Generative AI has moved past the pilot stage in pharma. Molecules are being designed by algorithms, trial protocols are being simulated before a single patient is enrolled, and regulatory submissions are being drafted with AI assistance. Yet a striking data point from MIT's 2025 research should give every Product and Program Manager pause: Nearly 95% of enterprise GenAI pilots failed to deliver measurable business impact — not because the models were weak, but because they stayed disconnected from real workflows, data foundations, and organizational ownership.

That gap is the real opportunity. The technology exists. What's missing is a delivery model built to capture its value. A recent milestone shows what capturing it looks like: On August 19, 2026, Moderna and Merck announced that their personalized mRNA melanoma vaccine, intismeran, combined with Keytruda®, met its primary and secondary goals in a Phase 3 trial of over 1,100 high-risk melanoma patients - slowing recurrence and distant spread of the disease after surgery. It's the first randomized Phase 3 trial to demonstrate a benefit from a personalized "neoantigen" cancer vaccine, and it validates a decade-long computational pipeline: sequencing a patient's own tumor mutations, using algorithms to predict which mutations the immune system will recognize, and generating a bespoke mRNA vaccine for that individual - all inside a manufacturing timeline that makes personalized treatment commercially viable rather than a lab curiosity.  

Here's how program managers in life sciences should rethink their roadmaps in light of this shift.

1. Stop treating GenAI as a feature; treat it as a workflow redesign

Bolting a chatbot or summarization tool onto an existing process rarely moves the needle. The programs generating real returns succeeded because AI was embedded into the discovery and design workflow itself, not layered on top. The Moderna-Merck program is the clearest case: neoantigen prediction, sequence design, and per-patient manufacturing scheduling are all computational steps folded directly into the treatment pathway, not a side tool consulted occasionally.

Insilico's AI-originated candidate reaching Phase II in roughly three years instead of the usual twelve to eighteen tells the same story. As a PM, this means your discovery-phase deliverable shouldn't be "evaluate a GenAI tool." It should be "redesign the target-identification-to-candidate workflow with AI as a native step," with success metrics tied to cycle time and hit rate, not tool adoption.

2. Build data readiness into the roadmap as its own workstream

Most failed pilots trace back to fragmented, poorly governed data.

Before committing a program to GenAI-driven molecule generation or trial simulation, budget a genuine data-foundation phase - normalized genomic, proteomic, and clinical datasets, clear lineage, and access governance. Treat this the way you'd treat infrastructure in a platform migration: a prerequisite gate, not a parallel track that can slip.

3. Reframe trial design as a computational sprint, not a linear milestone

Virtual patient twins, synthetic control arms, and adaptive protocol simulation let trial design become iterative, rather than a single big design-freeze event. For program roadmaps, this means shorter, simulation-driven design sprints ahead of protocol lock, with computational biologists and GenAI engineers embedded alongside clinical operations from day one - not consulted after the protocol is drafted.

4. Plan for "batch of one" manufacturing and logistics

The Moderna-Merck program isn't just a discovery win - it's a delivery-model win. Personalized vaccines mean each patient's product is unique, so the program had to solve individualized manufacturing scheduling, cold-chain logistics, and quality control at scale, not just a single validated formulation. PMs running similar programs need roadmaps that treat "per-patient production" as its own workstream - supply chain, QC, and regulatory sign-off processes built for continuous individualized batches rather than a handful of large, standardized lots. This is a fundamentally different program shape than a traditional drug launch, and it needs to be planned as one from the start, not retrofitted once trial data reads out.

5. Create new risk and governance checkpoints

AI-generated candidates and AI-assisted regulatory content introduce risk categories most program plans don't yet account for: model validation, explainability for regulators, and audit trails for AI-assisted decisions. Build these as formal stage gates in your program roadmap, the same way you'd gate a safety review, rather than treating them as compliance afterthoughts.

6. Structure cross-functional squads, not tool rollouts

For organizations to capture, they should value pair computational scientists, GenAI/ML engineers, and domain (chemistry, biology, clinical) experts in standing squads with shared KPIs – just have not a central AI team serving requests from disconnected business units. For PMs, this changes resourcing plans: fewer one-off "AI proof of concept" projects, and more permanent squad structures owned jointly by R&D and Technology.

7. Rethink portfolio prioritization

With AI compressing certain phases (discovery, early trial design) more than others (regulatory review, manufacturing scale-up), portfolio-level roadmaps need asymmetric timelines. Don't apply a uniform "AI will speed everything up 30%" assumption - model each phase's compression potential separately when sequencing your program portfolio.

The Takeaway

GenAI's opportunity in pharma R&D isn't a technology question anymore - it's a delivery model question. The programs that win will be the ones where product and program managers redesign workflows, data foundations, governance gates, and team structures around AI from the start, rather than retrofitting it onto roadmaps built for a pre-AI world.

Disclaimer: AI chatbots were used to help perform research, gather contents and summarize thoughts into better and succinct language.

References:

- MIT NANDA, The GenAI Divide: State of AI in Business 2025 (July 2025): The source of the "95% of pilots deliver no measurable P&L impact" finding, based on 300+ initiative reviews, 52 interviews, and 153 survey responses.  ai_report_2025.pdf

- Moderna and Merck, joint press release, August 19, 2026: Announcement that the Phase 3 trial of intismeran (personalized mRNA cancer vaccine) plus Keytruda® met its primary and secondary endpoints in resected stage high-risk melanoma. Merck and Moderna Announce Phase 3 INTerpath-001 Trial of Intismeran Autogene Plus KEYTRUDA® Met Endpoints of Recurrence-Free Survival (RFS) and Distant Metastasis-Free Survival (DMFS) in Patients With Completely Resected Stage IIB-IV Melanoma