AI has moved rapidly up the agenda for Life Sciences leaders. The potential is significant: faster decisions, greater automation, better use of scientific and operational knowledge, and new ways to improve productivity and performance.
But you can’t layer AI on top of poor data foundations and expect results.
We’ve compiled a short (and absolutely not exhaustive) list of useful terms that every life sciences leader should know when preparing the right data foundations and investigating the role of AI in their business.
1. Machine Learning
A type of AI that, rather than following rules written by people, learns patterns from historical data – including numerical data, text and images – and uses them to classify information, detect anomalies or predict outcomes. It can identify patterns across large and complex datasets or images that would be difficult for people to detect consistently – from recognising anomalies in manufacturing data to classifying images, detecting defects or identifying patterns in microscopy and other scientific imaging.
2. Generative AI
AI that learns patterns from existing data and uses them to create new content, such as text, images, code, audio or summaries, in response to instructions or prompts. Large Language Models (LLMs) are a major type of Generative AI used for language, although the terms are not interchangeable.
Because Generative AI is dependent on the quality of the data supplied, it can be wrong – even though it may read as confident and convincing. This means in highly regulated environments it must be reviewed by someone with the right expertise before it is relied upon.
3. Agentic AI
AI that can carry out a sequence of tasks towards a defined goal, using other systems and tools, and adapt its next steps based on what happens, with appropriate human oversight and controls. Also referred to as AI agents or autonomous agents.
While Generative AI can produce something for you, Agentic AI can do something for you. It enables businesses to move from AI that provides information or creates content to AI that gets work done.
4. Ontologies
A structured way of defining what data means and how different concepts relate to each other, creating a shared understanding that both people and machines can interpret. For example, an ontology can define that a batch is made from specific materials, on a piece of equipment, at a specific site, and may have deviations raised against it.
Without an ontology, systems and AI can struggle to understand its context and relationships and that limits the ability to connect knowledge across different systems, functions and datasets.
5. Semantic Layer
Where an ontology defines what things mean and how they relate, the semantic layer then applies those definitions to an organisation’s actual data – so people and tools can use them every day.
It enables businesses to create a shared understanding of their data across different systems and functions – so a term such as “batch”, “customer”, “revenue” or “deviation” means the same thing wherever it is used. This makes data easier to find, combine and interrogate, and gives analytics and AI more reliable business context. In simple terms a semantic layer helps turn data that systems store into intelligence the business, and AI, can understand consistently.
6. Data Standards
Agreed rules for how data is defined, formatted, structured and exchanged so that different people, systems and organisations interpret it consistently. Without defining standards data from different systems, sites or partners can mean or look different, making it difficult to combine, compare, exchange and use reliably at scale and limiting the adoption of AI.
Where to start: Identify the critical data that needs to move between systems, functions or organisations, check whether recognised industry standards already exist (for example CDISC for clinical data, Allotrope for laboratory data, and ISA-88 and ISA-95 for manufacturing) and agree common definitions, formats, units and structures where they don’t.
7. Structured Data
Data organised in a consistent, predefined format so that machines can reliably identify, interpret and use it. For example, results held in a laboratory information system, fields in an electronic batch record or records in a clinical trial database. By contrast, PDFs, emails, SOPs, reports and images are unstructured – the information is there, but a machine cannot reliably pick it out Without it, valuable information can remain trapped in documents, spreadsheets and inconsistent formats, requiring people to find, interpret and restructure it before it can be used at scale.
Where to start: Identify the critical data your business repeatedly extracts, re-enters or reconciles manually, and establish common fields, formats and standards for capturing it consistently at source.
8. Data Security
Ensuring data is protected and only available to the people and systems authorised to use it. Without it, you can’t safely increase access to data or scale its use across connected systems and AI applications.
Start by identifying who can access the data? What are they permitted to do with it? Where has it come from? Where is it going?
9. Data Pipeline
The route data takes from its source to where it is ultimately consumed or analysed. Without a well defined pipeline, moving and preparing data remains dependent on repetitive manual work, slowing down analysis, automation and AI.
As data moves through it, duplicate records can be removed, values checked, formats standardised and additional information calculated. Done well, a data pipeline replaces repetitive human intervention with a consistent, controlled flow – moving organisations from people repeatedly preparing data to data being systematically prepared for reuse.
10. Data & AI Governance
The rules, responsibilities and controls governing how an organisation creates, defines, manages, accesses and uses data. Without effective governance, complexity accumulates, standards diverge and all those individual workarounds build up and become future enterprise-wide data problems. As AI is adopted, the same discipline needs to extend to AI itself: which uses are approved, how outputs are checked, and where an expert stays in the loop – an appropriately qualified person who reviews AI-supported outputs and remains accountable for the
decisions made.
Start with your business outcomes and identify the data that matters most to achieving them. Identify the critical data, what “good” looks like and understand how it moves.
The ability to exploit AI tomorrow will depend heavily on the decisions organisations are making about their data infrastructure today.
Get those foundations right and an organisation isn’t simply preparing for AI, it is creating better data practices now: reducing manual intervention, improving trust, increasing accessibility and making information easier to reuse across the business.
We’ve seen how the right data foundations alone can unlock the data you already collect into new value for the business.
