams ask natural-language questions — like "What concerns do diabetes patients have about starting injectable therapy?" — and receive integrated, attributed answers drawn from all sources simultaneously. A configurable filtering layer lets users scope queries by disease area, stakeholder type, journey stage, demographics, and geography. The synthesis engine identifies convergent themes across sources, surfaces divergent perspectives, and flags evidence gaps — cross-source patterns invisible when each source is searched in isolation. For example, concerns dominating patient forums may be entirely absent from published literature, revealing unstudied issues. The poster shares deployment learnings, including how to structure multi-source synthesis for different use cases, present findings with appropriate confidence levels and limitations, and build organizational capability around unified patient intelligence.

Reaching the Right Physician Before the Diagnosis: A Predictive HCP Targeting Case Study

Many rare diseases affect an estimated 5,000–10,000 patients in the U.S., making efficient HCP identification critical for any commercial strategy. Traditional claims-based targeting triggers activate only after diagnosis — a point at which the majority of prescribing decisions have already been made. This poster presents a multi-year predictive targeting program developed for a major Pharma brand, in partnership with PulsePoint.

Using a multi-signal intelligence framework integrating clinical procedure patterns, endemic content engagement, prescription history, and patient demographics, PulsePoint built dynamic HCP segments designed to identify high-probability prescribers 35–60+ days before traditional clinical triggers fired. The resulting segmentation narrowed the addressable HCP universe, while maintaining 50–60% coverage of historically verified prescribers — and concentrated media spend 5–8x among physicians most likely to encounter relevant patients in the near term.

This poster will detail the methodology, signal hierarchy, and validated campaign outcomes, and explore the implications for rare disease marketing strategy at scale.

To enable successful commercial launches, organizations should prioritize timely access to high-quality Marketing Mix (MMX) data through a standardized and automated data quality (DQ) framework. Traditional approaches relying on fragmented, manual validation across multiple stakeholders often delay workflows, limit KPI visibility, and force teams into reactive issue resolution, ultimately impacting launch readiness and decision-making.

Organizations should implement an integrated, AI-powered DQ operating model comprising two key capabilities: an Automated Data Quality Engine and a Data Quality Watch Tower. The engine should embed comprehensive business and technical validation rules across datasets, brands, and marketing channels to enable rapid, scalable validation and eliminate manual effort. Complementing this, real-time dashboards should provide continuous KPI monitoring, allowing proactive detection of data issues such as missing values, schema mismatches, and inconsistencies.

By shifting to automated validation and real-time visibility, teams can improve data trust, enhance operational transparency, and accelerate decision-making during critical launch windows. This approach also frees up resources for strategic initiatives while strengthening governance and cross-functional alignment.

Overall, adopting a scalable, rules-driven, AI-enabled DQ framework should materially improve data readiness, ensuring faster, more confident commercial execution in complex environments.

Marketing Mix Modeling (MMM) has long served as a cornerstone for strategic budget allocation in life sciences and other regulated industries. However, traditional MMM approaches are inherently episodic, slow to refresh, and expert-dependent, limiting their ability to support in-quarter course correction and rapid decision-making in increasingly dynamic market environments. As channels proliferate, data latency shrinks, and marketers demand faster insights, the gap between MMM outputs and real-world decision needs continues to widen.

This presentation introduces an always-on, agentic marketing mix capability, currently being implemented at Bristol Myers Squibb (BMS), that reimagines MMM as a continuous decision-support ecosystem rather than a periodic analytical exercise. The proposed framework combines continuous data ingestion, automated modeling, and agentic orchestration to enable near–real-time planning, simulation, and optimization of marketing investments. Instead of relying on quarterly or annual re-runs of MMM, the system learns incrementally as new data arrives, supporting mid-quarter budget reallocations and rapid performance course-correction.

