Before Adding AI, Fix The Process

By Dan Schell, Chief Editor, Clinical Leader

Before DPHARM, I spoke with Kaniel Cassady, Ph.D., executive director, Clinical Science Strategy and Excellence at Regeneron, about the senior leadership panel he was scheduled to join on “doing more with less” in clinical operations. With six people on the panel, Cassady knew there was no guarantee every point he wanted to make would find its way into the conversation. So, I asked him what he would talk about if given the opportunity.

Not surprisingly, AI was on his list. But as our conversation went on, I realized he wasn’t talking about simply the wonders of AI; he was focused on what should happen before AI even gets anywhere near a clinical development workflow. “There’s so many ‘shiny objects’ out there today that are AI enabled,” he said. “Regeneron’s approach, is to rethink the process first and then bring the technology on top of it.”

I’ll admit, that whole “investigation of the process” concept seemed kind of … well, boring (Sorry, Kaniel) when I first heard it. But as he went on, I tried putting myself in the shoes of some of the clinical development/sciences execs who may have been in the audience listening, and then it hit me. You don’t just one day wake up and say, “We’re going to add this piece of technology to our company” without first considering how that change will affect your current processes.

Find The Bottleneck First

Cassady has watched Regeneron grow rapidly, particularly in hematology, where he said the portfolio expanded from roughly five clinical trials to more than 50 in about six years, including more than 20 Phase 3 studies. Supporting that growth required more than simply adding people.

Cassady described their three-step approach:

1. Understand the work.
2. Identify the bottlenecks.
3. Design the organization and technology around solving those problems.

I figured in ClinDev, one of the biggest obstacles is actually writing the clinical documents (e.g., protocols). While that may be true, that’s not the bottleneck that caught Cassady by surprise. “The biggest bottleneck in document generation is the review cycles,” he said. “We still do two-dimensional authoring. It’s like how I wrote my thesis or how we wrote papers back in school.”

Sure, you can use LLMs to accelerate the drafting of protocols, synopses, and CSRs (clinical study reports), but that does little good if the document then crawls through multiple rounds of reviews. As a writer who has created many documents over the years, his questions regarding the process were refreshingly basic: Does everyone need to review every section? Should reviews be divided by expertise? Can previous reviewer feedback be used to identify problems earlier?

You know what isn’t mentioned when asking these types of questions at this point in a process? The technology. That comes later.

He said Regeneron is applying the same thinking to areas such as site monitoring. Rather than find an agentic monitoring tool and then determine what to do with it, teams are looking at the process end to end, identifying gaps, and then determining whether technology belongs there.

Do Smaller Biotechs Really Have An Edge?

As he described all of that, I started thinking about the smaller biotech executives who may have been in the DPHARM audience. If you have 20 processes being handled by 20 different groups, who is supposed to wrangle all of that together? Cassady said one of the best ways to tackle this is for leaders to start within their own functions and work outward. He added that, in some cases, this kind of change may be easier for smaller companies since they have less people and red tape.

Still, I’m not completely sold on the idea that being smaller automatically makes transformation easier. Fewer legacy processes don’t magically give a biotech more people, money, or time. But Cassady’s point is fair: Smaller organizations may have less organizational baggage to overcome.

He also believes clinical development has to move beyond a strictly linear resourcing model. “I think we’re past the model of one more study means one more clinical scientist,” he said. Technology, in his view, should help organizations match the right people, tools, and operating model to the work instead of simply adding headcount.

Maybe “Less” Really Should Mean Less

Cassady’s comment that stuck with me most may have been the simplest: “Why collect data if you don’t plan on using it?” That sounds obvious, but clinical research has struggled with exactly that problem for years. Recent Tufts CSDD research found that Phase 3 protocols now collect an average of about 5.9 million datapoints, with total data volume rising roughly 11% annually since 2020. Nearly one-third of procedures and datapoints were classified as non-core or non-essential.

I wrote about the same problem after DPHARM 2025, when Tufts’ Ken Getz showed that data points collected in trials had increased 283% between 2012 and 2020. We have known for years that protocol complexity and unnecessary data collection create burden for patients and sites. Knowing what should come out of a protocol, though, is much easier than getting a room full of stakeholders to agree on it.

After talking with Cassady, I came away thinking that “doing more with less” may be the wrong way to describe what he’s advocating. In some cases, it’s really about figuring out what we shouldn’t be doing at all: unnecessary reviews, duplicative processes, rote work, and data collection that adds burden without answering an important scientific question. Yeah, AI might help with that. But first, somebody has to be willing to question the process.

Editor’s Note: Cassady knows a thing or two about some of the burdens patients experience because he has been a clinical trial participant. My colleague Abby Proch explored his perspective in this Clinical Leader article.