.png)
In this episode of the Verisian Community Podcast, Richardus Vonk discusses the transformative role of AI in clinical trials. He explores how AI is reshaping trial planning and execution, improving early disease detection like cancer, and the challenges it still faces in drug development. Richardus also highlights the benefits of automation in freeing up time for deeper scientific exploration and elaborates on the tactical versus strategic use of AI and GenAI. The conversation touches on the rise of open-source tools, balancing data sharing with patient privacy, while offering a vision for the future of clinical trials over the next decade.
Guest: Richardus Vonk, former Vice President at Bayer, over 30 years of experience in clinical drug development and statistics, formerly led both statistical programming and medical writing functions.
Host: Tomás Sabat Stöfsel, CEO & Co-Founder, Verisian
Topics: Richardus's career path from statistics into medical writing leadership, tactical versus strategic applications of AI in clinical trials, the "not in my trial" cultural resistance to innovation, building psychologically safe environments in a risk-averse, zero-mistake industry, open source versus commercial software tooling, lessons from successful and failed innovation projects, coaching upcoming leaders, predictions for clinical trials over the next 5-10 years
The full episode can be found on YouTube, Apple Podcasts or Spotify.
Richardus's career has spanned statistics, programming, and data management throughout, but his interest in AI's practical application sharpened while leading Bayer Oncology's medical writing function alongside his statistical responsibilities. Watching medical writers struggle to produce clinical study report narratives, a highly standardized but time-consuming task built almost entirely from structured database fields, he began thinking seriously about where generative AI could genuinely help, a line of thinking that expanded as he considered the simultaneous explosion in data diversity (clinical data, real-world data, wearables) statisticians and programmers now have to work with.
Richardus's central framework separates AI's current, largely tactical applications, automating narrative writing, generating tables/listings/figures, reducing the time and effort spent on standardized clinical trial deliverables, from a more strategic future use he thinks the industry hasn't yet seriously engaged with: applying AI at the level of an entire clinical program rather than a single trial. That means using AI to help predict trial outcomes, identify the right population for a program, determine how to move from phase one to phase three faster with greater certainty, and ultimately increase impact for patients. He argues today's AI conversation in pharma is almost entirely tactical and opportunistic, and that a genuine strategic conversation, program-level rather than task-level, still needs to happen.
Asked about building a culture that embraces new or unproven technology, Richardus describes a consistent pattern: people are broadly enthusiastic about new methodology in principle, right up until it's proposed for their own specific trial, at which point the response shifts to "great idea, but not in my trial," since a failure there has detrimental effects reaching all the way to patients. He argues overcoming this requires engaging decision-makers, clinicians, management, and regulators together, and that as a leader his core job is honestly assessing real risk: is a potential failure repairable, and repairable quickly? If not, the right move is smaller, incremental steps rather than a single large leap, gradually increasing the size of the steps as trust and evidence build.
Richardus is skeptical of fully automating processes with "humans out of the loop," arguing scientific progress moves faster than any standardization effort can track, which means domain experts, clinicians, statisticians, and programmers who can evaluate an algorithm's output and risk, remain essential. His framing for the human-AI relationship draws directly on the human-computer chess parallel (referencing Deep Blue's 1997 defeat of Garry Kasparov and the subsequent rise of "centaur chess," where human-computer teams still outperform computers alone): AI should function as a partner or enhancement, a "second physician" or programming assistant that helps debug code or find faster algorithms, not a replacement. The payoff, in his view, is freeing statisticians and programmers from routine, standardized tasks (writing SAPs, checking data structures, generating routine TLFs) so they can focus on higher-value scientific and strategic questions.
Richardus resists taking a side in the open-source-versus-commercial-software debate, framing himself as a statistician who simply needs the right tool for a given question, sometimes commercial software (sample size calculation, simulation), sometimes open source. He pushes back on the idea that open source is inherently cheaper, noting that validating, reproducing, and hardening open-source code for a GCP-controlled environment carries its own real cost, "there's no free lunch," whether you pay upfront for commercial software or later for validation work. His practical view: a statistician's toolkit should include SAS, R, and Python side by side, chosen by task (Python for AI work, R for advanced simulation, SAS for its mature, well-documented procedures), and he sees growing interoperability between them, SAS incorporating R, R reading SAS datasets, as a more productive direction than picking a winner.