At the core of the approach is an agentic architecture, where specialized AI agents manage data validation, model refresh, scenario simulation, budget recommendation, and governance. Together, these agents enable the platform to act as a 24/7 AI analyst, capable of performing multi-step causal analysis, identifying key performance drivers, and generating actionable recommendations. This shifts the role of MMM from retrospective reporting toward proactive decision intelligence—moving beyond “What happened” to root cause analysis and “What’s next.” As a result, brands and omnichannel leads gain always-on access to insights that continuously adapt to changing market conditions.

A self-service, user-centric interface allows marketers to interactively validate data, explore scenarios, and understand trade-offs—dramatically reducing dependency on centralized analytics teams while increasing speed and confidence in decision-making. The framework is designed to scale across brands and markets through modular, reusable components, enabling consistent governance while preserving brand-specific nuance.

By shifting from “analyze, optimize, and deploy” to “deploy, learn, and adapt,” this agentic, always-on analytics platform provides a foundational layer for enterprise decisioning. It enables faster, smarter marketing mix decisions at scale, improves organizational responsiveness, and lays the groundwork for a broader Agentic Always-On Analytics Platform that supports continuous insight generation across brands, channels, and planning horizons.

What if field operations didn’t require stitching together multiple systems, manual inputs, and delayed insights?

This poster introduces a Unified AI-Enabled Field Ecosystem that reimagines the end-to-end field engagement lifecycle, from planning and travel to HCP interactions and performance management. Instead of fragmented workflows, the model brings enterprise data into a single cloud-based intelligence layer, enabling AI-augmented orchestration at every step.

AI prioritizes HCPs, recommends content, optimizes travel routes, and supports in-call interactions through contextual prompts and automated capture. Post-call activities are streamlined through zero-touch CRM synchronization and AI-generated next best actions, while performance management shifts from static reporting to real-time, scenario-based insights.

Deployed in a global pharmaceutical setting, this approach demonstrates measurable improvements in preparation efficiency, administrative reduction, and travel optimization while also improving alignment and quota visibility.

Know more about how a unified, AI-native ecosystem can transform field execution into a responsive, continuously learning system.

Pharmaceutical commercial teams are increasingly challenged to make data-driven decisions due to the sheer volume and complexity of available information. While analytics platforms have evolved to offer dashboards and self-service BI, decision-making can still be delayed by reliance on technical experts for data access and rigid, predefined views.

This presentation outlines agentic AI-powered Advanced Analytics &Visualization solution. This solution is an intelligent analytics assistant that interprets natural-language business queries, delivers structured insights, and generates visual representations without requiring technical expertise or manual report building.

This solution leverages a multi-agent architecture, in which specialized agents (for example, query processing, &insights reasoning agents) work together to deliver end-to-end advanced analytics and visualization experiences, with the ability to save generated analyses and visualizations for future reference and reuse. The session will also showcase how our agent can support commercial tasks like tracking performance, analyzing brands and customers, and assessing market and competitive trends by generating understandable outputs in textual, tabular, and visual formats. The solution follows industry standard governing principles such as access control, audit logging, and data protection.

Pharma commercial teams operate across fragmented data ecosystems spanning HCP engagement, omnichannel marketing, CRM, patient services, and external vendors. While data volume has grown, both AI systems and users struggle to consistently interpret and apply it due to dispersed institutional knowledge such as metric definitions, vendor nuances, business rules, and usage constraints. This fragmentation reduces AI reliability, increases variability in outputs, and drives dependence on manual prompt engineering and specialized expertise.

This presentation introduces an AI-enabled framework that acts as an enterprise prompt-augmentation layer powered by domain-specific knowledge graphs. The framework structures relationships between datasets, metrics, vendors, use cases, and constraints into a reusable knowledge layer that can be dynamically retrieved and injected into AI workflows at runtime.

This enables consistent, context-aware interactions across applications including BI, analytics, data science, and agent-driven use cases. Solutions access contextual intelligence while attending to natural language user queries, synthesizing structured knowledge and documentation into clear, actionable insights.

