Jana Hersch, Head of Corporate Scientific Research at Genedata.
getty
Historically, pharmaceutical organizations treated early discovery research and later regulated development as separate worlds. That distinction is becoming harder to maintain as AI models learn from discovery data and increasingly inform candidate selection, developability assessments, manufacturability predictions and other decisions that guide a molecule’s path through development. New regulations are also emerging around how AI is governed, documented and traced.
These shifts make high-quality discovery data matter far more than it once did. That increased value also brings a new risk: validation debt, a hidden liability that accumulates when scientific data and AI models lack the documentation and traceability needed for future use in good practice (GxP) environments, where requirements for data integrity, traceability, auditability and validation apply. The longer this debt accumulates, the harder and more costly it becomes to address.
Where Discovery Ends And GxP Begins
GxP standards come into play as candidates enter regulated safety testing, clinical development and manufacturing phases. Discovery research, by contrast, sits outside this framework. The boundary was once treated as clear: research generated knowledge, while development generated regulated evidence.
Discovery data can inform development decisions and, when properly contextualized and traceable, support downstream regulatory activities. Creating that flow requires data integrity and traceability in discovery, rather than retrofits at the GxP boundary, so research data remains interpretable and defensible as it moves downstream.
AI adds another dimension. Under the EU AI Act, systems classified as high-risk, defined by their intended use, must meet requirements for data governance, documentation, logging and life cycle oversight. The Commission’s draft Annex 22 also proposes expectations about certain AI models used in critical manufacturing applications, although its scope is still evolving.
Neither framework labels discovery as GxP. However, when an AI system falls within their regulated scope, its data, operation and lifecycle must be documented and controlled appropriately for its use. Where discovery data is used to train or test such a model, its provenance and integrity may become critical to the model’s defensibility in a regulated setting. Traceability cannot be reliably reconstructed after the underlying evidence has been created.
As AI models increasingly connect discovery data, scientific decisions and development outcomes, discovery data is increasingly subject to questions that most legacy electronic lab notebooks (ELNs) and other documentation systems were never designed to answer.
Research data has a lifespan beyond the experiment that produced it. It is reused during method validation, referenced in developability programs and sometimes included in regulatory submissions years after it was first created. It also serves as a critical part of the evidence package used to move promising molecules from discovery into development. For this transition to succeed, every assay, method and result generated during discovery must be transferred with sufficient context, traceability and documentation for the receiving team to trust, defend and build upon the data with confidence.
The Demands Of Data Integrity
Regulators commonly apply the ALCOA++ framework. It questions data in the following ways:
1. Who created it, when was it created and can we prove it?
2. Was it recorded when the event happened, not reconstructed later?
3. Is it the original, and if it is a copy, is the original preserved and where?
4. Is it accurate, and can we defend the method that produced it?
5. Is it complete, consistent, enduring and available on demand?
6. Is it traceable?
These principles ultimately come down to whether data can be understood, verified and trusted throughout its lifecycle. Traceability is central to that confidence because it supports all other data integrity attributes. If a result cannot be traced back to the instrument, method and analyst who produced it, then regulators cannot verify that result.
Data Integrity Gaps In Legacy Systems
In early-stage teams, individuals often use whatever tools are most readily convenient to achieve scientific results. This includes using spreadsheets for assay results, single-laboratory ELNs, custom Python scripts and ad hoc cloud tools, along with instrument software that exports data to local drives. While each method can yield useful outcomes, few can answer all ALCOA++ questions.
Spreadsheets are easy to read, but they often lack proper attribution and contemporaneity. While local instrument exports may consist of original data, they can lack traceability throughout the research process. Custom analytical code may be accurate when initially created, but it can become unverifiable over time. AI assistants and autonomous agents can accelerate research, but without appropriate governance, they may introduce gaps in traceability, provenance, reproducibility and oversight.
While none of these issues present problems in discovery, they can become compliance violations as soon as this data falls under inspection.
The Strategic Reframe
The best time to address validation debt is before data enters regulated use. Establishing integrity, context and traceability early can reduce duplicated work, validation delays and inspection risks later in development.
Data and validation gaps can also become material during due diligence, where the provenance, reproducibility and future usability of scientific data and models may affect technology-risk and remediation assessments. Addressing these gaps after a transaction may introduce unplanned costs, efforts and delays. The consequences of delay can be severe: McKinsey’s analysis suggests that a one-month delay in submitting a $1 billion asset could erase approximately $60 million in net present value.
Discovery is not becoming GxP. But as discovery data, analytical methods and AI models increasingly inform regulated activities, decisions made early in R&D can have consequences much later. Building integrity, context and traceability into scientific data from the start reduces validation debt and creates a stronger foundation for AI, development and regulatory success.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?











