Artificial intelligence dominated the conversation at CDMO Live Europe 2026. But compared to just a few years ago, the AI discussion is no longer about whether AI belongs in pharmaceutical manufacturing. It’s now about AI’s practicality in pharmaceutical manufacturing.
Where is AI already creating measurable value? Which applications still require caution? And how should pharmaceutical leaders decide where AI belongs within their organizations?
During his presentation at CDMO Live Europe, Ben Locwin, PhD, MBA, Vice President at Reliant Life Sciences, shared a practical framework for evaluating AI across pharmaceutical development and manufacturing. Rather than make broad predictions on AI replacing knowledge workers, he identified AI’s best current technology applications. He also recognized where experienced human judgment remains indispensable.
“AI can be incredibly powerful, but only when organizations understand both its strengths and limitations – as well as how to deploy it within a proper framework of data governance.”
- Ben Locwin, Vice President
Reliant Life Sciences
One key message had widespread appeal, touching manufacturing operations, quality systems, CMC, pharmacovigilance, and technology transfer consistently:
Organizations seeing the greatest AI return aren’t deploying it everywhere. They’re selective. They identify the highest-value use cases first. All the while, maintaining human oversight where it matters most.
AI Has Entered a New Phase
Manufacturing organizations have spent decades using statistical process control (SPC), process analytical technology (PAT), and advanced analytics to improve quality, consistency, and operational performance.
AI represents the next evolution, not because it replaces these disciplines, but because it can recognize relationships traditional statistical models often miss.
Locwin described how today’s AI tools are moving beyond retrospective analysis.
Newer models are not just limited to simply identifying why a batch failed. Now they can also begin better recognizing upstream process conditions that may contribute to downstream quality outcomes, long before issues become visible.
For manufacturers, that’s a big shift.
When supported by high-quality data and experienced interpretation, AI can deepen process understanding, identify emerging risks earlier, and improve operational decision-making.
The key distinction, however, is that better predictions don’t automatically produce better decisions. Organizations still need scientific expertise to determine whether those predictions should influence manufacturing, quality, or business decisions.
Four AI Applications Delivering Real Value Today
One of the biggest misconceptions surrounding AI is that every application’s value is interchangeable.
It’s not.
Locwin advised evaluating AI one use case at a time rather than viewing AI as a universal solution. Several applications are already demonstrating significant operational value.
1. Pattern Recognition Across Manufacturing Data
Modern pharmaceutical manufacturing generates enormous amounts of information through:
- Process analytical technology (PAT)
- Manufacturing execution systems (MES)
- Laboratory testing
- Equipment sensors
- Batch records
- Environmental monitoring
AI excels at identifying relationships across these disconnected datasets that may be difficult for individual teams to recognize manually.
Rather than replacing scientists or engineers, AI provides additional visibility into potential process relationships. AI gives experts stronger information for investigation and decision-making.
2. Equipment Performance Optimization
Locwin identified equipment optimization as one of AI’s strongest current applications.
Metrics such as Overall Equipment Effectiveness (OEE) and Total Effective Equipment Performance (TEEP) become more valuable when AI evaluates thousands of operating variables simultaneously.
These insights can help manufacturers:
- Identify maintenance opportunities earlier (more precise preventive maintenance)
- Reduce unplanned downtime
- Improve equipment utilization
- Support more efficient manufacturing operations
3. Documentation and Knowledge Management
Shifting gears over to another conference, Pharmacovigilance USA 2026, similar themes were echoed.
Namely, that AI creates the most value when it reduces administrative burden rather than replaces professional expertise.
Organizations are already finding practical applications for AI in:
- Preparing documentation
- Summarizing deviation histories
- Retrieving technical information
- Organizing large datasets
- Accelerating literature searches
- Supporting report preparation
These activities consume valuable expert time, but don’t necessarily require a specialized skillset to perform well (but they do require expertise in interpretation, to provide valuable context).
By reducing manual effort, AI allows scientists, engineers, quality professionals, and safety experts to focus more attention on higher-value analysis and decision-making.
