AI For Pharma Growth

Digital twins have become one of the most promising tools in Alzheimer’s research, but the bigger story is what happens when they scale across disease areas. In this episode, Dr Andree Bates interviews Aaron Smith, Founder and Head of Machine Learning at Unlearn AI, about how “digital twin generators” can transform trial design by modelling realistic patient progression and improving statistical power without compromising the fundamentals of randomised controlled trials.
Aaron shares his journey from academic mathematics into computer vision and machine learning, then into biopharma, where Unlearn began by building generative models that learn the joint distribution of clinical variables. In practice, that means the model can take baseline patient measurements and generate likely future progressions that are as indistinguishable from real clinical records as possible.
The conversation dives into a key misconception: digital twins are not only about replacing control arms. Aaron explains a regulatory friendly approach where you keep standard trial structure, but add counterfactual information for every patient into the analysis. Unlearn’s best known method, ProCOVA (prognostic covariate adjustment), summarises a predicted control outcome per patient and uses it for covariate adjustment, creating more efficient treatment effect estimates. The headline result is simple: you can increase power, or reduce recruitment burden while maintaining power, potentially speeding time to results.
Finally, Aaron explains why scaling across diseases is genuinely hard. Data structures differ wildly by indication, missingness can block transfer learning, and areas like oncology require modelling complex treatment histories. He also highlights that combining sources is not just “more data”, it demands careful harmonisation and context modelling to avoid biased predictions, especially when bringing in real world evidence.

Topics Covered
What “digital twin generators” are in clinical trials

Generative modelling of clinical records and disease progression

Counterfactual prediction under standard of care

Why replacing control arms is not the only use case

ProCOVA and prognostic covariate adjustment

Getting more statistical power and reducing trial size

FDA openness to digital twins in trials and what it enables

Why scaling across disease areas is not just parameter tuning

Missing data, confounding context, and data harmonisation

CNS versus oncology modelling challenges

Real world evidence and how to validate digital twin models

About the Podcast
AI For Pharma Growth is the podcast from pioneering Pharma Artificial Intelligence entrepreneur Dr Andree Bates, created to help pharma, biotech and healthcare organisations understand how AI-based technologies can save time, grow brands, and improve company results.
This show blends deep sector experience with practical conversations that demystify AI for biopharma leaders, from start-up biotech right through to Big Pharma. Each episode features experts building AI-powered tools that are driving real-world results across discovery, R&D, clinical trials, medical affairs, market access, regulatory, insights, sales, marketing, and more.

Dr. Andree Bates LinkedIn | Facebook | X

AI in drug development is moving beyond “failure prevention” into something much bigger: redesigning how we discover, develop, and deliver medicines. In this episode, Dr Andree Bates speaks with Vitalay Fomin of Numenos about biomarker discovery, patient stratification, and why the next breakthroughs come from breaking down data silos across diseases, modalities, and even species.
Vitalay shares his background across biotech and pharma, including work in biomarker discovery, translational medicine, and data science, and how frustration with existing approaches led her to build a new architecture for clinical genomic insights. A core theme is that traditional methods often oversimplify biology by forcing outcomes into binary labels and treating each disease area as an isolated box, even when the available data is too limited to answer meaningful questions well.
The conversation explores how foundation model approaches can unify clinical, genomic, transcriptomic, proteomic and imaging signals to create a fuller “biological fingerprint” of each patient. Vitalay explains how this can enable earlier insight from single-arm trials by effectively benchmarking against standard-of-care cohorts, helping teams enrich later-stage trials with the right subpopulations sooner, and reducing time and cost.
They also discuss the real blockers to adoption: not only scientific conservatism, but commercial uncertainty around how Big Pharma structures deals with tech-bio companies that bring platforms rather than single assets. Vitalay argues that explainability is non-negotiable in this space, because clinicians, scientists, patients, and regulators will not trust black-box predictions.
Topics Covered
Why AI is shifting from failure prevention to pipeline redesign

