Mayo Clinic Laboratories is applying AI where it helps most today, strengthening daily workflows while keeping experts firmly in control. From faster specimen intake and triage to data‑informed scheduling, first‑pass support for complex analyses, draft interpretive reporting, and reliable digital pathology quantification, teams use AI to organize information, reduce repetitive steps, and improve consistency. Human‑in‑the‑loop oversight, fit‑for‑workflow design, and regulatory‑aware communication keep the focus on safe, practical gains that benefit patients, clinicians, and laboratorians.
Across Mayo Clinic Laboratories, teams are applying AI where it directly supports daily work, specimen intake, planning and forecasting, first‑pass analytical review, interpretive drafting, and digital pathology quantification. The emphasis is practical: tools help organize information, reduce repetitive steps, and standardize routine processes, while experts remain in control of verification and final decisions.
Leaders and project teams consistently describe these efforts with the same guardrails, human‑in‑the‑loop oversight, clearly defined thresholds and flags, and fit‑for‑workflow design so the right people see the right information at the right time.
“AI in the labs is increasingly about transforming our operations, pre‑analytics, planning, and daily workflows that support the testing, rather than changing the test itself,” says Christopher Garcia, M.D., chief digital information officer of Mayo Clinic Laboratories. “When we scope tools to solve practical problems, they become part of how the lab works rather than a novelty.”
“AI in the labs is increasingly about transforming our operations, pre‑analytics, planning, and daily workflows that support the testing, rather than changing the test itself. When we scope tools to solve practical problems, they become part of how the lab works rather than a novelty.”
Christopher Garcia, M.D.
Moving samples faster with AI‑assisted intake and triage
At specimen intake, AI supports teams by capturing label details from slides and paperwork, organizing them into structured fields, and proposing routing suggestions that technologists verify before moving a case forward. In pilots that paired optical character recognition with language models, a single image could prepopulate information that previously required manual typing, helping reduce bottlenecks at the front door of the lab. Staff remain in control of verification and corrections, ensuring the process fits laboratory quality standards while streamlining the receiving, routing, and bench placement sequence.
Teams emphasized that real value comes from integrating this support into day‑to‑day workflows with clear operational and IT ownership, as well as realistic expectations about performance. The aim is fitness for purpose, not perfection, with humans explicitly in the loop to confirm or redirect as needed.
“We showed that OCR tied to large language models can pull key fields from incoming paperwork and labels, and with a single image, populate what used to be hand‑typed,” Dr. Garcia adds. “The technology is there, and the work now is fitting it into real workflows with clear ownership and expectations.”
“It is really just augmenting them, not replacing them,” says Patrick Day, MT(ASCP). “There is still a human involved in the process, but reducing the manual lookups and data entry helps work move forward faster.”
Data‑informed scheduling and capacity planning
Lab leaders are introducing decision support that brings forecasting discipline and feedback loops to daily planning. Rather than relying solely on rule‑of‑thumb scheduling, models incorporate relevant signals, such as operating room scheduling and planned capacity changes, to help anticipate needs. The intent is to provide transparent, evidence‑guided recommendations that supervisors can accept, adapt, or override, preserving judgment while improving standardization and knowledge transfer across teams.
This shift is designed to scale expertise, not supplant it. Decision support remembers when a veteran scheduler is unavailable, and it learns from outcomes over time. The result is planning that travels better across settings, improving the way instruments and staffing are aligned to daily volume without changing who makes decisions.
“The most useful thing is the feedback loop that lets tools learn from what actually happens and improve the forecast,” Dr. Garcia says. “It also means we can incorporate external signals like OR schedules and capacity changes that humans might not consistently track.”
First‑pass analytical support on complex signals
In areas like flow cytometry, spectral interpretation, and cell classification, AI acts as a first pass that organizes signals, proposes starting points, and flags cases that merit extra scrutiny. Confidence thresholds and flags determine what proceeds to expert review and what pauses for manual triage, keeping control with technologists and laboratory directors. The approach reduces repetitive compensation adjustments and elevates attention to rare or ambiguous patterns, adding consistency across shifts while preserving expert judgment.
