Digital technologies are evolving rapidly and offer benefits for bio/pharmaceutical manufacturers and developers who seek digital transformation. New artificial intelligence (AI) and machine learning (ML) tools help manage, transform, and analyze data—including data coming from process analytical technology (PAT) tools—so it can be better used to develop process insights or for process control. Early adopters of these technologies have been process development laboratories, but implementation in commercial manufacturing is emerging.
“Historically, PAT has been used to define processes and build control strategies, but its integration into manufacturing has often been limited,” says Michalle Adkins, director of life sciences strategy at Emerson. She points to the emergence of inline spectral PAT, which makes it easier to embed spectral waveform data directly into control systems, as a significant advance.
“We’re also starting to see OEMs [original equipment manufacturers] including PAT in their offerings with their bioreactors, which signals a broader readiness to integrate advanced analytics not only in process development but also in commercial manufacturing,” Adkins says. “These initial steps are building the foundations to lead to these technologies rolling out to bigger scale solutions in the future.”
Currently, AI-based tools are often used to create process insights that people on operations teams can use to improve processes, but eventually AI tools will allow more automated control, suggests Ryan Thompson, senior specialist, Industry 4.0 at CRB.
“Huge strides are being made in bioreactor productivity (e.g., higher titers) and chromatography efficiency using these tools,” Thompson says. “The next big step will be [to take] the person out of the middle and have truly AI-based closed-loop control.”
The convergence of AI and ML with advanced process control is helping to address some particularly difficult measurement and control areas, adds Adkins.
“Additionally, regulatory bodies are increasingly open to and encouraging the use of these more automated approaches, particularly as companies are demonstrating success in both process development and targeted projects,” Adkins says. “With early successes demonstrated in bulk drug substance processes such as fermentation and purification, along with pilot-scale manufacturing, broader adoption is expected across the manufacturing lifecycle.”
Challenges and solutions for commercial manufacturing
Digital Transformation for Deviation Management
A recent initiative by the BioPhorum IT Digital and Data collaboration tackled the need to need to improve collaboration and transparency between sponsors and contract manufacturing organizations (CMOs) in the area of deviation management.
“Historically, deviation management has been largely manual and fragmented, leading to inefficiencies, delays, and miscommunication,” says Ciera Clayton, Global Change Facilitator at BioPhorum. “Many companies still operate in the ‘predigital’ stages—where deviation information is exchanged manually, often via email, spreadsheets, or paper-based systems. These methods are prone to delays, errors, and lack of traceability. BioPhorum members recognized that digitizing this process could significantly enhance data integrity, accelerate issue resolution, and foster a more proactive quality culture across organizational boundaries.”
The BioPhorum IT Digital and Data, Digital Integration of Sponsor and Contract Organizations (DISCO) team published a paper in July 2025 to share its vision for digital deviation management (1). The group found that gaps between digital maturity of CMOs and sponsors caused difficulty in data exchange and decision-making.
“We have also learned that interoperability and shared digital standards are key to bridging these gaps,” says Clayton. “Even when one partner is more advanced, establishing common data formats, governance models, and secure exchange protocols can enable meaningful collaboration. Trust and transparenc are key; when partners align on expectations and invest in interoperability, they can overcome maturity mismatches and unlock significant value.”
Next steps from the DISCO team members aim to translate their vision into practical and scalable solutions, Clayton says. For example, one task is to develop standard data structures and integration specifications, with a focus on interoperability, good practice compliance, and scalability across multiple products and sites.
Work by other teams in the BioPhorum IT Digital and Data group will also inform the deviation management project. For example, the AI Validation team is publishing guidance to define frameworks for implementing AI in regulated environments, which will ensure that digital deviation data and other manufacturing insights can be leveraged responsibly, Clayton explains.
Reference
1. BioPhorum. Digital Deviations: Improving Sponsor-CMO Collaboration. BioPhorum, July 2025.
Moving from process development to commercial manufacturing requires technology to be robust and designed for use by operators on the factory floor.
“If new technologies add uncertainty or additional steps, they’re unlikely to be widely adopted,” notes Adkins. “Historically, that has been a challenge, but we’re starting to get over those complications as the technologies themselves mature. Today’s solutions are far more robust and designed for high-performance environments.”
Technology transfer is another concern. “Teams need to be sure that when they change the scale, it won’t change the outcomes,” Adkins says. “They need collaborative software solutions designed specifically to help navigate scale-up and scale-out.”
Data handling is also a crucial issue. “Process data on the factory floor is generally unstructured and lacks context,” says Thompson. “It takes a lot of time and effort to search, organize, and clean data before [they are] useful for AI/ML algorithms.”
Large quantities of data are needed, however, because AI/ML accuracy and insights improve with the quantity of data.
“Data modeling is a necessary first step to meaningful AI/ML, where process data and modern architectures, such as a unified namespace, should be considered,” Thompson suggests.
“One of the most critical—and often underestimated—elements of effective data analytics in pharma manufacturing is context,” agrees Adkins. “Teams must be empowered to generate and preserve context as data flows across the enterprise. Whenever context needs to be reconstructed, the value of that data diminishes. It becomes harder to use, more expensive to manage, and slower for driving meaningful action. That’s not just a technical challenge—it’s a strategic one.”
