CLN Daily
As artificial intelligence (AI) tools continue to evolve, there is growing interest in incorporating them into clinical workflows. But successful implementation does not end when an AI tool goes live. In Monday afternoon’s scientific session at ADLM 2026, “Life after AI deployment: A conversation on effective real-world performance monitoring,” Nick Spies, MD, from the University of Utah and ARUP Laboratories, and Christopher Garcia, MD, from Mayo Clinic, will address key considerations for using AI safely and effectively in clinical practice.
“As AI applications become more prevalent in our field, we have found that the immense efforts required to implement them into clinical workflows often stop at the ‘go-live’ date,” said Spies. “Ongoing monitoring is an often overlooked, but crucial, piece of using these systems safely and effectively.”
Spies will discuss one of the biggest gaps in the current landscape: the lack of established evidence and best practices for monitoring AI in clinical labs. The newness of this technology in lab medicine means monitoring frameworks are not yet built in; thus, careful planning is required before AI tools are implemented. Waiting until a system has already failed, disrupted clinical workflows, or affected patient care is a reactive approach that lab professionals simply cannot afford to take.
To help ADLM attendees think this process through, the session will explore what AI monitoring involves, why it matters, and how performance can be assessed from technical and statistical perspectives. The speakers will cover practical challenges in algorithm management, including bias surveillance, version control, governance strategies, and data drift.
The presenters will also address questions of accountability. For example, who is responsible for making critical “stop-the-line” decisions when performance drift occurs? What protocols should be in place to detect, evaluate, and correct drift before it compromises lab operations or patient care?
As AI becomes more common in clinical settings, its deployment should be viewed as the beginning of an oversight process rather than the final milestone. For laboratorians who have not yet implemented AI monitoring tools but plan to do so in the future, Spies offers straightforward advice: “Engage stakeholders early and often. The front-line staff and leadership know their workflows best.”
By involving staff, leadership, technical experts, and clinical stakeholders from the beginning, clinical lab professionals can design practical and sustainable monitoring routines that fit into their daily operations. This collaborative approach can help protect workflow integrity, support responsible AI use, and maintain the quality of patient care.
Spies and Garcia hope attendees emerge from their session with a clearer understanding of what is required after an AI model is deployed. First, the work does not stop at go-live. Second, effective monitoring requires ongoing collaboration, organizational buy-in, and clearly defined processes to ensure AI tools continue to perform as intended.