AI is moving quickly across life sciences, but speed is different from readiness. In regulated industries, the real test is not whether an organization can launch a pilot. It is whether the organization can explain the use case, assign ownership, manage the risk, and keep learning after the first deployment. That work is less visible than the technology itself, but it is where sustainable value is created.
The 95% Problem
AI adoption in life sciences is accelerating. Production impact is not keeping pace. Estimates suggest that 80 to 95 percent of pharmaceutical AI pilots never reach production (Sakara Digital, 2026). At the same time, AI-agent adoption in life sciences has surged from 11 percent to 54 percent in just 18 months (KPMG’s Generative AI transformation in life sciences 2026 Pulse Survey). Yet 87 percent of leaders are scrambling to improve their workforce, and 63 percent still require human validation of AI outputs (KPMG, 2026).
Those numbers tell a story I have seen many times in transformation work. The technology may be new, but the failure pattern is familiar. The issue is rarely only the model. More often, it is unclear ownership, uneven adoption, a weak change management discipline, and a governance process that arrives after work is already in motion.
Why Most AI Initiatives Stall
After decades of leading transformation in regulated environments, I have learned to first look at the operating conditions around the organization’s technology. AI initiatives in pharma often stall because the enterprise has not answered a few practical questions:
- Where is AI already being used? Shadow AI can spread through workflows before leaders have a reliable inventory of use cases.
- Who owns the decision when something goes wrong? A steering committee may sponsor the effort, but accountability still needs to land with a named person who has authority, context, and capacity.
- What happens after the pilot? A model that is acceptable at launch still needs monitoring, change control, requalification, and a defined path to retirement.
These are not new concerns. They are closely aligned with the FDA’s Good Machine Learning Practice principles, including total product lifecycle management, human-AI team performance, and continuous real-world monitoring (U.S. Food and Drug Administration, 2025). The EU AI Act, with high-risk requirements phasing in through December 2027, also makes meaningful human oversight a design expectation rather than a late-stage compliance activity (USDM, 2025).
The direction from regulators is not the hard part to understand. The harder work is building habits, roles, and review practices that people can use while the business keeps moving.
A Practical Way to Start
The strongest AI governance work I have seen does not start with a new bureaucracy. It starts by adapting disciplines pharma already knows. GxP risk thinking, Prosci ADKAR, and PMI benefits realization can give leaders a practical foundation without making teams feel as if they are starting over.
First, build a use-case inventory before debating the model. Capture the intended use, data inputs, business process, vendor, accountable owner, and any boundaries the system must not cross. A spreadsheet is enough at the beginning. What matters is that leaders can see the work and discuss it with the same facts.
Next, sort the use cases by risk in language the organization already understands. A marketing content assistant should not carry the same review burden as a clinical decision support tool. I would look at GxP impact, patient or product risk, data sensitivity, and whether the use case appears to fall into a limited or high-risk category under emerging AI regulation.
Next, treat AI-enabled capability as something that will change over time. Document the training data and assumptions, define change-control triggers, monitor drift, revisit performance, and decide in advance when the system should be paused, requalified, or retired. Traditional validation discipline still matters, but AI requires leaders to pay closer attention to what happens after go-live.
Finally, make human oversight real. For higher-risk use cases, someone must be able to interpret the output, challenge it, override it, or stop the process. That person needs to be named in the workflow, not implied in a committee charter.
The Change Leader’s Contribution
One example from my consulting work, later featured in Dr. Harold Kerzner’s Global Project Management Playbook, highlights this dynamic. In a Fortune 500 biopharmaceutical organization facing IT operational inefficiencies, the answer was not simply to introduce another tool. The progress came from combining PMBOK-based project management, Prosci ADKAR change practices, and Lean Six Sigma methods in a way people could follow (Kerzner, 2026). Reduced operational costs, standardized workflows, and stronger documentation came from disciplined execution and adoption.
That experience still shapes how I think about AI. Tools can accelerate work, but they cannot compensate for ambiguous decision rights, lack of sponsorship, or processes that people do not trust.
What Change Leaders Can Do This Quarter
- Start with an AI use-case inventory. Within 30 days, identify where AI is being used or seriously considered. Keep the format simple. The point is visibility, not perfection.
- Bring Quality, IT, Legal, and key business leaders into the same discussion. Select the most important use cases, classify the risk, and document the rationale so future decisions are easier to defend.
- Name one accountable person for each higher-risk use case. Committees can advise, but ownership should be clear enough that teams know who to go to when judgment is needed.
- Measure adoption as well as performance. Usage, override frequency, exception patterns, and user trust can reveal whether the capability is creating value or simply adding another layer of work.
AI will continue to accelerate. The organizations that benefit from it will be the ones that operationalize it most effectively using disciplined governance, human-centered design, and sustained change leadership.
Cheryl Howard is Principal Consultant at Howard Consulting LLC, where she advises Fortune 500 life sciences and healthcare organizations on AI-enabled transformation, governance, and structured change leadership. She holds PMI-CPMAI, PMP, and Prosci ADKAR credentials and serves as a PMI Standards+ volunteer (2026). Her work is featured in Dr. Harold Kerzner’s Global Project Management Playbook, Frameworks, Methods, and Best Practices. Previously, she served as IT Strategy Director at Gilead Sciences.
References
Global Project Management Playbook: Frameworks, Methods, and Best Practices, by Harold Kerzner, 2026; Howard Consulting: IT Operational Biopharmaceutical Efficiency case study; Wiley.
Generative AI Transformation in Life Sciences: Q1 2026 Pulse Survey, KPMG International, 2026
https://kpmg.com/kpmg-us/content/dam/kpmg/pdf/2026/generative-ai-transformation-in-life-science.pdf
“From AI Pilots to Production: Why 95% of Pharma AI Projects Fail” article, Sakara Digital, April 4, 2026
https://sakaradigital.com/blog/pharma-ai-pilots-fail-production-3/
“EU AI Act Compliance for Pharma and Life Sciences: What to Prepare Before August 2026” article, USDM Life Sciences, July 9, 2025
https://www.usdm.com/resources/blogs/the-eu-ai-act
“Good Machine Learning Practice for Medical Device Development: Guiding Principles,” U.S. Food and Drug Administration, December 19, 2025.