Emerson’s new digital platform will enable integration of disparate industrial automation technologies, according to Adkins. A centralized data model will allow teams to collect full context and share the master data across solutions without needing redundant definitions in separate systems (e.g., the distributed control system, the manufacturing execution system, and others), Adkins says.
“The enterprise operations platform will eliminate that fragmentation by creating a common data platform that distributes consistent equipment information across all solutions via a shared data fabric,” Adkins says. “This not only simplifies data management but also strengthens the foundation for more autonomous and scalable operations.”
Using the approach of managing data as a product, the platform is designed to make data findable, accessible, interoperable, and reusable (i.e., the FAIR principles).
“This approach enables scalable, AI-ready data environments that support both operational efficiency and innovation,” says Adkins.
Predictive control strategies
Digital twins are tools that have found success in process development and promise additional benefits in manufacturing control strategies. The integration of PAT data with a digital twin of a process offers multiple benefits, says Andrew Whytock, head of Digitalization for Pharma at Siemens, which supplies digital tools including digital twins and the Simatic Sipat software platform for implementing PAT in pharmaceutical development and manufacturing.
“Integrating PAT enables right-first-time production, as real-time data and predictive models minimize errors and ensure quality from the very beginning,” Whytock explains. “This integration also leads to improved yields, with continuous feedback and optimization enhancing output.”
Whytock says that a digital process twin can minimize physical experimentation and significantly reduce development time through virtual testing and optimization in the digital twin. Although these tools can allow significant efficiency gains and faster time-to-market, a potential hurdle is the initial investment cost for infrastructure, software, and integration of disparate processes into a unified, data-driven approach.
In addition, says Whytock, “Overcoming the reliance on extensive physical experimentation, a long-standing practice in biopharma development, demands a paradigm shift in how processes are designed, tested, and scaled up. This transition also necessitates specialized expertise in data science, modeling, and automation, which may not be readily available internally, requiring significant training or external consultancy.”
Digital twins can be used for batch processes and are fundamental for continuous processes, for which process automation is crucial, suggests Whytock. Continuous manufacturing saves space, reduces costs, and speeds up the path from lab to patient, he says, but the shift from batch to continuous still faces resistance due to perceived risks and question marks about return on investment.
“Ultimately, the challenge lies in navigating the complexities of digital transformation while ensuring compliance, maintaining product quality, and realizing the full efficiency gains offered by these cutting-edge solutions,” he explains.
Although AI tools and digital twins may be used in various ways to improve quality and potentially enable closed-loop control and real-time release, process understanding remains crucial. In a 2024 survey of pharmaceutical and biotechnology companies, CRB found that intelligent devices and smart data (e.g., PAT) are indeed already being used by more than half of respondents for real-time batch analysis (i.e., process understanding) and by 40% of respondents for real-time batch release (1). A smaller segment (24%) reported using in-line or online sensor technology for real-time batch correction.
“Process understanding is a key component of being able to leverage AI into batch analysis, release, and correction,” cautions Thompson. “Relying on AI without a detailed process understanding will lead to implementations that cannot be validated—and of course, the rigor required increases as companies move from analysis, to release, to correction.”
A July 2025 publication from the International Society for Pharmaceutical Engineering (ISPE) GAMP Community of Practice, Software Automation and Artificial Intelligence Special Interest Group offers best practices for approaching AI projects integrated with current regulatory guidance (GAMP was formerly known as the acronym for good automated manufacturing practice) (2). The guide covers a broad range of use cases, including use of AI in data analysis for applications such as real-time release.
“An innovative approach combining PAT and AI models alongside robust control strategies can enable faster and more reliable product disposition by analyzing critical quality attributes in real-time rather than relying solely on end-product testing,” says Brandi Stockton, founder and managing partner of the Triality Group and lead of the ISPE GAMP AI Guide initiative.
“AI has the potential to enhance PAT by capabilities to identify subtle patterns in large amounts of data and support predictive control strategies by dynamic optimization of parameters to reduce process variability,” adds Martin Heitmann, consultant at the Triality Group and co-lead of the ISPE GAMP AI Guide initiative. “In process development, AI may reveal relationships between variables that traditional methods may miss, accelerating scale-up and tech transfer. Several key challenges to consider include data availability and quality, explainability and integration with existing control systems.”
Embracing AI
The pharmaceutical industry has a reputation for being slow to change, but the fast pace of AI development may be altering that paradigm. Thompson says that AI has become a strategic necessity in the drug discovery and information technology functions of pharma companies. Although adoption of AI has been slower in the manufacturing side, Thompson predicts rapid adoption will occur as service providers start demonstrating successful implementations and manufacturers see the benefits.
“In the past three to five years, I have seen real changes [from] pharma being an insular industry to one seeking ideas from other sectors,” suggests Thompson. “AI is absolutely being adopted rapidly in pharma, and companies are creating new organizational units to invest in and [obtain] value from AI. It is slower to be adopted directly on the factory floor, but pharma companies are embracing AI more than most other manufacturing sectors, as they have much more to gain than typical companies.”
References
1. CRB. Horizons: Life Sciences Report; 5th edition; CRB, October 2024.
2. Stockton, B.; Heitmann, M.; Staib, E.; et al. GAMP Guide: Artificial Intelligence; ISPE, July 2025.