Flow compensation illustrates the impact. What once required adjusting a large number of plots across interconnected fluorophore channels can be reduced to a quick verification step, easing a niche task that is time‑intensive to train and maintain. In spectral analysis, routine stone types can be handled more efficiently so staff can focus on subtle patterns that require seasoned interpretation. In both cases, the purpose is to reclaim expert time for cases where it matters most, while introducing a steadier baseline for routine work.
“Automating compensation turns a tedious manual step into quick verification so staff can handle many more cases,” says Wenchao Han, Ph.D. “They remain in the loop, and they can focus their time on the difficult problems.”
Draft interpretive reporting for pathologists
When interpretations must synthesize multiple assays and clinical details into a single narrative, AI helps by assembling a draft that pathologists review, edit, and sign. The emphasis is on context assembly rather than decision‑making, reducing the blank‑page burden and concentrating expert effort on nuance, accuracy, and clinical meaning. Teams describe a disciplined balance, where authorship and accountability remain with the pathologist, and where transparency about inputs and uncertainty is essential for trust.
This draft assistance is most useful when it is easy to shape into the final professional judgment and when it clearly signals what it considered. The result is steadier consistency across cases and a practical reduction in cognitive load, achieved without shifting responsibility away from the experts who sign out results.
“Drafting tools bring dozens of data points together so the pathologist does not start from scratch,” Dr. Garcia says. “They keep authorship and accountability exactly where they belong: with the expert who signs the report.”
Predictive readiness for blood products
Forecasting tools combine surgery type and specialty — and, when available, laboratory indicators such as hemoglobin or hematocrit — to estimate future demand for blood products. The design centers on the colleagues who make inventory decisions, providing a clearer preview of likely needs and reducing last‑minute stress. Human overrides are expected and encouraged, ensuring the tool augments rather than replaces the judgment that keeps service levels high.
This readiness support also has operational benefits. Better projections can help balance reliance on external suppliers with local donor contributions, aligning stewardship and day‑to‑day planning with clinical realities. The operational focus is deliberate, reflecting how forecasting can support the systems that, in turn, support clinical care.
“Surgery type and specialty are strong signals, and sometimes lab values like hemoglobin or hematocrit help refine the estimate,” says Thomas Tavolara, Ph.D. “If the forecast looks off, the staff steps in and overrides it, which is exactly how it should work.”
“Surgery type and specialty are strong signals, and sometimes lab values like hemoglobin or hematocrit help refine the estimate. If the forecast looks off, the staff steps in and overrides it, which is exactly how it should work.”
Thomas Tavolara, Ph.D.
AI in the clinical lab
Explore how Mayo Clinic Laboratories approaches AI in the clinical lab, focusing on workflow modernization, quality, and collaboration. Hear perspectives from leadership on using AI to streamline tasks, integrate diverse data, and support staff while maintaining human oversight and ethical guardrails.
Digital pathology, quantification, and sign‑out support
In digital pathology, AI assists with tasks like immunohistochemistry quantification, for example, counting positive and negative cells for markers such as Ki‑67, across large image fields. The goal is reproducible counting that adapts to new markers, with numbers presented reliably so pathologists can integrate them with morphology and clinical context. This work aims to reduce tedious manual tallying while preserving expert judgment about significance and next steps, and discussions about specific platforms remain out of scope to keep the focus on clinical workflow and consistency.
“Counting positive and negative cells for markers like Ki‑67 is crucial and time‑consuming for teams," Dr. Han says. “The aim is reliable counts that adapt to new markers, while experts decide what those numbers mean clinically.”
Together, these use cases show AI operating where it’s most practical in the lab: organizing information, easing repetitive steps, and supporting consistent decisions while keeping experts firmly in control. The emphasis is on strengthening daily workflows so patients, clinicians, and laboratorians all benefit from clearer, faster, and more reliable processes. With human oversight, fit‑for‑workflow design, and regulatory‑aware communication as the guardrails, this work reflects a measured approach to modernization.
