By Nick Spittal, Chief Operations Officer, Velocity Clinical Research
How many times have you heard the term AI this week? Sponsors are investing in it, CROs are building for it, and tech vendors are popping up left, right, and center to sell it. Sites are fully in the crosshairs, with hundreds of hours of procurement conversations aimed at making them more productive, more efficient, and more accurate.
The problem is that most of those tools are built for someone else’s understanding of how trials run. Sites get handed systems designed by people who have never managed a screen fail spike or run three sponsor-mandated platforms alongside their own. The result is parallel workflows, duplicated data entry, and capable people spending time navigating software instead of running studies.
The proliferation of AI means that, inevitably, sites will be expected to use it. At Velocity, we decided to build exactly what we need. Over the last several years, we’ve invested in proprietary tools designed around the realities of trial delivery, software our own teams shaped around the problems they actually face. At the scale we operate, across 70+ sites globally, that investment compounds. A tool that works across the full network doesn’t just solve a problem once; it delivers small gains for hundreds of staff every day.
Building for the Network, Not the Brief
Payment reconciliation is a good example. The revenue cycle in clinical research is fragmented and manual, with invoice formats varying across sponsors, CROs, and sites. Payments are tied to milestones that shift over the course of a trial. Velocity’s finance team was one of the first to benefit from our AI investment, with a Gen AI solution that reconciled close to $75 million in payments and over a million line items in its pilot phase, automating what would otherwise have been nine months of manual work. Sponsors don’t interact directly with that tool, but they benefit from a partner that doesn’t have to dedicate skilled staff to wrangling spreadsheets.
The same is true of customer-facing operations, as well. We recently introduced a Feasibility Agent, which draws on longitudinal data across 70+ global sites (e.g. PI and staff experience, performance history, local disease prevalence, patient behavior, and enrollment rates) to assess site suitability. Sponsors and CROs receiving feasibility submissions from Velocity now get a probabilistic recommendation that shows how well-matched a site is to the protocol requirements, producing more reliable results downstream, informed by those closest to the data.
What a Site Evaluation Doesn’t Tell You
Building these tools has also taught us a lot about AI adoption. To work well, it must come from two directions: leadership setting a culture of trying new things and modeling behavior, and teams building solutions to the problems they face every day. The organizations that get it right consolidate those efforts, so they can adopt from end users doing the work, and don’t end up with a proliferation of point solutions, but instead build towards an enterprise intelligence. Critically, they design AI to absorb the work people least want to do, so staff feel supported rather than replaced. As an example of this approach, Velocity is embarking on a weeklong AI hack-a-thon in July building agents and solutions from ideas raised by team members across the entire organization.
The fact that all of this is still invisible in a standard site evaluation is problematic. AI in clinical research will keep moving fast, and the networks investing now are building a data and operational advantage that grows with every study. The ones waiting to be “sold to” will inevitably fall behind. It’s worth knowing which kind you’re partnering with.