By combining human expertise with knowledge graphs and generative AI, this approach transforms static documentation into a scalable, governed context layer that improves the consistency, reliability, reusability, and impact of AI-enabled decision-making.

Regulatory Affairs teams face mounting pressure to deliver faster, higher-quality submissions amid rising document complexity, frequent regulatory changes, and lifecycle management challenges. Despite digitization efforts, many organizations still depend on fragmented tools and manual processes, with critical knowledge remaining trapped in individuals rather than systems.

This poster explores how an AI-native platform like Harmony is transforming dossier authoring and lifecycle management for ANDA, NDA, INDs, and DMFs. It moves beyond traditional digitization to intelligent, context-aware regulatory authoring, enabling seamless preparation, revision, querying, and management of dossiers throughout their lifecycle.

Aligned with the PMSA 2026 theme "Convergence of Data, Talent &AI" the poster demonstrates how integrating structured regulatory data, human expertise, and AI intelligence creates connected, knowledge-driven systems. This allows professionals to shift focus from routine tasks to high-value strategic work, improving speed, consistency, and submission readiness.

Key Takeaways include limitations of document-centric approaches, benefits of AI-assisted authoring for faster submissions and variations, content reuse strategies, intelligent search capabilities, and practical AI adoption considerations while ensuring compliance.

The session offers Regulatory Affairs professionals and digital transformation leaders actionable insights on building AI-ready organizations, identifying immediate value opportunities, and planning effective AI integration to reduce timelines, enhance quality, and support scalable future growth.

This poster presents Agentic Forecasting as a next-generation approach that transforms traditional biopharma forecasting from a static, periodic process into a dynamic real-time decision engine. It highlights how conventional forecasting methods often struggle to keep pace with rapidly changing market conditions, competitive events, access dynamics, and internal business decisions. In response, the poster introduces a governed agent-based framework where GenAI-powered agents continuously monitor signals, refine assumptions, and support forecast recalibration.

The poster outlines the conceptual architecture behind this approach, showing how diverse data inputs, forecasting engines, and a central system of record are connected through specialized agents. These agents perform functions such as identifying historical analogs, incorporating stakeholder insights, tracking competitive intelligence, evaluating launch plans, and simulating the financial impact of strategic decisions. Together, they enable a more responsive and transparent forecasting environment.

Field Pro is an AI assistant embedded within the CRM workflow to help field teams with faster, better pre-call and territory planning. It was designed around real rep workflows and addresses the main barriers that often limit GenAI adoption in the field: intent misalignment, weak grounding, fragmented data, and slow response times. Using a domain-infused multi-agent architecture, Field Pro combines structured and unstructured data, applies guardrails for safer and more relevant responses, and returns concise, call-ready insights. Since launch, it has shown strong adoption and usage, suggesting that workflow-embedded, evidence-grounded AI can move beyond experimentation and deliver practical value for commercial teams. The poster highlights the product design choices, data and orchestration approach, and lessons for evolving from an “insight assistant” toward a more proactive field copilot.

This presentation introduces a Medical Affairs data strategy centered on purposeful integration and fit-for-purpose data products. The framework moves organizations from episodic reporting to continuous insight generation through deliberate alignment with critical business questions.

This session presents a real-world architecture journey from a regulated pharmaceutical environment, where the primary barrier to adopting Generative and Agentic AI was not model capability, but enterprise trust. The organization faced fragmented lineage, inconsistent data quality, and limited traceability across data, prompts, embeddings, and model outputs, making it impossible to scale AI beyond isolated proofs of concept without introducing compliance risk.

To address this, we implemented a Medallion architecture, reinforced with Data Vault principles, as a practical governance and traceability framework for AI. In this design, Medallion was not just a data-refinement pattern—it functioned as a control and observability layer for governed AI, treating GenAI artifacts as first-class data assets.

