A few years ago, most of my conversations with clinical development leaders about AI started with some version of “Does this matter?” Today, that question is settled.
Teams are already using AI to draft regulatory documents, support data review, inform trial design and make sense of increasingly complex evidence. The question leaders ask me now is much harder: What happens when a promising pilot has to become part of how clinical work actually gets done?
For most organizations, that’s where progress stalls. Pilots are easy to start and hard to scale, and plenty of clinical AI efforts prove the concept but never become a validated, repeatable part of the process. Adoption is everywhere, but the ability to scale within most environments is still a challenge.
The gap is rarely the algorithm
Don’t get me wrong: Many clinical AI use cases are genuinely promising, with real value behind the buzz. Yet when a pilot fails to scale, the model is rarely the problem. What I see over and over again is an environment that can’t support the level of trust clinical work demands.
Clinical teams work across multiple tools, languages and systems, and every handoff is a chance to lose traceability and oversight. As AI speeds up analytical work, manual reconciliation, disconnected validation processes and fragmented documentation get harder to sustain.
The research bears this out. In the 2026 Data and AI Impact Report from SAS and IDC, only 17.5% of organizations across industries report running their data infrastructure at the highest maturity level. Most are deploying AI on infrastructure that can’t fully support the explainability, lineage or validation agentic AI requires.
Organizations that do the foundational work are seeing the difference. In life sciences, the same report found that 55% of those with the strongest trustworthy AI practices expect strong or high ROI from AI, compared with just 11% of those with the weakest. A better model won’t close that gap.
The clinical standard doesn’t relax for AI
Anyone who works in clinical development knows the questions: Where did the data come from? How was the analysis performed? What changed along the way? Can the result be reproduced? And the one I hear most: Can we trust the data, the environment and the output together? AI doesn’t loosen those expectations. It raises them.
The answer isn’t less oversight or waving a magic wand – it’s building traceability, reproducibility and governance into the way work happens, so no one has to reconstruct them later. Done correctly, governance stops being a brake. Teams move faster because they know how an output was produced and whether it’s fit for use, and they deliver results that regulators, clinicians and ultimately patients can trust.
Modernization isn’t an IT project anymore
Here’s the catch: Trust is hard to build into an environment that wasn’t designed for it. Most clinical environments predate AI, multilingual programming and today’s pace of analysis.
Those environments got us here, but they won’t take us further.
A modern clinical framework puts data, code and documentation in one governed place, so the record builds as the work happens. Teams can work in the languages they know, and AI runs inside validated workflows instead of beside them.
To be frank, modernization isn’t a “just” technology refresh. It’s a leadership decision, and it’s what makes scaling AI possible.
Before you modernize, find out if you’re ready
This is the part I wish we talked about more. Modernization can stall too, often because we jump to the solution before taking an honest look at where we’re starting from. Put a new platform on top of fragmented data, unclear ownership and inconsistent processes, and you’ll have the same problems in a newer environment.
Before anyone starts comparing platforms, I’d encourage leaders to ask their teams a few uncomfortable questions:
- If a regulator asked how a result was produced, could we show them without a scramble?
- Where does traceability break down when work moves between tools, languages or teams?
- Could our current environment support one more AI use case without one more manual workaround?
- Who is accountable when an AI-assisted output informs a decision?
- Have we given up any rights to our data to get AI-driven insights, and what did that compromise?
- Was our governance built with AI in mind, or are we stretching a process that predates it?
If the answers change depending on who you ask, that’s not a failure. That’s your starting point.
To help leaders get a clearer picture, SAS worked with Amazon Web Services (AWS) to develop the Life Sciences Modernization Readiness Assessment. It evaluates an organization’s current maturity level and provides a custom report to help guide its modernization journey. Wherever you land, you’ll have a better sense of what to tackle first.
Readiness comes down to the patient
It’s easy to talk about all of this in terms of platforms, pilots and processes. But every validated output, every submission and every safety signal eventually reaches a person.
For someone waiting on a new therapy, an avoidable delay isn’t an operational metric. And a result no one can trust isn’t a result anyone should act on.
We’ve settled whether AI matters. The question now is whether your clinical environment is ready to make it work at scale – for your teams, for regulators and, most of all, for the patients counting on you.