Biomarker discovery beyond binary responder vs non-responder labels

Breaking disease silos to learn across indications

Multimodal integration: DNA, RNA, protein, imaging, and clinical data

Using foundation models to bridge trial data and real-world data

Patient stratification and trial enrichment from early studies

Reverse translation and identifying unmet need before target hunting

Explainability, trust, and regulatory readiness

Adoption barriers: culture, champions, and deal structures for tech-bio

Misconceptions about AI in drug development and why “press a button” is a myth

About the Podcast
AI For Pharma Growth is the podcast from pioneering Pharma Artificial Intelligence entrepreneur Dr Andree Bates, created to help pharma, biotech and healthcare organisations understand how AI-based technologies can save time, grow brands, and improve company results.
This show blends deep sector experience with practical conversations that demystify AI for biopharma leaders, from start-up biotech right through to Big Pharma. Each episode features experts building AI-powered tools that are driving real-world results across discovery, R&D, clinical trials, medical affairs, market access, regulatory, insights, sales, marketing, and more.
Dr. Andree Bates LinkedIn | Facebook | X

Enterprise AI is shifting from experiments to infrastructure, and that changes everything. In this episode, Dr Andree Bates interviews Jocelyn Houle, Senior Director of Product Management at Securiti.ai, to explore the future of enterprise AI, agents, automation, and the single biggest blocker to scale: trust.

Jocelyn shares what she is seeing across highly regulated industries as organisations move beyond proof of concept into production. She explains why the agent era raises the stakes: when you add LLMs into workflows, systems become non-deterministic and harder to trace end to end. In her words, once the data goes in, you cannot easily untangle it, so organisations need stronger controls around permissions, auditing, and policy enforcement.

A major theme is that data foundations matter more than ever. Jocelyn warns that agents will not magically repair messy data, they will expose weak data quality immediately. From there, she outlines how trust can be won or lost at the prompt layer, both outbound (what the model says to customers) and inbound (what users share with the organisation). She also discusses “toxic combinations”, where overlapping access can accidentally leak sensitive information, plus the growing need for prompt screening and tracking to reduce risk.
The conversation also digs into explainability and auditability, with Jocelyn being refreshingly honest that the perfect solution is not here yet. Instead, enterprises are using practical approaches like benchmarking releases side by side, cataloguing AI agents in use, and building governance that is starting to look more like modern cybersecurity: baked in from the start, not added as an afterthought.
Jocelyn closes with clear advice for leaders: start “left” with raw data controls, build a truly cross-functional team, and begin setting up auditability even with imperfect tools, because regulators are catching up and they will expect responsible behaviour.

Topics Covered
Where enterprise AI adoption really stands today

What makes AI agents different from traditional automation

Non-determinism, traceability, and why permissions matter

Data mapping, policy controls, and reducing sensitive data leakage

Prompt security: outbound and inbound trust risks

“Toxic combinations” and exposure through agent workflows

Explainability, benchmarking, and parallel release testing

AI governance becoming as essential as cybersecurity

Top 3 pieces of advice for CTOs starting their AI journey

Why enterprise AI will become so embedded we stop talking about it

AI For Pharma Growth is the podcast from pioneering Pharma Artificial Intelligence entrepreneur Dr Andree Bates, created to help pharma, biotech and healthcare organisations understand how AI-based technologies can save time, grow brands, and improve company results.

This show blends deep sector experience with practical conversations that demystify AI for biopharma leaders, from start-up biotech right through to Big Pharma. Each episode features experts building AI-powered tools that are driving real-world results across discovery, R&D, clinical trials, medical affairs, market access, regulatory, insights, sales, marketing, and more.