Using Medallion’s traditional color scheme as brown for Bronze, gray for Silver and yellow for Gold layer, Bronze served as the immutable system of record, capturing raw structured and unstructured data alongside prompts, embeddings, model outputs, model metadata, and execution telemetry, with full auditability and temporal traceability.

Silver applied Data Vault–aligned normalization, validation, semantic enrichment, and business rule enforcement, transforming raw AI and enterprise inputs into trusted, policy-compliant, explainable assets with clear lineage across hubs, links, and satellites.

Gold published certified features, curated knowledge representations, and governed inference datasets that downstream models and agents could safely consume in a GxP context.

This architecture enabled the organization to reproduce model behavior, trace hallucination sources, detect drift, and enforce human-in-the-loop controls, meeting regulatory expectations for transparency and explainability.

Critically, this relationship was bidirectional and grounded in a repeatable architectural pattern. The pattern refers to the systematic treatment of GenAI artifacts—such as prompts, responses, embeddings, metadata, and telemetry—as governed data assets managed through Medallion and Data Vault refinement layers. While these refinement patterns enabled trusted, enterprise-scale GenAI adoption, GenAI was simultaneously applied to automate and enhance the architecture itself, supporting metadata enrichment, data quality validation, schema change detection, documentation generation, and lineage interpretation across Bronze, Silver, and Gold layers. Agentic workflows further assisted with rule authoring, anomaly triage, and governance operations, reducing manual effort without compromising control.

The pattern itself is not proprietary; its value lies in operationalizing governance by embedding AI artifacts into established data-refinement patterns, enabling GenAI to transition from isolated experimentation to safe, repeatable, and compliant enterprise deployment.

AI transforms the end-to-end business process around commercial data operations. What are the typical use cases, implementation paradigms and benefits for this transformation?

A Bridge Data Layer (BDL) for Conversational AI serves as an intelligent middleware architecture that mediates between large language models (LLMs) and heterogeneous enterprise data sources. Rather than allowing AI systems to query databases, APIs, or knowledge repositories directly, the BDL acts as a semantic translation and orchestration layer — normalizing data schemas, enforcing access controls, and transforming raw structured data into context-rich representations that conversational models can reason over effectively.

Despite rapid advances in AI capabilities and significant investment, many organizations struggle to move beyond pilots to sustained impact. The challenge is not model performance, but organizational readiness. While employees increasingly embed AI into their day to day work, leadership often continues to treat AI as a technology deployment initiative—rather than a transformation of roles, workflows, and decision rights. This disconnect, combined with inconsistent training and unclear governance, causes promising efforts to stall before scaling.

This session reframes AI adoption in commercial forecasting and insights as a human AI collaboration and talent transformation challenge, not a tooling problem. It introduces a practical five stage AI maturity framework—from foundational automation to safe, constrained autonomy—that helps leaders assess where their organizations are today and what must change to progress. Each stage is evaluated across key dimensions including human roles, decision ownership, workflow integration, governance, and business impact. The framework will be brought to life through a critical industry use case: forecasting for a product launch.

Life sciences organizations generate advanced analytics for commercial strategy, but field teams often rely on static plans and disconnected systems. This gap limits impact. Meanwhile, sales representatives are evolving into data-driven decision-makers, requiring effective human-AI collaboration. Intelligent systems should enhance—not replace—human judgment by embedding insights directly into workflows. There is a growing need for platforms that bring analytics to the point of action while supporting talent transformation. To design and implement an AI-enabled, dynamic HCP route planning platform that integrates analytics into daily workflows, enabling real-time collaboration and improving productivity, prioritization, and decision-making. The platform replaced static territory plans with dynamic, AI-driven tools. Representatives achieved faster planning, better HCP prioritization, and improved adaptability to real-world factors like traffic and weather. Managers gained better visibility into execution, enabling targeted coaching, while real-time insights supported home office decision-making.

