Modeling & Simulation: Why Adoption Stalls in Pharmaceutical Development and Manufacturing

Editor’s Take: M&S can deliver real value, but adoption still hinges on trusted data, organizational buy-in and confidence in the models.

In a race defined by time to market, modeling and simulation (M&S) has outgrown its former reputation as a nice-to-have. Across industries, and in conjunction with advanced manufacturing initiatives, the evidence that M&S can reduce experimental burden, improve quality, and de-risk decisions is clear and well understood. Regulators encourage its use, industry leaders talk about it openly, and a growing body of case studies documents tangible impact.

Adoption, however, remains stubbornly uneven. If the value is so well established, why hasn’t M&S become a routine part of day-to-day product and process development?

The shortfall is not one of tooling. Mature, commercially available modeling platforms exist across scales – from molecular and unit operation models to plant-wide simulations and digital twins. Nor is it a shortage of justification. Most organizations already have strategic initiatives focused on process improvement, supply chain resilience, sustainability, and risk reduction, which are precisely the areas where M&S delivers value.

The real barriers lie elsewhere, and this article examines four of the most significant:

The True Cost of In-House M&S Capability

When organizations set out to bring M&S in-house, the discussion often begins and ends with software licensing, yet licensing is only a fraction of a much larger investment.

Consider a modest team of five to six specialists in a mid-cost region: in our experience, the annual bill can easily approach $2 million, and for each full-time employee, the outlay goes well beyond salaries. A realistic in-house setup typically carries:

• Specialized talent that spans process engineering, chemistry, data science, and advanced modeling

• Onboarding, training, and knowledge transfer

• Multiple commercial-grade M&S software suites

• High-performance computing infrastructure, whether on-prem, cloud, or hybrid

• Ongoing IT support, maintenance, and upgrades

Importantly, much of this cost is fixed. M&S demand rarely holds steady, and much of the work is front-loaded, with scope and problem definition, model setup, and validation, followed by long compute runs and waiting cycles. Utilization rates of roughly 60% are typical, and that idle capacity is what inevitably puts these teams and programs at risk. Whether the project pipeline is high or low, the organization shoulders the same infrastructure and personnel costs.

Total Cost of Ownership: A More Informed Perspective

Companies often delay M&S adoption because the upfront and recurring costs feel disproportionate to near-term needs, even when the long-term value is clear. A total cost of ownership (TCO) model offers a more sophisticated way to weigh the decision. TCO generally resolves into three categories:

Software and Compute

• Licensing costs, which usually renew annually, vary widely, as do compute costs whether on-prem or in the cloud
• Scale and usage models are the primary dimensions to consider, alongside depreciation, energy usage, and data management, all of which climb as model complexity grows

Personnel

• Engineering talent is usually the single largest contributor, typically eclipsing the combined cost of hardware and software

External Expertise

• Even with strong in-house capability, organizations commonly bring in external specialists for domain-specific expertise or when extra capacity is needed

Framed this way, the build-versus-buy question changes shape. It’s not just about the upfront costs; it is a question of how much permanent capability the business truly needs.

On-Prem, Cloud, or Hybrid? Costs Still Add Up

Many organizations look to infrastructure choices in the hope of softening the investment, but each path has its own set of trade-offs.

On-prem systems offer control and can align with legacy IT policies, but they require capital expenditure, maintenance, upgrades, and internal expertise to manage performance and reliability. Cloud-based compute reduces upfront hardware costs and offers flexibility, but introduces recurring usage fees, data management considerations, and licensing constraints. For compute-heavy simulations, costs can escalate quickly. Hybrid models blend the two, and with them comes additional complexity, since integration, governance, and security still require internal ownership.

The M&S Data Readiness Gap

Over the past decade, the application of Artificial Intelligence (AI) and Machine Learning (ML) has demonstrated remarkable success in digital-native industries, where data is abundant, cheap, and generated continuously. The same playbook, however, doesn’t apply to pharmaceutical process development and manufacturing.

