Life Sciences is investing heavily in AI but investment is not the same as impact. While 90% of organizations are increasing their AI spend, 72% of surveyed Life Sciences CIOs say their organizations have either broken even or generated a negative ROI (Gartner 2025).
That gap should be a wake-up call for clinical data sciences. The question is no longer where AI can automate another task. It is whether organizations are bold enough to redesign the work, roles and behaviors around it, to move beyond isolated productivity gains and start delivering long-lasting value.
The boundaries of clinical data work are changing.
Clinical development has traditionally relied on highly siloed, specialist functions with relatively defined responsibilities.
Data Management ensures clinical data is collected, cleaned and of high integrity. Clinical Programming transforms that data into standardized datasets and outputs. Biostatistics defines and applies the statistical methods through which evidence is interpreted. AI does not make those disciplines irrelevant. It makes their interdependencies increasingly important.
The use cases already emerging illustrate why.
In Data Management, AI and LLMs can support medical coding, automated data review, query management and standards mapping. In Clinical Programming, opportunities extend across code generation, SDTM and ADaM dataset creation, automated validation, quality control and TFL generation. In Biostatistics, applications include automated SAP drafting, predictive analytics, decision support and safety monitoring.
Individually, these can look like opportunities to automate specific activities. Collectively, they point toward a bigger trend: the clinical data workflow itself is becoming more connected, intelligent and increasingly orchestrated through AI.
This creates a more connected clinical data workflow.
A change in how data is reviewed can influence query management. Standards mapping affects downstream dataset creation. Programming and validation decisions ultimately influence tables, listings, figures and regulatory outputs.
AI therefore creates an opportunity not simply to optimize each activity, but to reconsider the workflow connecting them. That is where the transition from clinical data delivery to clinical data science starts to become real.
Where human contribution is moving upstream.
One of the most misconstrued narratives around AI is that machines perform more work and people perform less. This misses the point. As AI takes on more execution, human contribution moves upstream.
For a data manager, value increasingly comes not only from identifying an anomaly but from assessing the significance of an AI-identified signal, understanding its context and deciding what requires intervention.
For a clinical programmer, expertise moves beyond writing code toward directing AI-generated code, validating its logic, understanding dependencies and ensuring that outputs remain robust.
For a statistician, AI can accelerate analytical activity, but the value of statistical expertise becomes even more concentrated around study context, interpretation, scientific judgment and the decisions that follow. While execution decreases, judgement does not.
This calls for an evolution of capabilities.
Clinical data professionals increasingly need to know when an AI-generated output can be trusted, when it should be challenged, what an exception means, and when human expertise needs to take precedence.
The future clinical data professional will not simply be an expert in a task. The future clinical data professional will increasingly need to understand the system around it.
This requires a different combination of capabilities. AI literacy matters. But so do critical thinking, systems thinking, adaptability, curiosity and accountability.
An operating model shift, not just a skills shift.
Organizations cannot solve this transition through training alone. If roles change but workflows do not, AI will simply be layered onto existing processes. If activities become more automated but accountability remains unclear, faster execution may create more review and reconciliation downstream.
If Data Management, Programming, Statistics, Data Science and Technology each develop independent approaches to AI, we risk automating the silos we should be breaking down.
The operating model has to evolve with the technology. Teams need greater clarity around who directs AI-enabled work, who validates outputs, who owns exceptions and who ultimately remains accountable for decisions. Ways of working need to become more connected as activities that were traditionally manual and sequential become increasingly automated and interconnected.
The future is not human or AI
What should technology execute? Where is human judgment essential? How should those two forms of contribution come together across an end-to-end clinical data workflow?
AI can take on more of the generation, transformation and initial review of information. Human expertise can move further toward interpretation, scientific judgment, quality stewardship and decision-making.
That does not make clinical data expertise less important. It makes expertise more critical in different places.
The future of clinical data sciences will not be defined simply by how much work we can automate. It will be defined by how intelligently we redesign the work around it.
Next in the series: Beyond Automation: The Rise of Agentic Clinical Data Workflows.
Sources
- Gartner. (2025, March). Top Technology Trends for Life Sciences CIOs for 2025. Whitepaper. https://www.gartner.com/en/documents/6258551
- Cleary, E. (2026, July). Value Starts with People: Turning AI into Impact in Clinical Data Analytics. PHUSE SDE Presentation. https://phuse.s3.eu-central-1.amazonaws.com/Archive/2026/SDE/US/West+Windsor/PRE_WestWindsor06.pdf