THE WHY (The Purpose): We believe our sales force's most valuable asset is their time spent in front of customers, not behind a dashboard.

The current reality is that our reps are bogged down by "analytical friction"—spending hours deciphering siloed reports to find their next move. We believe that by removing the burden of data analysis from the field, we empower our reps to do what they do best: sell.

THE HOW (The Process): We will bridge the gap between financial motivation and daily execution through contextual AI nudges.

Our approach isn't just about showing data; it’s about **translating** it. We will deliver insights at the moment of highest impact—when a rep is checking their IC payout. By converting complex performance metrics into a conversational dialogue, we provide the shortest path between Where is my commission?" and "Which customer do I call next?"

THE WHAT (The Result): An AI-driven "Performance Coach integrated directly into CRM.

This is a conversational interface that acts as a personal analyst for every rep. It identifies high-probability opportunities, suggests specific actions to hit IC milestones, and eliminates administrative waste.

The Outcome: A results-driven sales force that spends less time on spreadsheets and more time hitting the targets that drive our business growth.

Pharmaceutical forecasting doesn’t fail because of a lack of data—it struggles because the world keeps changing. Regulatory decisions, competitive launches, pricing shifts, and clinical news can reshape demand overnight, forcing forecasters into slow, manual cycles of model rework. That challenge sparked the collaboration between Ferring and Axtria that led to SMART Extrapolator.

Built for real‑world complexity, SMART Extrapolator abandons the “one‑size‑fits‑all” forecasting mentality. It adapts its methods to each product’s data reality—whether sparse or rich, short‑term or long‑range—and automatically evaluates which models work best. It goes further by intelligently testing external drivers like pricing and competition, using them only when they genuinely improve predictions.

What truly sets the platform apart is its event awareness. By continuously scanning pharmaceutical news across regulatory, clinical, and market domains, SMART Extrapolator turns emerging signals into actionable scenarios. Layer in hierarchical forecasting across geographies and conversational analytics that cut hours of analysis to seconds, and forecasting moves from reactive to strategic.

The result: forecasters gain speed, scale, and insight—without losing control. SMART Extrapolator doesn’t replace expertise; it amplifies it.

Oncology biologics require navigating complex, dynamic patient populations characterized by evolving treatment pathways, biomarker-driven segmentation, and variability in access and provider behavior. Traditional analytics approaches—often based on static segmentation and point-in-time reporting—are limited in their ability to support forward-looking, decision-oriented planning, particularly during launch and early inline optimization.

This poster presents a patient-segment digital twin framework designed to address these challenges. Rather than modeling individual patients, the approach focuses on clinically and commercially meaningful patient segments, creating continuously updated representations of how these populations evolve over time. By integrating real-world data sources such as claims, EHR, biomarkers, access data, and commercial engagement signals, the framework enables simulation of patient flow, treatment transitions, and response to strategic interventions.

The methodology combines transparent cohort construction, time-to-event and transition modeling, and scenario-based simulation to generate decision-ready insights. These include estimates of uptake timing, duration of therapy, and sensitivity to evidence, access, and competitive dynamics. Outputs are presented as ranges to reflect uncertainty and support robust decision-making.

By emphasizing explainability, governance, and human-in-the-loop validation, this framework demonstrates how advanced analytics can move beyond descriptive reporting to enable predictive and prescriptive insights in oncology.

Pharmaceutical commercial analytics teams operate across fragmented data ecosystems, including structured databases, dashboards, internal documents, and external research. Insights are often generated through disconnected tools and manual synthesis, leading to delays, duplication, and limited context. Existing platforms typically optimise for a single modality, lacking a unified and explainable intelligence layer. ConvergeAI™ Analytics Engine addresses this gap as an agentic, multimodal AI platform that unifies data, dashboards, and enterprise knowledge into a single conversational experience. It leverages four specialised agents: Insight (structured data analysis), Lexi (unstructured text interpretation), Vista (visual/dashboard reasoning), and Cognify (advanced analytics like trends and correlations). An orchestration layer dynamically activates the right agents based on user intent. Users can ask complex business questions and receive structured, explainable outputs, including insights, trends, SQL queries, visualisations, and recommendations. The platform ensures traceability and trust, critical in regulated environments. A human-in-the-loop model allows SMEs to update context and definitions without developer support. With GenAI-assisted onboarding, ConvergeAI™ reduces analytical effort by ~90%, saves ~2 hours daily per analyst, and accelerates data integration from days to hours—enabling faster, proactive decision-making.

