By Martin Ball
Sr. Product Marketing Manager, Automation and Digital, Cytiva
Ashley Howard
Senior Product Director, Automation and Digital, Cytiva
Modern biopharma manufacturing faces many challenges, including process execution, observability, digital transformation, and reliability. While each topic addresses a different aspect of manufacturing performance, they all contribute towards a common objective: creating greater confidence in how decisions are made.
Over the last decade, biopharma manufacturers have invested heavily in automation, connectivity, data platforms, and digital technologies while navigating the challenges of integrating legacy infrastructure and modern manufacturing systems to produce increasingly diverse therapeutic modalities. As a result, organizations today have access to more process, equipment, and operational data than ever before. The challenge has moved from collecting information to using that information to anticipate outcomes before they occur.
Historically, manufacturers have focused on understanding what happened. Increasingly, the goal is to understand what is likely to happen next. In many ways, this represents the next stage in the digital evolution of the industry, where confidence in predictive systems becomes just as important as the technology itself.
Moving beyond hindsight toward model-driven bioprocess control
For decades, manufacturing improvement has been built upon hindsight. Data is collected, deviations are investigated, root causes are identified, and corrective actions are implemented. Valuable insights are generated, but often only after an event has already occurred. While this approach has driven significant progress, it has become increasingly challenging to apply the same process in the face of today’s demands on productivity, flexibility, and efficiency, especially for advanced drug modalities. The process is often labor-intensive, reliant on specialist expertise, and difficult to scale. Lessons learned in one investigation are not always easily transferred across sites, processes, or manufacturing networks.
Organizations are increasingly looking for ways to become more proactive. The ability to predict process behavior, anticipate operational risks, and evaluate potential outcomes before they affect performance creates opportunities to improve productivity, reduce variability, and make better-informed decisions.
Model-driven manufacturing supports this shift by combining process knowledge, historical data, equipment information, and operational performance into models that can help predict how a process is likely to behave under different conditions. In this context, a model is a digital representation of how a process is expected to behave, developed using scientific understanding and manufacturing data. Its effectiveness depends on the quality of the underlying data, the way it is trained and calibrated, and how effectively it is maintained throughout its lifecycle.
In practice, model-driven manufacturing can take many forms. Advanced process control (APC) systems can use predictive models and real-time process data to automatically adjust operating conditions. Mechanistic models can help manufacturers understand how biological and physical processes interact, while digital twins can create virtual representations of manufacturing processes that allow organizations to evaluate potential scenarios before implementing changes in the real world.
Models are only as good as the foundations beneath them: enabling advanced process control in bioprocessing
While model-driven manufacturing offers significant potential, the effectiveness of any model ultimately depends on the quality of the foundations supporting it, not just the quality of the data behind the model. For example, APC is made up of algorithms that continuously adjust process parameters based on real-time data, to maintain optimal performance. In bioprocessing, an APC might, for example, adjust feed rates or temperature in real time in response to sensor data about cell growth or metabolite accumulation, all without manual intervention.
However, models that feed APC algorithms are not created from data alone. They require process understanding, subject matter expertise, and an ability to distinguish meaningful signals from operational noise. As manufacturing environments evolve, models must evolve with them. Changes to equipment, processes, operating conditions, or therapeutic modalities may require models to be refined, recalibrated, and continuously improved to ensure they remain relevant and trustworthy. For larger organizations, this is increasingly becoming a strategic capability. However, developing and maintaining the expertise, infrastructure, and governance required to support model-driven decision-making may represent a greater challenge for smaller organizations.
Ultimately, successful model-driven manufacturing depends as much on process understanding and organizational capability as it does on technology.
The foundations have already been built: empowering mechanistic models and digital twins for biopharma manufacturing
Although model-driven manufacturing is often discussed as a future concept, many organizations have spent years building the capabilities required to support it. Investments in automation have improved data collection. Observability has improved process understanding. Digital transformation has increased connectivity across people, data, and systems. Reliability initiatives have improved the consistency and predictability of manufacturing operations.
