Laboratories generate vast amounts of scientific data, yet much of it remains fragmented across instruments, software platforms, and research sites, limiting its long-term value. As artificial intelligence (AI) and machine learning (ML) become increasingly important research tools, creating connected, context-rich data environments has become a foundational requirement rather than a future aspiration.
David Hardy
Lead Market and Innovation Strategist
Thermo Fisher Scientific
In this Innovation Spotlight, David Hardy, lead market and innovation strategist, digital science and laboratory automation at Thermo Fisher Scientific, discusses how connected workflows, automation, and strong data governance can help scientists spend less time managing information and more time advancing scientific discovery. He also explains why human expertise remains central as laboratories adopt increasingly intelligent digital tools.
Why are siloed and fragmented data among the biggest barriers preventing laboratories from efficiently turning scientific insight into experimental action?
Siloed data, in many ways, is dark data. It may have been used once, but it then becomes difficult to find, access, or reuse, even as it continues to cost the organization in terms of storage and maintenance. Fragmented data compounds the problem even further. Before the data can be put to work, additional time and resources must be spent reconciling, contextualizing, and standardizing it.
The FAIR principles capture this precisely: data is only truly valuable when it is findable, accessible, interoperable, and reusable. That framing applies directly to AI and ML because any model depends on high-quality, well-contextualized data. When that data is invisible, incomplete, or unreadable, the barrier to reuse becomes too high. The result is that potential insights go unrealized, and opportunities for acceleration are lost before the science even begins.
At Thermo Fisher Scientific, we often see this challenge play out not only within individual laboratories but across large, complex research organizations where data is generated across instruments, sites, applications, and teams. If scientific context is lost as data moves, scientists spend valuable time reconstructing what happened instead of deciding what to do next.
Connecting and contextualizing data is therefore not just an informatics exercise. It is a way to give scientists more confidence, more usable knowledge, and more time for higher-value scientific thinking. When data can move with its context intact, laboratories are better positioned to turn insight into action.
What does a truly connected, closed-loop laboratory workflow look like in practice?
Like any sound scientific process, the connected workflow begins by asking, “What are we trying to answer, and why?” AI applications can help scientists develop and refine those questions, while digital tools can support initial in silico analysis, helping predict likely outcomes and shaping experimental design before a single sample is prepared.
In an automated digital lab, data-quality checks are critical. Once the initial foundation is set, ML and statistical techniques help identify low-quality results, while a human-in-the-loop approach provides oversight and review where needed. Those insights then support higher-level analysis such as trending, cross-species comparisons, and broader pattern detection.
The loop closes when data, results, and insights from one round of experiments are available to inform the next. Most laboratories are not starting from a blank slate, and connected workflows need to integrate existing technologies, preserve scientific context, and evolve as new methods and analytical tools emerge. Each experiment generates results and adds to a growing knowledge base that helps scientists design subsequent experiments with greater speed and confidence.
AI-powered digital tools can help scientists analyze connected laboratory data, refine experimental design, and make more informed research decisions while keeping scientific judgment at the center.
©iStock, FG Trade
What foundational capabilities need to be in place before AI can deliver meaningful value in the laboratory?
AI cannot generate meaningful value on a weak data foundation, so the first step often comes earlier than most organizations expect, with connecting data sources, capturing context at the point of work, and building trust in the data before AI enters the picture. From our experience supporting customers, the most successful programs sequence those steps deliberately, focusing first on the workflows and data assets that matter most scientifically, then expanding toward broader standardization, governance, and AI readiness.
In practice, that means ensuring rigorous data capture, including metadata that places results in their scientific context, reliable storage and access, and a clear approach to standardization across formats, naming conventions, and integration standards. For more complex environments, it may also require knowledge graphs or ontologies to represent the relationships between samples, methods, instruments, results, and outcomes. Data governance cannot be an afterthought, particularly around provenance, audit trails, and access control. Where AI and ML models are deployed, model performance needs to be tracked over time, because models may require retraining or revalidation as new data accumulates or scientific context shifts.
There is also a cultural dimension that is easy to underestimate. Building genuine confidence, skills, and trust in AI takes time, and leadership needs to recognize that the technical capabilities and the quality data foundation must advance together. When those foundations are in place, AI becomes part of a trusted scientific workflow, helping scientists find relevant prior work, evaluate options, and make more informed decisions without displacing scientific judgment.
How can automation support research teams without disrupting existing workflows?
Scientists already have established instruments, protocols, systems, and decision points. The right goal for automation is to remove the manual, repetitive, or error-prone steps from those workflows without changing the fundamental way the science is done. The most productive place to start is identifying where scientists lose time today: manual transcription, repeated searches for prior results, file conversion, sample handoffs, and routine reporting. Addressing those friction points first creates value without asking scientists to abandon familiar processes.
Achieving this requires connecting with the systems laboratories already use, including instruments, laboratory information management systems (LIMS), electronic lab notebooks (ELNs), data stores, and analysis platforms, so that data flows more seamlessly throughout the lab. Automation can handle data movement, metadata capture, quality checks, file conversion, sample tracking, and routine reporting without requiring scientists to relearn how they approach their work.
Automation can also help teams make better use of the data they already have. Previous results, protocols, and contextual data can be surfaced when relevant, helping scientists avoid duplication and make more informed decisions at the design stage. A human-in-the-loop approach remains essential throughout. Automation should suggest, flag, route, and orchestrate, while experimental design, interpretation, and final decisions remain with the scientist.
How is Thermo Fisher Scientific helping laboratories adopt more connected and automated digital workflows?
At Thermo Fisher Scientific, we approach connected laboratory automation as an ecosystem challenge. Most laboratories are not starting from a blank slate. They operate complex environments with existing instruments, automation platforms, LIMS, ELNs, data stores, security requirements, local processes, and scientific variation. Our role is to help customers connect those environments in a practical, scalable way so they can create value without forcing scientists to rebuild the way they work.
A key principle is openness. Our vendor-agnostic approach to integration is designed to connect with the scientific instruments, laboratory automation equipment, and digital systems that laboratories already have in place. From there, we help customers strengthen data capture, preserve metadata and context, automate routine steps, improve data movement, and build the governance needed for AI-ready workflows. Because we work with customers across a wide range of scientific disciplines, applications, and operating models, we see the practical challenges that arise when organizations try to scale digital transformation beyond a single workflow or site. That experience helps us support customers as they move from isolated automation projects toward connected scientific ecosystems that can grow with their needs.
When labs have connected and automated digital workflows, scientists spend less time moving, formatting and reconciling data, and more time asking better questions, designing better experiments, and interpreting results with confidence.
How do you see connected automation, AI, and real-time data orchestration changing the role of scientists and the future of laboratory research?
The most immediate change will be in how scientists spend their time. Working in an automated digital lab, with advanced tools like AI and ML, will free them from manual data handling, reformatting data, and troubleshooting disconnected workflows, allowing them to focus more on the science itself. Legacy data will become more accessible and useful, and scientists will be able to find, compare, and interpret it more efficiently.
As connected scientific ecosystems mature, scientists will increasingly act as orchestrators of data, automation, and AI-enabled workflows, deciding which questions to pursue, which data to trust, and how automated recommendations should be tested. Their role becomes more focused on scientific judgment, experimental design, and interpretation.
Increasingly, scientists may provide instructions to digital agents to help address specific scientific or automation-related problems. Those agents may return suggested approaches, but it is the scientist who uses their expertise to determine what is scientifically valid, practical, and appropriate.
