As the life sciences industry races to implement AI across clinical development, many sponsors are about to discover a fundamental problem: the data AI needs most isn’t in their systems.

The industry has spent decades building visibility into sponsor-side operations through systems like EDC, CTMS, and eTMF. These platforms capture study outcomes, milestones, and documentation. But they miss something critical: the day-to-day operational work happening at research sites.

Visit preparation, consent discussions, document review cycles, staff workflows, and regulatory readiness rarely exist in structured systems of record. They live in email threads, paper binders, spreadsheets, and the institutional knowledge of coordinators and site staff. Sponsors only see what eventually enters their systems. Everything upstream of that entry point remains largely invisible.

This creates what Florence calls the data shadow: a hidden layer of operational activity that impacts trial outcomes without ever becoming visible to the organizations responsible for them.

Sponsors see the shadow every day. They see delayed activations, missed enrollment targets, protocol deviations, and site performance variability. What they don’t see are the site-level workflows casting those shadows.

The problem isn’t AI. It’s the data shadow.

When pharmaceutical teams discuss AI, the vision is compelling. The ability to predict site performance, flag enrollment risk, accelerate activation timelines and reduce monitoring burden appear enticing.

They are valuable use cases. But they all depend on data that most organizations don’t currently collect.

Traditional sponsor systems excel at measuring outcomes. They can report enrollment rates, protocol deviations, query volumes, and activation milestones. What they often cannot see are the process signals that drive those outcomes: how quickly documents move through review cycles, whether visit preparation is completed on time, how consistently workflows are followed, or where operational bottlenecks emerge before performance suffers.

In Florence’s analysis, these operational process signals represent a needed source of untapped intelligence. Depending on the use case, incorporating site-level workflow data can increase the amount of usable operational signal by 40–80%. Site performance prediction models, for example, may gain approximately 60% more signal, while site activation readiness models may gain as much as 80%.

The implication is important: clinical AI models are frequently trained within a data shadow. They see outcomes while remaining blind to the processes that produced them. The result is an industry trying to predict behavior from traces instead of causes.

That’s not an AI limitation. It’s a data infrastructure limitation.

Clinical trials have no persistent operational memory

The challenge runs deeper than missing data points.

Today’s clinical trial ecosystem is built around studies, not operational continuity. The only time meaningful site workflow data is typically captured is when a sponsor deploys a specific technology solution for a specific protocol. When the study ends, the visibility ends with it.

The next trial often begins with little understanding of how that site actually operates, what workflows perform best, where delays emerge, or which operational patterns predict success. Years of practical experience disappear into disconnected systems and one-off implementations. As a result, the sponsor repeatedly starts over with each new study.

AI exposes this problem because machine learning systems depend on persistent operational memory. They improve when organizations continuously capture, retain, and learn from process-level data across studies. Most clinical trial infrastructure was not designed for that purpose.

The industry’s AI ambitions are colliding with its lack of operational continuity.

The digital adoption gap is widening

Reaching the promise of AI requires unlocking data from sites through digitization. The question is: how prepared are sites today?

Florence’s 2026 State of the Industry Report mapped clinical trial technology adoption across the Geoffrey Moore technology adoption curve and found the industry at an inflection point. Approximately 24% of sites qualify as technology leaders, while another 24% remain technology laggards. Between them sits a meaningful adoption gap.

Technology-forward sites are already generating structured operational data through digital workflows. Paper-based sites are not. As AI capabilities improve, organizations with digital infrastructure will compound their advantage while others fall further behind.

For sponsors, this creates a significant constraint: the industry’s current approach effectively limits AI readiness to roughly one-quarter of sites that have invested in digital workflows on their own.

Interestingly, the ¼ are in jeopardy, too. Sponsors continue to rely on a setup-and-takedown model, deploying study-specific technologies, competing with the most reliable study-to-study infrastructure already available at these “Leader” sites. Meanwhile less mature sites frequently fail to adopt sponsor tech, tools were designed primarily for sponsor requirements rather than site operations. 

The result is a fragmented technology strategy that struggles to serve either audience. One group doesn’t want the technology being deployed. The other doesn’t meaningfully adopt it. Meanwhile, sponsors remain trapped in a cycle where operational visibility disappears at the end of every study and must be rebuilt from scratch in the next.

A digital mandate for clinical trials in the AI era

Sponsors seeking to unlock AI’s potential should focus first on building a foundation of persistent operational intelligence.

That requires a site-conscious approach to digitization. In doing so, sponsors improve probability of digital adoption across their site networks.

1. Capture existing site intelligence

For digitally mature sites, the priority is integration rather than replacement.

Sponsors should connect to the systems sites already use and leverage the operational data being generated every day. This creates visibility into workflows without disrupting established processes.

2. Digitize site workflows

For less mature sites, the first step should be digitizing workflows that exist across every study.

Regulatory document management through eReg/eISF systems is a natural starting point. It is universal, operationally critical, and easy to standardize. More importantly, it creates the foundation for broader workflow digitization and future AI applications.

3. Build persistent operational memory

The ultimate goal is not merely digital adoption. It is creating a durable layer of operational intelligence that persists beyond any single study.

When site workflows are continuously captured, sponsors gain insight not only into the current trial but into future site selection, startup planning, risk identification, and operational forecasting. Each study contributes to a growing body of intelligence rather than disappearing when the protocol closes.

Thereby eliminating the data shadow and providing the foundation AI requires to operate successfully.

Underlying all three mandates is a simple reality: sites must be willing to use the technology generating the data.

Sponsor-mandated tools may satisfy study requirements, but they rarely create the sustained engagement needed to build a meaningful operational data layer. The strongest signal of future success is often site preference itself. For example, Florence has achieved 90% eISF adoption rates, suggesting that when technology aligns with site workflows, participation follows, and the data foundation required for AI can finally take shape.

What’s next? The 12% signal

According to Florence’s 2026 research, only 12% of respondents report using AI consistently today.

At first glance, that number appears small. In reality, it may represent the leading edge of a significant competitive shift.

The organizations moving now are not simply experimenting with AI. They are building connected workflows, digitizing operational processes, and creating the data infrastructure necessary for AI to improve over time. Every workflow digitized today becomes training data tomorrow. Every process captured contributes to a growing intelligence advantage.

It’s clear that the data flywheel rewards early participants.

AI will not create an advantage for sponsors that simply purchase new technology. It will create an advantage for organizations that build the operational foundation required for those technologies to learn.

The question is no longer whether AI will transform clinical operations.

The question is which organizations will have the data, infrastructure, and operational memory required to benefit from it.