Together, these capabilities create the foundation for more predictive ways of working. With integrated data systems and improved process understanding, companies can develop in silico models of their processes. For example, a digital twin of a bioreactor can allow engineers to simulate how different feed strategies might impact cell growth, titer, and product quality before testing those changes in the lab or at manufacturing scale. Mechanistic models for bioprocess optimization are the predictive foundation for these simulations.
Over time, the longer-term opportunity is to progress from predictive recommendations towards model predictive control, where routine process adjustments can increasingly be performed within defined operating parameters while people retain oversight and responsibility for managing exceptions.
The value does not derive from the model itself. It comes from combining process understanding, operational expertise, and predictive insight to enable tighter process control, reduced variability, improved yield, and more consistent manufacturing outcomes. In many cases, the greatest value lies not only in preventing failures, but in identifying opportunities to optimize performance within an established operating space and a set of agreed parameters or guardrails.
Building trust in predictive operations for closed-loop bioprocessing
Perhaps the biggest challenge facing model-driven manufacturing is not technology, but in determining how much trust organizations are willing to place in artificial intelligence (AI)-supported recommendations and predictive systems. Closed loop and/or continuous manufacturing necessarily require more trust in the data and models developed to remove humans from the equation. However, proving to regulators that these algorithms can be trusted and validated is still an ongoing challenge for many organizations.
Other industries have already demonstrated how this transition can occur. Modern aircraft operate with extraordinary levels of automation. Throughout much of the flight, automated systems navigate, monitor performance, and respond to changing conditions. Yet pilots remain firmly at the center of the operation.
The reason goes back to trust. Pilots trust automation to perform routine activities consistently and reliably, while they provide judgement, contextual understanding, and intervention when unusual circumstances occur. Trust was not established overnight. It was built through years of proven performance, transparency, and predictable outcomes.
Biopharma manufacturing is likely to follow a similar path. Predictive, model-driven systems will increasingly identify risks, evaluate potential outcomes, and recommend actions before performance is affected. Human expertise will remain essential, but it will increasingly be focused on oversight, optimization, innovation, and managing exceptions rather than routine operational decision-making.
The future role of human expertise in model-driven bioprocess control
One of the most common misconceptions surrounding AI, predictive analytics, and model-driven manufacturing is that they are intended to replace people. In fact, the reality is that roles will change, not be made redundant. Scientists, engineers, and manufacturing specialists possess knowledge, experience, and contextual understanding that no model can fully replicate. The objective is not to eliminate human involvement, but to reduce the need for routine intervention and increase the value of the expertise people provide.
As these technologies mature, demand is likely to increase for skills in data science, modelling, process analytics, systems integration, and digital manufacturing. The future workforce may spend less time collecting information and more time interpreting, optimizing, and improving how we use it.
This creates opportunities for people to focus their expertise where it delivers the greatest value, including process optimization, continuous improvement, innovation, technology transfer, manufacturing strategy, and accelerating the delivery of therapies to patients. In many ways, the ultimate goal of model-driven manufacturing is not automation itself, but enabling human expertise to be applied more effectively.
An emerging and untapped competitive advantage
As access to data, automation, and digital technologies continues to expand, most organizations will eventually collect similar information from similar systems. Manufacturers that can anticipate outcomes, evaluate alternatives, and act before problems occur will gain advantages in productivity, quality, reliability, and operational performance. As processes become more predictable, organizations can spend less time managing uncertainty and more time driving improvement.
In many ways, this may become the next manufacturing battleground. The organizations that outperform will not necessarily be those with the most sophisticated technology. They will be those that can translate predictive insights into trusted decisions and consistent execution.
From uncertainty to confidence in model-driven biomanufacturing
The future of biomanufacturing is not simply connected, automated, or digital. Key elements work together to build a foundation for the future. Closing the execution gap will enable action. Increasing bioprocessing observability will create understanding. Digital transformation will lead to connectivity. Reliability will increase confidence. Model-driven manufacturing will enable anticipation, or, proactive versus reactive processes. Together, they represent the industry's journey from reacting to events towards shaping outcomes.
We may never completely trust the models behind predictive manufacturing. However, processes may become predictable enough that human expertise will eventually focus less on managing uncertainty and more on creating improvement, innovation, and value.
CY61869-27SEP26-AR