These observations were previously mentioned in The Human Decisions Behind Stronger Pharma Partnerships and Smarter AI,which examined why AI is most effective when experienced leaders can interpret its insights, apply sound judgment, and remain accountable for the decisions that follow.
4. Decision Support
Perhaps AI’s greatest opportunity lies in supporting, not replacing, decision-making.
Whether evaluating manufacturing trends, identifying quality indicators, or organizing complex datasets, AI can help leaders process information more efficiently.
Final decisions, however, still require scientific expertise, organizational context, and professional accountability.
Four Areas Where Leaders Should Proceed with Caution
Locwin cautioned against assuming every manufacturing challenge is ready for automation.
Several high-profile applications still require significant human oversight, like:
Quality Investigations
Quality Management Systems (QMS) contain enormous amounts of valuable information, including:
- Deviations
- CAPAs
- Change controls
- Investigations
- Audit findings
Although these datasets appear well-suited for AI analysis, current models often reorganize existing information rather than generating genuinely new insight.
AI can neither determine if an investigation reached the correct conclusion nor whether organizational judgment was applied appropriately.
Governance remains a human responsibility.
Manufacturing Forecasting
History repeats itself, but not always in an easily predictable manner. AI, like humans, can struggle to make sense of uncertainty. Demand forecasting, technology transfer planning, and capacity planning remain difficult because AI can only learn from the data and patterns available to it. Many of the variables that shape these decisions fall outside manufacturing data and may have little or no historical precedent.
Market demand can change unexpectedly. Clinical programs may fail or change direction. Funding priorities can shift quickly. Competitive conditions evolve, and customer behavior changes as new needs emerge in a highly digitized, globally connected world.
Because these outcomes are driven by business decisions, external events, and human behavior, AI may identify patterns in existing data but still struggle to predict major changes that do not resemble the past.
Regulatory Documentation
Generative AI can produce documentation remarkably quickly. Speed, however, should never be confused with accuracy.
Large Language Models (LLMs) may fabricate supporting references, misinterpret scientific findings, or express uncertain conclusions with unwarranted confidence.
Regulatory documentation demands a comprehensive expert review before submission.
Clinical Judgment
Perhaps the clearest limitation remains clinical decision-making.
No AI model can yet fully replace the medical expertise required to evaluate patient safety, interpret clinical context, or determine appropriate regulatory actions.
These responsibilities also remain firmly within human expertise.
The Biggest Risk Isn’t AI. It’s Bad Data.
Perhaps the most important caution involved is data quality.
Modern AI systems are exceptionally good at generating answers, even when the underlying information is incomplete.
As Locwin emphasized, AI rarely tells users when available data is insufficient. Instead, it often produces confident-looking outputs regardless of data quality.
That creates risk.
If organizations rely on incomplete, inconsistent, or inaccurate information, AI accelerates flawed decision-making.
Successful AI programs depend on disciplined data governance and sophisticated algorithms to thrive.
Choosing the Right Level of Human Oversight
Many of these same themes resurfaced at Pharmacovigilance USA 2026. Here, drug safety leaders emphasized agentic AI should augment pharmacovigilance professionals, not replace them.
One discussion centered on selecting the appropriate governance model.
Some activities are well-suited for greater automation:
- Document routing
- Information retrieval
- Report preparation
- Workflow management
Others demand direct human involvement:
- Medical adjudication
- Safety signal evaluation
- Regulatory decisions
- Clinical interpretation
- Benefit-risk assessment
Rather than asking whether AI should always operate with a Human-in-the-Loop (HItL) or Human-on-the-Loop (HOtL) model, organizations should evaluate the potential consequences of error and assign oversight accordingly.
The goal isn’t maximum automation. It’s appropriate automation.
Smarter AI Starts with Smarter Decisions
Organizations should stop asking: Can AI do this? And instead ask: Should AI do this?
The pharmaceutical companies creating the greatest value from AI are making more disciplined decisions about where AI belongs, where human expertise remains essential, and how to balance innovation with accountability.
For life sciences organizations, AI is becoming an extraordinarily powerful decision-support capability.
Its greatest value, however, will continue to depend on the people responsible for applying it wisely.