This poster presents a comprehensive MLOps framework designed to automate and govern a machine learning-driven solution for healthcare professional (HCP) prioritization. The solution integrates ten interconnected models leveraging nearly one hundred engineered features sourced from over fifteen heterogeneous datasets, including prescription records, digital promotion metrics, sales call logs, and HCP demographic details. Advanced ETL pipelines ensure systematic data cleansing, standardization, and structural consistency across all inputs.

Model development utilizes Automated Machine Learning (AutoML) tools within the Dataiku platform, enabling streamlined hyperparameter tuning and efficient model selection. A fully automated MLOps pipeline manages the entire model lifecycle - from data ingestion and training through testing, validation, deployment, and continuous monitoring - eliminating manual intervention to reduce costs, turnaround times, and human errors.

Central to the framework is a robust governance structure encompassing continuous performance evaluation, quality assurance, and compliance management. Comprehensive metrics assess model accuracy, stability, robustness, fairness, and ethical standards, with automated alerts identifying anomalies for timely optimization, ensuring alignment with business requirements and regulatory standards.

By integrating rigorous automation with comprehensive governance, the solution enables data scientists to focus on new initiatives while maintaining reliable, scalable, and compliant ML models in production, unlocking the full potential of machine learning for healthcare business applications.

Ex‑US pharmaceutical markets present a complex promotional landscape shaped by diverse regulatory environments, multichannel engagement patterns, and therapeutic‑area heterogeneity. As commercial teams continue to expand omnichannel strategies, the need for rigorous, data‑driven Marketing Mix Measurement (MMM) and investment optimization has become critical for maximizing promotional effectiveness and return on investment (ROI). This work outlines an enterprise‑scale MMM initiative conducted across EU5 markets—France, Germany, Italy, Spain, and the UK and Japan spanning multiple therapeutic areas including HIV, Oncology, and Liver. The objectives were to quantify promotional impact, evaluate channel‑level ROI, surface strategic trade-offs, and optimize commercial investments to meet brand‑specific revenue and forecasting goals.

This initiative aimed to revolutionize patient services field force deployment in Specialty Pharmaceutical Care by developing and implementing AI-driven models to: (1) identify patients at high risk of not starting or staying on therapy, (2) elucidate key risk drivers at the individual patient level, and (3) generate prescriptive recommendations enabling targeted interventions through a transformed field force approach. A comprehensive AI framework was developed utilizing machine learning algorithms and generative AI capabilities, with hundreds of features engineered across four critical domains: patient characteristics, provider attributes, specialty pharmacy, and payer attributes. The AI solution generates patient-specific risk scores highlighting key reasons why a patient is unlikely to start on therapy, enabling field managers, case managers, and specialty pharmacy partners to address risk drivers through targeted interventions. The overarching goal was to shift from traditional reach-and-frequency tactics to surgical, data-driven interventions that optimize customer interactions and drive measurable improvements in patient access and therapy initiation.

This poster presents a practical, data-driven framework designed to improve commercial field effectiveness beyond traditional targeting and Next Best Action (NBA) approaches. While existing models largely rely on structured data and predefined HCP lists, they often miss emerging signals from scientific, digital, and clinical activity, limiting both reach and relevance.

