AI for Drug Development and Portfolio Management – with Leaders from Intelligencia AI and Novartis

This interview analysis is sponsored by Intelligencia AI  and was written, edited, and published in alignment with our Emerj sponsored content guidelines. Learn more about our thought leadership and content creation services on our Emerj Media Services page.

Rising costs, long development cycles, and uncertain success rates continue to challenge pharmaceutical R&D efficiency. According to McKinsey & Company, pharmaceutical R&D spending reached $260 billion in 2023, yet the average time from Phase I to market launch remains at ten years, with per-launch costs rising to $4 billion in 2022. These widening gaps between investment and outcomes demand new strategies to optimize efficiency and improve returns.

Interest in AI as a tool to improve decision-making and efficiency in drug development has surged, driven by the need to bridge the gap between investment and outcomes. While AI shows promise in accelerating discovery, persistent challenges remain. Harvard Catalyst highlights issues such as data quality, model transparency, and integration hurdles as key obstacles to adoption.

Data complexity is a major barrier—approximately 80% of biomedical data remains unstructured, complicating integration into machine learning pipelines designed for structured inputs. Model transparency also poses difficulties; regulatory agencies like the FDA emphasize the importance of explainable AI to ensure accountability in drug development. According to a 2024 McKinsey survey, 40% of respondents cited explainability as a critical risk when adopting generative AI, yet only 17% are actively addressing it.

Emerj Artificial Intelligence Research recently spoke with Panos Karelis, Director of Customer Experience and Insights at Intelligencia AI, and Scott Bradley, Vice President of AI and Innovation at Novartis. Together, they explore practical solutions to these challenges, drawing on their experience in deploying AI within pharmaceutical R&D.

Intelligencia AI develops software that assesses the risk and probability of success across drug development programs. Its platform integrates curated biomedical data and machine learning to support more informed portfolio and investment decisions in the life sciences. The company collaborates with R&D, clinical, and business teams to enhance how organizations assess assets and allocate resources.

Throughout their conversation, both leaders emphasized the importance of embedding AI thoughtfully into existing decision-making frameworks. Their insights underscore the importance of success not only in technical accuracy but also in transparency, change management, and the ability to translate predictions into actionable guidance for teams operating across clinical and business domains.

• Building trust through transparency: How making AI models interpretable at both input and output levels helps overcome skepticism, supports validation steps in early-stage drug development, and builds confidence across R&D teams.
• Improving portfolio decision-making: How AI enables sponsors to evaluate trade-offs across pipeline assets earlier and with greater accuracy—helping R&D leaders identify high-potential assets, reduce risk, and allocate resources more strategically.

Emerj thanks Panos Karelis and Scott Bradley for sharing their expertise on AI adoption, risk 

management, data transparency, and change leadership in pharmaceutical R&D. Below, enterprise leaders will find a detailed breakdown of the key insights from their conversation.

Transparency as the Foundation of AI Adoption in Pharma

Episode: Accelerate and Mitigate Risk in Clinical Development  – with Panos Karelis of Intelligencia AI

Guest: Panos Karelis, Director of Customer Experience and Insights, Intelligencia AI

Expertise: AI in life sciences, clinical development optimization, data-driven decision-making

Brief Recognition: Panos Karelis serves as Director of Customer Experience and Insights at Intelligencia AI. He works closely with pharmaceutical and biotech partners to integrate AI across clinical workflows and strategic decision-making. With a strong background in life sciences, customer-facing AI engagement, and data analytics, he helps bridge the gap between predictive models and real-world R&D outcomes.

Panos Karelis opens the discussion for Emerj’s executive audience by underscoring that the success of AI in drug development fundamentally depends on high-quality, well-structured data and clear transparency throughout the process. 

He emphasizes that such levels of transparency mean that “good data” must be:

• Comprehensive
• Recent and relevant
• Harmonized with organization data
• Carefully engineered in a “unified source of truth”

Panos defines that “unified source of truth,” as one that:

• Supports thorough analysis
• Generates meaningful, actionable insights

According to Panos, building systems according to his framework requires adopting an engineering mindset to biological data, transforming disparate datasets into parameterized and consistent forms that both analysts and AI systems can reliably use.

He highlights that building trust in AI systems extends beyond just technical performance — it requires the kind of transparency and explainability whereall stakeholders, from scientists to decision-makers, can understand, verify, and have confidence in the insights produced. 

 Transparency defined so rigorously is essential not only for internal validation of algorithm accuracy but also for fostering broader user confidence and industry acceptance.

AI’s core value, Panos explains, lies in transforming complex risk assessments and probabilities into actionable business decisions. His team develops solutions that assess the likelihood of both technical and regulatory success for drug candidates, enabling pharmaceutical companies to prioritize their resources more effectively. 

By analyzing extensive, multi-dimensional datasets, these AI models can uncover patterns that inform better clinical trial designs, optimize drug development portfolios, and improve the accuracy of outcome predictions. Capability allows companies to:

• Identify weaker candidates early in development
• Reallocate investments toward higher-probability programs
• Address rising drug development costs described by Eroom’s Law

Panos cautions against viewing AI as a standalone solution. Instead, he emphasizes its role in augmenting human expertise and judgment. “AI isn’t a magic bullet; it augments human capabilities rather than replaces them. The most successful implementations blend AI insights with human context and experience to enhance decision making.”

