Beyond the Demo: What It Really Takes to Scale AI in Clinical Research

Reflections from the 2026 Veeva R&D Summit

Sweta Pittala
Vice President, Technology Portfolio, IT R&D

AI is advancing rapidly across clinical research, but moving from promising demonstrations to sustainable, real-world impact remains a significant challenge.

At the 2026 Veeva R&D Summit, Chief Operations Officer Nicholas Spittal and I shared Velocity Clinical Research's experience putting AI into production, including the lessons we've learned along the way. What made the discussion especially interesting was hearing where other organizations are in their own AI journeys.

Where is the industry today?

We opened by asking the audience: Where is your organization today with AI in clinical operations? The responses offered a revealing snapshot:

This was an informal poll, but the message was clear: AI enthusiasm is widespread, yet scaling beyond pilots is still a work in progress.

We then asked a second question: What makes AI hardest to operationalize? The word cloud was telling. Trust, security, cost, legal considerations, hallucinations, and training stood out, alongside validation, alignment, and other operational concerns. Notably, these challenges aren't mainly about what AI models can do. They're about what organizations need to build around AI to make it work reliably.

What operationalization has taught us at Velocity

We've deployed 26 AI agents across our operational workflows, and the experience has reinforced an important distinction: automating a task is not the same as transforming a business process.

To illustrate, we shared our accounts receivable operations (AROps) implementation. The goal was to prevent revenue leakage, improve invoicing, and strengthen cash collection by connecting contractual terms, operational activities, billing triggers, and payment reconciliation. On paper, the process looked straightforward. In practice, it meant navigating inconsistent contract formats, connecting information across systems, interpreting operational triggers, and handling exceptions.

Several lessons emerged that reach well beyond AROps:

1. Connected workflows create real value. Individual AI agents can perform tasks effectively, but meaningful business outcomes depend on how those capabilities work together across systems and processes.
2. Exceptions need to be designed for, not worked around. Real-world operations rarely follow a perfect sequence. Missing information, inconsistent data, and ambiguous conditions call for clear escalation paths and human oversight.
3. AI economics can't be an afterthought. Once AI is in production, document complexity, repeated processing, and token consumption can significantly affect costs. Technical performance has to be matched by economic sustainability.
4. Adoption is as important as automation. One of our more surprising findings was that employees sometimes perceived AI-enabled workflows as slower, even when time studies showed measurable improvements. Efficiency alone doesn't guarantee adoption. Trust, transparency, training, and change management matter just as much.
5. Autonomy must be earned. Our focus has shifted from simply increasing automation to improving accuracy, recognizing uncertainty, and knowing when human intervention is appropriate. Reliable AI isn't about eliminating human judgment; it's about applying it where it matters most.

Where do we go from here?

The audience discussion reinforced what we've experienced firsthand: scaling AI is as much an operational and organizational challenge as a technology one. Real progress will take collaboration across research sites, sponsors, CROs, and technology partners to reduce unnecessary variation, improve interoperability, and design workflows that work for both people and technology.

At Velocity, our ambition isn't simply to deploy more AI agents. It's to create connected, intelligent workflows that improve operational performance while giving our teams more time for patients, quality, and higher-value decisions.

Ultimately, the measure of AI success isn't how impressive the technology looks in a demo. It's how reliably it delivers value in everyday operations. AI can only scale as well as the people, processes, and ecosystem around it.

Thank you to everyone who joined our session at the Veeva R&D Summit and contributed to the conversation. It was encouraging to see such a shared focus on moving beyond experimentation toward practical, responsible AI adoption.