Ask Adrián Figueroa how much of an AI system’s work a person should check, and his answer is far less than most companies assume.
“With AI as it stands now, the volume of output is more than a human can process,” he said. “Reviewing everything turns the human into the bottleneck, but human judgment is still needed for high-risk or low-confidence decisions.”
Figueroa, an electrical engineer and Stanford Graduate School of Business graduate, applies that thinking to one of healthcare’s least glamorous problems, the logistics of getting medical products where they need to be.
He founded Flexio, a tech-enabled logistics services company specializing in healthcare, and in 2024 launched Routina, a platform that uses software and AI to help medical product distributors and other healthcare organizations connect information across their systems and cut down on manual coordination.
The need is easy to see. A single medical supply moves through distributors, pharmacies, clinics and laboratories, each working from its own systems, spreadsheets, emails and calls, and people end up stitching the information together by hand.
When that breaks, care stops. A VA Office of Inspector General report published in September found that the main operating room at a veterans’ hospital in Augusta, Georgia, closed four times in five months, once for 27 days, because of incomplete surgical kits, expired supplies and unreliable restocking.
The hospital had already paid for a $3.2 million automated inventory system that was never fully installed.
Oversight with teeth
Routina’s answer is to have the AI assign every result a confidence score and a risk score. People review and approve only the outputs flagged as low confidence or high risk, while the system tracks how it performs on everything it treats as safe.
Figueroa was clear that this cannot become a rubber stamp. “Human oversight has to be real, not a person clicking ‘approve’ all day,” he said. “Specific people own those flagged decisions, they have the authority to override the system, and every override is logged and reviewed.”
That mirrors Article 14 of the EU AI Act, which requires high-risk AI systems to allow effective human oversight, including the ability to override them.
He added two more conditions. Teams need to know what the system is measuring and why, and revisit those metrics often. “If the input data is biased, AI will apply that bias faster and at scale,” he said. And the last word stays with people. “What AI shouldn’t do is make the final decision.”
From pilot to multinational test
The model is moving into a live supply chain in the Caribbean. Surgical Solutions International recently launched a new line of medical products in the Dominican Republic with Cardinal Health, and it will run them through Routina.
“We will be implementing Routina’s technology at Surgical Solutions to manage all of Cardinal Health’s products and those of future suppliers,” Figueroa said.
Routina also plans to pitch Cardinal Health directly “as a solution to manage all of its distributors in the different countries where it operates,” he said. If that lands, a system designed to know when to hand a decision to a person will be tested across one of the largest medical distribution networks in the world.
Featured image: Alex Shuper via Unsplash+
Disclosure: This article mentions a client of an Espacio portfolio company.