By combining machine analytics with domain knowledge, companies ensure decisions remain grounded in scientific rigor and clinical realities, improving overall decision quality.

Beyond technical challenges, Panos addresses the organizational and cultural hurdles involved in AI adoption. He emphasizes the necessity of a mindset shift and the development of AI fluency throughout an organization, noting that adoption is as much about people and processes as it is about technology.

Panos advises companies to start small with focused use cases to build confidence, demonstrate value before scaling, invest in training programs to close skill gaps and secure executive sponsorship to overcome cultural resistance.

Looking forward, Panos envisions AI driving a transformation from traditional data collection to what he calls “decision intelligence.” This shift will enable R&D leaders to allocate resources more strategically and efficiently, guided by real-time, actionable insights derived from AI-powered analytics.

He also sees generative AI and large language models playing a pivotal role in democratizing insight generation — making complex, high-level data analysis more accessible across all levels of the industry. However, he warns that companies must proceed cautiously, given risks like hallucinations in generative AI models and biases embedded in training data.

Ultimately, Panos says companies should embrace AI as a critical tool to de-risk the drug development process, improve cost-effectiveness, and increase the success rate of bringing new therapies to market. That mission drives ongoing efforts to accelerate patient access to innovative treatments by combining advanced AI technologies with rigorous data management and human expertise.

From Gut Instinct to Data-Driven Strategy: Balancing AI and Expertise in Drug Development

Episode: Building Agile R&D Strategies Through Predictive Analytics – with Scott Bradley of Novartis

Guest: Scott Bradley, VP of AI and Innovation, Novartis

Expertise: AI-driven R&D innovation, protein design, portfolio strategy

Brief Recognition: Scott Bradley serves as Vice President of AI and Innovation at Novartis, leading initiatives that apply AI and predictive analytics across drug discovery, clinical development, and portfolio management. He is a seasoned life sciences executive and angel investor with decades of experience driving digital transformation and AI-powered strategic decision-making in the pharmaceutical industry.

Early in the drug development process when leaders are deciding what potential medicines to invest in, Scott Bradley notes that a data-driven mentality is becoming increasingly central, particularly among researchers in leadership positions. He explains that such leaders in these conversations ask “cause and effect” questions to understand why an AI model produces specific recommendations. 

According to Scott, “I don’t think anyone is putting all of their trust and faith in those outputs. We still are following these early pre-clinical steps in the development cycle to validate those results.” 

Like anyone new to AI, building trust takes time and experience as teams learn to recognize and rely on the positive, reliable outputs AI can provide. With his years of experience in drug portfolio development, Bradley is noticing a fundamental shift underway in how  R&D teams advise these investments: moving from gut-based decisions to data-driven strategies.

“AI is fundamentally changing how we make portfolio decisions in drug development, moving us from gut-based approaches to data-driven strategies that can decode protein complexity in minutes rather than months. This shift allows us to advance more targeted therapies for specific patient populations with higher predictability of success, ultimately getting more effective treatments to patients faster. The companies that will win are those that can leverage AI to identify both successes and failures early in the process, dramatically reducing the cost and time of bringing new drugs to market.”

— Scott Bradley, VP of AI and Innovation at Novartis

He also notes a critical challenge in scaling these advances through clinical development: “We could be sitting on a gold mine of many promising new future drugs but cannot get those through to regulatory approval and into the hands of patients. The clinical trial process hasn’t scaled proportionally, so there’s a pipeline bottleneck.” 

 Pipeline bottlenecks shape tough portfolio decisions about how to strategically segment patient populations, ensuring that micro-targeted therapies do not cluster in overlapping segments while leaving gaps in treatment coverage, according to Bradley: “Ultimately, that’s the goal — to prevent and cure disease and improve health and life.”

Scott elaborates that AI’s impact goes beyond portfolio management, addressing deep technical challenges throughout the drug discovery value chain, but particularly in R&D. Bradley notes, “There are so many problems on the research side being tackled with AI right now — it’s probably the most significant area of investment for deep technical AI beyond the chat GPT-level — to understand clinical pathways and, crucially, the protein molecules themselves. 

According to Scott, a lesser-known breakthrough is AI’s role in synthesizing proteins: identifying candidates is one thing, but manufacturing them is a complex process. AI is now unlocking the underlying chemistry, solving problems that were previously unsolvable, and demonstrating AI’s potential to drive innovation at both molecular and strategic levels.

He underscores the importance of combining AI analytics in life sciences R&D with expert judgment:

“The magic happens at the intersection of AI analytics and human expertise. The algorithms can process vast amounts of data and identify patterns. Still, it’s the scientists and clinicians who provide context, interpret results, and make the final calls based on their experience and judgment.” 

— Scott Bradley, VP of AI and Innovation at Novartis

Collaborative approaches reflect the mission Panos described: to augment human decision-making with advanced analytics, ensuring data-driven insights are always grounded in domain knowledge.

Another place for human talent in AI development that Bradley points to is the critical role of AI fluency in enterprise leadership ” Leaders don’t need to code,” he tells the Emerj executive audience. “But they do need to understand AI’s capabilities and limitations to drive strategic implementation.” 

Novartis invests heavily in upskilling its teams and cultivating a culture that values and rewards data-driven decision-making. Enterprise-level organizational readiness aligns with Karelis’s recommendation to start AI adoption with focused use cases that build confidence before scaling.