Our approach integrates internal data such as sales, CRM notes and NBA alerts with external sources such as publications, conferences, clinical trials, and digital engagement to create a more dynamic view of HCP behavior. Using natural language processing and topic modeling, unstructured data is translated into actionable insights at an individual HCP level. These insights are prioritized through a relevance-scoring layer and connected to brand and disease strategies to provide clear “why now” context.

The framework is operationalized through targeted alerts and short summaries that help field teams prepare for more meaningful interactions. A structured conversation model further supports reps in delivering focused, compliant engagements.

Early application shows improved identification of high-potential HCPs beyond traditional lists, better visibility into emerging engagement trends, and stronger alignment between analytics and field execution, ultimately enabling more confident and context-driven decision-making in the field.

Forsyth Health’s poster introduces a Prescribing Event algorithm that links upstream administrative signals—benefit checks, prior authorizations (PAs), and pharmacy rejects—with medical and pharmacy claims to create a longitudinal, near–real-time view of prescriber intent through patient adoption. Building on prior work showing that many prescribing journeys are invisible in claims alone, this poster maps “10 Points of Failure” where patients drop off between initial brand interest and therapy initiation—an attrition gap that can approach 45% for specialty brands.

Attendees will see how Prescribing Events are constructed (anchored by the first observable prescribing-related activity after a clean period), how downstream outcomes are attributed (e.g., on-brand initiation, reversal/abandonment, PA approval with no pickup), and how normalized PA denial reasons can enable scalable, AI-ready analytics. The framework helps brand, sales, and field reimbursement teams quantify demand that never becomes a claim and pinpoint where access friction disrupts adoption, enabling more targeted interventions to recover lost prescribing demand.

HCP engagement is undergoing a fundamental transformation as digital channels increasingly complement traditional in‑person sales representative interactions. HCPs now operate within complex omnichannel ecosystems encompassing personal promotion (e.g., sales calls, emails, speaker programs) and non‑personal promotion (e.g., digital advertising, social media, EHR, and point‑of‑care media). As channel preferences vary widely across HCPs, uniform engagement strategies are becoming less effective, underscoring the need for scalable, data‑driven personalization.

To address this challenge, an advanced analytics framework leveraging artificial intelligence and machine learning was developed to quantify and predict HCP‑level channel affinity across both reached and unreached populations. The framework was applied to a GSK respiratory brand by integrating data from more than ten promotional channels and analyzing engagement patterns across over 90,000 HCPs. For reached HCPs, observed reach and engagement metrics were used to generate channel affinity scores. For unreached HCPs, multiple AI‑based collaborative filtering approaches—including singular value decomposition (SVD), neural collaborative filtering (NCF), and k‑nearest neighbor (KNN) models—were evaluated to infer channel affinity using prescribing behavior, specialty, practice characteristics, and peer‑based similarity.

By combining observed and AI‑predicted channel affinity into a unified HCP‑level view, the framework enables scalable personalization across the entire target universe, improves promotional efficiency, and supports hyper‑personalized engagement strategies in an increasingly digital pharmaceutical environment.

Modern pharmaceutical marketing operates in a complex omnichannel ecosystem, where campaigns span multiple personal and non-personal channels across diverse customer segments. Traditional test-and-control approaches often result in low match rates and insufficient statistical power, limiting reliable campaign-level measurement.

This poster introduces Synthetic Test &Control (T&C), a novel framework designed to address these challenges and enable robust impact measurement. Developed and validated in collaboration with GSK, the approach uses advanced matching techniques to construct high-quality test &control groups, even with constrained control populations and thresholds.

Compared to traditional methods, Synthetic T&C demonstrates 4–10x improvement in match quality, higher match rates even with smaller control group sizes (10–15% of the population, compared to ~30% typically required in traditional approaches), and up to 80% reduction in analysis turnaround time.

Real-world case studies demonstrate how the framework enables reliable and scalable campaign-level measurement across brands.

Leveraging Agentic AI for driving personalized engagement at HCP, Patient, and Payers across diverse therapeutic areas.