In this episode of Innovation to Impact: Ruminations & Ramblings, Szczepan Baran, Brian Berridge, and Nick Kelley tackle one of the biggest problems in modern drug development: we keep adding more technology, more data, and more complexity, yet clinical attrition remains painfully high. Across discussions on AI, NAMs, digital biomarkers, animal models, translational science, and organizational culture, they argue that innovation fails when tools become the strategy instead of serving clearly defined patient-centered decisions. The conversation explores why reverse translation matters, how AI should function as a decision-support system rather than a magic oracle, why “decision warranties” may become essential in AI-enabled science, and how the industry continues to confuse activity with progress. This is a candid, often uncomfortable discussion about predictivity, accountability, translational learning, and what it would actually take to build a drug development system optimized for patient outcomes instead of platform hype.

Click here to watch a video of this episode.
Drug development workflows only get heavier, and subtractive change requires predictivity and decision-grade evidence that can survive regulatory science scrutiny. In this episode we talk about how AI and translational science can earn the right to delete steps, not decorate them. The tension is uncomfortable: are you willing to remove something you have always done, or will you just add another layer and call it progress? We introduce the idea of a Predictivity Ledger, a simple way to make miss rates visible and force clarity about what is being claimed, in what context, and with what failure modes. Deletion is not a vibe. It is a governance decision with receipts. Takeaway: pick one legacy step, define deletion criteria, and start logging misses like they cost time and money, because they do.

• (00:00) - Episode purpose, resolutions, and AI adoption risks
• (02:00) - Subtractive strategy and replacement opportunities in practice
• (04:00) - Evidence authority, familiarity, and higher validation bars
• (07:00) - Add-on endpoints and why adoption stalls
• (08:00) - Decision bias and the need to challenge interpretation
• (10:00) - Predictivity ledger components for accountable decisions
• (12:00) - AI capabilities, limits, and accountability requirements
• (14:00) - Administrative AI example and subtractive realization questions
• (18:00) - Process efficiency versus biology, fast-fail critique
• (25:00) - Breaking silos, expert finding, and organizational knowledge
• (29:00) - Historical knowledge, foundational literature, and AI leverage
• (33:00) - Closing remarks and next episode sign-off

Click here to watch a video of this episode.
In drug development, go/no-go decisions concentrate translational risk and expose real consequences. When AI or digital biomarkers influence that call, decision-grade evidence in a regulated environment is not optional. This episode sits in the moment every executive recognizes: the slide is on the screen and someone asks, “Do we advance?” Here’s the tension. We love the word validated, but what happens when the next dataset disagrees? We introduce a practical discipline we call the decision warranty: clear scope, clear evidence chain, clear boundaries, and explicit triggers for pause, rerun, or escalation. Someone has to own that call. Takeaway: if you cannot write the stop triggers and the decision owner on one page, do not let the tool move the decision.

• (00:00) - Forward Looking Themes for 2025
• (01:00) - Reproducibility Crisis and Foundational Biology
• (02:00) - Digital Measures and Preventive Health
• (03:00) - Multimodal AI and Foundation Models
• (06:00) - Closed Loop Data Generation and Innovation
• (09:00) - Decision Making as a Core Capability
• (13:00) - Human in the Loop and Accountability
• (16:00) - Micro Physiological Endpoints and Decision Use Limits
• (21:00) - Democratizing Data and Model Access
• (24:00) - Biology as Foundation for Modeling Systems
• (28:00) - Opportunity, Effort, and Due Diligence