When donating the blood that gets separated into plasma used for life-saving therapies, donors probably aren’t thinking that such an intensely physical procedure is the start of a highly regulated series of processes that are ripe for automation. But Aaron White does.
As Senior Director of I&T Plasma Digital Products & Services at CSL Behring, White has spent the past two decades navigating one of the most process-heavy environments imaginable. Once it leaves a donor’s arm, every step of the journey plasma takes demands accuracy, traceability and consistency. And automation has become a vital way to make that happen.
White has watched CSL Behring, a pharmaceutical company that manufactures medications from plasma proteins, shift from paper-bound workflows to RPA, and now toward fast-advancing agentic systems. And the difference, he explains in an interview with Automation Today, is not subtle. When he started exploring automation for plasma operations more than two years ago, labor shortages, competitive donor markets, and cost-sensitive manufacturing pipelines were squeezing margins across the industry. According to White, the question was never whether to automate—it was how fast automation could relieve the bottlenecks.
Ripe for Agentic Automation
White notes that while the major pharma firms in the plasma space follow nearly identical processes, differentiation comes down to minutes saved and errors avoided.
He emphasizes that early RPA work targeted repetitive internal tasks like checklists, data lookups, and back-office processes. But those improvements were only the beginning. The move from automated tasks to agentic decision support marked a much larger transformation.
“Determining the suitability of donors is where we wanted to take it next,” White explains. “Now I can just ask one question to my tool and the tool’s going to spit back everything the staff needs to know.”
The tool he’s referring to is Autopilot, implemented through UiPath. It is not automating decisions made by trained medical staff. But it is eliminating the need for those professionals to hunt for required documents, which previously defined the screening process. Instead of relying on staff to manually search through a 120-page rulebook of medical conditions and procedures, the conversational agent instantly consolidates the relevant guidance consistently, and in seconds.
White claims this shift has already changed donor outcomes. Not because humans were careless, but because humans were drowning in documentation. “God help you if you spelled ‘angioedema’ incorrectly,” he says. “If you did, you were going to miss what you needed.”
The impact, he notes, runs in both directions. Sometimes the agent flags issues earlier, reducing costly downstream complications at the manufacturing stage. In other cases, it prevents unnecessary donor deferrals by surfacing rules more accurately than manual lookups.
‘Don’t Waste Your Time on Low-Hanging Fruit’
CSL Behring’s first agentic use case wasn’t low-risk.
White insists the company intentionally started with one of the most complex processes in the workflow precisely because solving the difficult problems produces downstream leverage. It’s not advice you usually hear from automation experts implementing a new system.
“Pick the hardest one,” he says. “Don’t waste your time on low-hanging fruit.”
But complexity alone didn’t dictate pace. Regulation did. Everything in plasma collection touches donor safety, product quality, or patient outcomes, meaning the internal review process was intense. White recalls that after the agent itself was built in roughly seven to eight months, the organization spent an additional three to four months navigating internal change controls before moving into a formal pilot.
Convincing regulators is the next frontier. White says that while executive leadership is pushing aggressively for more AI, the challenge lies with stakeholders who must authorize its use inside safety-critical workflows.
“Now we’ve got to start working the regulators,” he explains. “Let us explain to you what human-in-the-loop means. Let us explain to you how we’re going to attack this.”
Even so, once the pilot proved accurate and consistent, the rollout moved fast. A three-site pilot with fewer than 20 users expanded to more than 2,000 users across 330 U.S. plasma donation centers in under a month.
White believes the rapid adoption ultimately came from seeing the tool in action. Early user skepticism centered around job loss fears, and AI uncertainty gave way to enthusiasm as staff realized the agent simply removed the most tedious portions of their day.
“We love this tool, it’s fast, it makes my life easier,” White reports users saying.
Speed and Consistency
White describes how the medical interview process once required staff to type symptoms, medications, or procedures into a PDF search field, hoping for an exact match. Misspellings, synonyms, and multi-factor scenarios all posed challenges.
“Now I ask one question to this tool and it gives me everything I need,” he says.
He highlights the speed—usually under four seconds—and the consistency. The agent returns the same interpretation every time, based on the same underlying documentation. Accuracy improves upstream donor qualification. Confidence improves downstream manufacturing. And the donor experience improves when staff aren’t juggling huge reference documents while conducting the interview.
The first agent may have been the most complex, but it won’t be the last. White’s team is now extending the model laterally. The internal learning and development group recently began piloting a version of the tool tailored to training content. Additional agents are in development, and orchestration will soon become essential. Multiple agents, operating behind the scenes, must appear unified to the frontline staff.
Advice for Automation Newbies
When asked what he would advise someone earlier in their automation journey, White doesn’t hesitate.
“I am a firm believer in ‘start hard’,” he says.
He argues that tackling the highest-value, most complicated process first yields broader organizational momentum and reusable design patterns. Low-complexity tasks, he believes, will increasingly be handled by baseline agents without needing major engineering resources.
His second point is cultural: involve business stakeholders early, especially in AI-driven projects. “Bring them along for the ride,” he says. He attributes some internal delays to not engaging quality and regulatory teams early enough to give them visibility and understanding of how the system worked.
And while the industry continues to debate AI budgets, consumption models, and governance, White appears focused on the only metric that matters in his world: getting the right information into the hands of the right people at the right time—fast, consistently, and safely.