Dr. Andree Bates LinkedIn | Facebook | X

About the Podcast

Clinical trials are a massive industry with brutal economics, long timelines, and failure rates that would be unacceptable in almost any other sector. In this episode, Dr Andree Bates is joined by Dr Joseph Geraci of NetraMark to break down why trials fail so often, how patient heterogeneity drives cost and uncertainty, and where AI can realistically shift the economics.
Joseph shares his unusual path from mathematics and mathematical physics into oncology and medical science, including a decision to move into hospital research rather than follow a more traditional academic route. That shift shaped his focus: not just discovering more molecules, but understanding why the same drug can work brilliantly for some patients and fail for others.
A central theme is that clinical trials are not “one disease, one patient type”. In many areas, disease definitions are too broad for trial design, making trials feel like trying to hit multiple dartboards with one dart. Joseph explains how NetraMark’s approach aims to identify meaningful subpopulations inside small datasets, finding the “pocket” where a drug’s true advantage shows up, without discarding patients as outliers.
The conversation also touches on regulators, including growing interest in innovation pathways, but also the fear pharma teams have about changing protocols and risking setbacks. Joseph argues that AI’s biggest economic value in trials is speed, using better insight from limited trial data to guide enrichment strategies, smarter substudy decisions, and faster iteration, especially in oncology and rare disease where time is everything.

Topics Covered
Why clinical trial economics are becoming unsustainable

Patient heterogeneity and why disease definitions break trials

Finding “pockets” of responders within small datasets

Trial enrichment and substudies that reveal a drug’s advantage

Why pharma adoption can be slow, even when failures are constant

Regulatory interest, guidelines, and sponsor risk aversion

Large language models vs mathematically augmented AI approaches

Speed as the biggest economic lever in trials

Practical examples across depression, schizophrenia, oncology, and beyond

What clinical trials could look like in five years with AI-driven insight

This show blends deep sector experience with practical conversations that demystify AI for biopharma leaders, from start-up biotech right through to Big Pharma. Each episode features experts building AI-powered tools that are driving real-world results across discovery, R&D, clinical trials, medical affairs, market access, regulatory, insights, sales, marketing, and more. 

Dr. Andree Bates LinkedIn | Facebook | X

AI capability has never been higher, yet most pharma AI programmes are still failing to create measurable business impact. In this solo episode, Dr Andree Bates breaks down why many pharma and biotech AI strategies are “broken before they even begin” and what a real AI strategic blueprint needs to include if you want adoption, scale, and outcomes, not just impressive pilots.

Dr Andree explains the core paradox: AI can now synthesise literature at speed, accelerate discovery, and outperform human experts in specific tasks, but the business results are often disappointing because the failure is rarely technical. It is strategic. She describes the “technical obsession trap”, where organisations spend months optimising models and benchmarking competitors while adoption remains low and teams are not operationally ready to act on the outputs.

She outlines three common failure modes:
Innovation Theatre, where disconnected pilots never compound into enterprise value

Competitor benchmarking, where companies copy use cases that do not fit their context

Technology first strategy, where tools are bought before priorities are defined

From there, she maps what a strong pharma AI blueprint must cover: grounding in business objectives, end to end deployment architecture (data, governance, capability, change), leadership and culture, rigorous financial modelling tied to revenue and ROI, and alignment across functions including commercial, medical, regulatory, R&D, market access, insights, and tech teams.

Dr Andree closes with a clear challenge for leadership: competitive advantage will come to organisations that build the most intelligent operating model around AI, not those with the biggest budgets. She also offers a 45 minute AI strategic diagnostic for pharma and biotech leaders who want an honest read on what to fix before investing further.

Topics Covered
Why pharma AI impact often disappoints despite powerful tools

The “technical obsession trap” and the AI strategy blind spot

Innovation Theatre, competitor benchmarking, and technology first mistakes

What a pharma AI strategic blueprint must include

Governance as a foundation for scale and regulatory trust

Leadership, culture, and adoption as the real differentiators

Financial modelling and prioritisation based on ROI and revenue impact

Organisational alignment across the full pharma value chain

Choosing the right advisory partner and avoiding generic frameworks

Why strategy must come before technology to build durable advantage

AI For Pharma Growth is the podcast from pioneering Pharma Artificial Intelligence entrepreneur Dr Andree Bates, created to help pharma, biotech and healthcare organisations understand how AI-based technologies can save time, grow brands, and improve company results.

This show blends deep sector experience with practical conversations that demystify AI for biopharma leaders, from start-up biotech right through to Big Pharma. Each episode features experts building AI-powered tools that are driving real-world results across discovery, R&D, clinical trials, medical affairs, market access, regulatory, insights, sales, marketing, and more.

Dr. Andree Bates LinkedIn | Facebook | X