Sensational headlines highlighting multi-million-dollar AI breakthroughs, such as generative drug discovery and AI-assisted genomic variant interpretation, often make advanced computing feel unattainable for mid-sized laboratories lacking expansive budgets. During their 2026 Leadership Summit session on AI adoption in labs, James Smagala, bioinformatics practice manager, and Adam Steinert, chief technology officer, of Yahara Software, demystified AI for contemporary laboratories. Their core message emphasizes that while grand-scale drug discovery makes the headlines, operational AI is what delivers immediate, high-value return on investment (ROI) for mid-sized, clinical, and industrial laboratories.
James Smagala, bioinformatics practice manager at Yahara Software.
According to studies highlighted during the session, a significant share of organizations utilizing AI report moderate to high ROI. The secret to capturing this value lies in shifting focus from speculative scientific breakthroughs to the foundational data generated every single day.
How does operational AI differ from generative tools?
To successfully implement AI, lab managers must first understand where the technology actually fits into their workflows.
While generative AI (like Gemini, ChatGPT or Claude) and advanced bioinformatics represent one end of the spectrum, laboratories are finding far more immediate value in targeted machine learning and machine vision. These operational tools address practical bottlenecks:
- Workflow Automation: Bridging the gap between disparate systems (like instruments and LIMS) to execute seamless laboratory workflow automation.
- Predictive Equipment Maintenance: Analyzing service records and technician usage to forecast instrument failures through predictive maintenance strategies before disruptions occur.
- Quality Control (QC) Automation: Streamlining raw data validation to automate quality control routines and identify anomalous batches early in the analytical pipeline.
- Machine Vision & Object Detection: Automating the counting, measuring, and morphological profiling of cells or particles.
By automating these processes, laboratories do not replace human experts. Instead, those experts are elevated. Smagala noted that automation removes drudgery from the workflow, yielding better quality, higher sample throughput, and the opportunity for experts to focus on critical thinking.
Why does data quality block machine learning?
The most critical baseline of any AI initiative can be summed up in a favorite maxim of the Yahara team: AI science is data science, and data science is science. Many laboratories have been operational for 15, 20, or 25 years and believe they are sitting on a goldmine of training data. However, when they attempt to run an AI project, they encounter what Steinert and Smagala call the “unfortunate surprise of AI.” Operational data is rarely clean, normalized, or linked to validated outcomes through standard method validation practices. In one real-world project highlighted during the session, a laboratory possessed 220,000 sample records, yet only 800 had sufficient, consistent metadata to be usable for machine learning.
Adam Steinert, chief technology officer at Yahara Software.
Over a decade or more, SOPs change, instruments are upgraded, software is patched, and different technicians input data differently. This causes "temporal drift." If un-normalized data is fed into an AI model, it introduces massive bias, yielding highly consistent but fundamentally incorrect results.
For lab managers, this means an AI project is, first and foremost, a data cleanup project. In many cases, data cleansing, normalization, and structuring can take up to a year, while the actual AI implementation takes only a fraction of that time once the data foundation is solid.
How can image analysis run without custom models?
Steinert presented a case study that addressed a common bottleneck: manual microscopy analysis.
Collaborating with Dr. Pippa Cosper, MD, PhD, at the Department of Human Oncology at the University of Wisconsin–Madison, the team looked at the workflow involving FISH (fluorescence in situ hybridization) microscopy. The research required graduate students to manually look at cell images, identify boundaries, and count chromosomes to find duplications and deletions. This process was slow, highly repetitive, and subject to individual human bias.
To automate this, the team designed a hybrid pipeline:
- Cell Segmentation via Pre-Trained AI: Utilizing MicroSAM, a pre-trained cellular image model, to identify cell boundaries in dense clusters without custom training datasets.
- Traditional Logic for Counts: Implementing a classic, highly stable bright-spot detection algorithm inside identified boundaries instead of complex AI models.
This hybrid approach illustrates a vital lesson for lab managers: Do not use a complex AI model when simple math or classical computer science will do. By pairing pre-trained AI vision with classical algorithms, they built a highly consistent, reproducible pipeline that outputs structured JSON data, allowing Dr. Cosper’s lab to scale research without manual bottlenecks.
Lab Quality Management Certificate
The Lab Quality Management certificate is more than training—it’s a professional advantage.
Gain critical skills and IACET-approved CEUs that make a measurable difference.
What are the long-term validation requirements?
For lab managers accustomed to purchasing Commercial Off-The-Shelf (COTS) software like a LIMS, AI requires a major mental shift in project management and budgeting.
AI initiatives behave differently. They often start small with minimal upfront costs, but operational expenditure (OpEx) scales as the system is tuned, integrated, and scaled across the organization. Because AI models operate on real-world, constantly shifting data, they require continuous monitoring.
During the session, the team warned that these are not set-and-forget projects, as the system tends to evolve and iterate over time, meaning it must be treated as an evolving entity. Lab managers must plan for ongoing change management, continuous data quality validation, and periodic model recalibration to ensure long-term accuracy.
This ongoing oversight is also critical for regulatory alignment. Agencies such as the Food and Drug Administration (FDA) and the Centers for Medicare & Medicaid Services (CMS) have become increasingly adept at evaluating software platforms. For instance, the FDA provides specific guidance on artificial intelligence in medical devices, focusing on lifecycle management, good machine learning practices, and predetermined change control plans.
Interested in lab leadership?
Similarly, clinical laboratories must align software validation with established CLIA program standards administered by CMS to ensure testing accuracy and reliability. Validating an AI tool therefore follows the same rigorous path of documented verification as any traditional laboratory instrument.
How should a team structure an initial roadmap
If a leadership team pushes for AI adoption, or if a bottleneck is identified in the laboratory, the Yahara team recommended launching an initial initiative with the following steps:
- Assess Data Readiness: Catalog existing data assets, mapping out where they reside, whether they are structured or semi-structured, and how formatting has changed over the last five years.
- Prioritize Data Quality Over Problem Size: Target the cleanest, most organized dataset for the first AI pilot rather than the most complex bottleneck to ensure a faster, smoother proof-of-concept.
- Focus on Consistency Before Correctness: Prioritize pipeline reproducibility first, and then collaborate with scientists to fine-tune parameters for absolute accuracy.
- Involve IT and Data Science Early: Treat security, data privacy (especially avoiding PII exposure), and hardware infrastructure as day-one requirements by engaging technology partners immediately.
AI is no longer an exclusive tool for tech giants and massive pharmaceutical firms. By focusing on operational data, cleaning up existing pipelines, and leveraging pre-trained models, laboratories can eliminate manual drudgery, optimize instrument usage, and let scientists focus on what they do best: real science.