Asked for an example of an innovation project executed smoothly start to finish, Richardus says he's never seen one, defining innovation as roughly "a good idea times implementation," where a genuinely different technology or workflow reveals limitations only once you try to apply it in practice. What separates good outcomes from great ones, in his view, isn't avoiding failure but how quickly and honestly a team adjusts when something doesn't work; he notes some of his own good ideas have failed simply because the timing wasn't right, either the organization wasn't ready or the underlying technology wasn't yet mature enough to support the idea.
Richardus describes his leadership coaching style as centered on stepping outside his own comfort zone and vision to genuinely ask what a given person's own ideas and strengths are, then helping develop those strengths rather than trying to fix weaknesses. His bluntest formulation: coaching is largely "asking a lot of stupid questions" and being kind about it, since the answer to what will make someone successful almost always comes from that person, not the coach. He connects this directly back to psychological safety: coaching only works inside a trusting environment where someone can say "I hear what you're saying, but I don't agree, and here's why," and argues that same safety is what allows people to admit mistakes even in the highest-stakes, closest-to-submission stages of a trial, since concealment of an error is categorically more dangerous to patients than the error itself.
Closing on the industry's trajectory, Richardus reaffirms his belief in randomized, double-blind trials as the enduring gold standard, but expects AI and generative AI to meaningfully streamline how trials are planned and run, particularly by integrating more external data sources into early-stage, program-level decision-making. He's explicit that this requires genuinely redesigned processes and strategy, not just faster versions of current workflows, but expects the industry to get there within five to ten years, resulting in faster, better outcomes for patients.
"Any standardization effort will not be able to keep up with that pace, with that scientific pace." — Richardus Vonk
"People see the challenges that we have in clinical development, and generally people are very open to try new things, but then if it comes to an important trial, the idea is this a great idea, let's do this, but not in my trial." — Richardus Vonk
"There's no free lunch. So either you spend the money here by buying the software, or you spend the money later by using the validation." — Richardus Vonk
"Coaching is just asking a lot of stupid questions... and being nice about it. Not so much about giving advice." — Richardus Vonk
"You need to create an environment where, if something goes wrong, people can say, 'Wait, this is a mistake. We need to redo this.'... otherwise decisions are being made on faulty interpretation of data. That's disastrous." — Richardus Vonk
What's the difference between tactical and strategic AI use in clinical trials, according to Richardus Vonk?
Tactical use automates standardized, time-consuming tasks within a single trial, like writing clinical narratives or generating tables and listings. Strategic use applies AI at the level of an entire clinical program, predicting trial outcomes, optimizing patient population selection, and speeding progression from phase one to phase three. He argues the industry is currently focused almost entirely on the tactical side.
What does Richardus Vonk mean by "not in my trial"?
A pattern where people broadly welcome new methodology in principle but resist applying it to their own specific trial, since a failure there carries direct consequences for patients and the program. He sees overcoming this as requiring honest risk assessment, incremental steps, and buy-in from decision-makers and regulators alike.
Does Richardus Vonk favor open source or commercial software for clinical statistical programming?
Neither exclusively. He frames tooling choice as secondary to patient outcomes and argues a statistician needs the right tool for each task, SAS, R, and Python each have strengths, and open-source code still carries real validation and hardening costs even though it's free to download.
Why does Richardus Vonk consider psychological safety essential in a zero-mistake industry?
Because an unsafe environment leads people to conceal errors rather than report them, and concealed mistakes, especially near submission, risk decisions being made on faulty data, which he considers far more dangerous to patients than an acknowledged and corrected error.
How does Richardus Vonk expect clinical trials to change over the next 5-10 years?
He expects AI and generative AI to meaningfully streamline trial planning and execution, particularly through integrating more external data sources into early-stage decision-making, while affirming randomized, double-blind trials remain the enduring gold standard.
The Verisian Community Podcast brings together experts in clinical trials to exchange innovative ideas and best practices central to clinical reporting, submission and review. Aligned with Verisian's mission to accelerate the evaluation and market launch of new medical treatments, each episode features expert insights, with guests ranging from statistical programmers to medical writers, to discuss the challenges and opportunities of the latest software and technology.
You can listen to us on YouTube, Apple Podcasts or Spotify.
See Verisian in action. Share a few details and we’ll be in touch to schedule your demo.
Your request has been received. A member of the Verisian team will contact you shortly to arrange a demo.