Pharmaceutical manufacturing processes are complex, highly regulated, and often developed under intense time pressure. Most historical datasets are sparse and fragmented, having been collected for quality assurance or regulatory compliance rather than predictive modeling applications. Data is scattered across laboratories, pilot plants, manufacturing sites, and contractors, and stored in incompatible systems with limited traceability. Even when data has been digitized, it tends to reflect a relatively narrow operating window, the “safe space” of development, rather than the broader design space needed to train robust AI or ML models.

This points to a need common to all industries looking to apply AI, ML, and data-driven models. The limiting factor is not algorithms, compute power, or even talent; it is data readiness. It’s why we see so many companies stuck in AI pilot mode.

M&S efforts stall for the same reason. Time and again, the foundational data required for purely data-driven approaches proves unavailable, untrustworthy, prohibitively expensive to generate, or fundamentally unsuitable for the decisions at hand.

Data Trustworthiness

Making data available is only half the battle; trusting it is the other. Pharmaceutical data is shaped by numerous sources of variability:

• Manual sampling

• Batch-to-batch inconsistency

• Equipment differences

• Operator effects

• Sensor drift

• Undocumented process changes

Compounding the problem, contextual information about why a run was performed, what constraints were present, and which deviations occurred is often missing or poorly recorded. Data-driven models excel at fitting patterns, including accidental or misleading ones, so without a strong foundation of trusted, contextualized data, these models are at risk of becoming high-confidence engines with low-quality insight.

The Cost of Data Generation

The challenge becomes even more pronounced when considering the cost of data generation. In contrast to digital industries, pharmaceutical data is expensive by design. Each experiment takes weeks or months of wet and dry lab work, costly raw materials, specialized equipment, GMP-grade facilities, and highly trained personnel.

Scale-up and technology transfer widen the gap even further. Much of the available data is generated at lab or pilot scale, yet the phenomena governing large-scale behavior, including mixing, heat transfer, mass transfer, and residence time distribution, are inherently scale dependent. Models trained on small-scale datasets frequently fail when extrapolated to commercial manufacturing, not because the algorithms are weak, but because the underlying physics have changed. In these scenarios, collecting more of the same data does not solve the problem; it merely reinforces the wrong assumptions and manufactures false confidence.

Improving Data Readiness

Given these inherent challenges, mechanistic and hybrid modeling approaches surface as key enablers for the pharmaceutical industry. Mechanistic models encode first principles, mass balances, energy balances, reaction kinetics, thermodynamics, and transport phenomena, providing inherent extrapolation capability and physical interpretability. They are grounded in reality rather than correlation, and critically, they do not depend on massive datasets to be useful; they hold up even when data is sparse, noisy, or imperfect.

Hybrid models, digital twins, and detailed mechanistic frameworks combine the strengths of physics-based modeling with data-driven techniques. Physics constrain the solution space, ensuring consistency with known laws and scale-dependent behavior, while data is deployed where it adds the most value: parameter estimation, model calibration, bias correction, and the capture of unknown or poorly understood effects. Rather than asking data to explain everything, hybrid models ask it to complement the understanding.

The payoff is not just a predictive model, but a decision-ready model – transparent, interpretable, scalable, and aligned with process understanding. Models of this kind support development decisions, enable more reliable scale-up, reduce experimental burden, and build confidence across technical, operational, and regulatory stakeholders.

Why Culture, not Capacity, Blocks M&S Adoption

Infrastructure cost and data readiness are formidable barriers to M&S adoption, but the most underestimated, and the most intractable, is company culture.

Most companies believe they are supportive of predictive modeling.  They endorse it in presentations and long-term vision statements, then resist it in practice.

Experimental Bias: “Show Me the Data”

Pharmaceutical development has been built on tangible experimentation for decades, and physical experiments carry an implicit legitimacy: what we see in the lab feels more real than what a model predicts. This bias runs deep, even when wet experiments are slow, expensive, or incomplete and, even though manually collected data can be inaccurate because of operator error, inappropriate test protocols, or faulty instrumentation.

In these scenarios, models arrive after the data and serve as explanatory tools rather than decision drivers. Modeling teams are brought in late, asked to fit the data, confirm what is already known, or explain why an experiment behaved the way it did, rather than to guide which experiments should be run a priori. Cast in this reactive role, the value of M&S is severely diminished.

True adoption requires a cultural shift from experiment led to decision led development, where models inform experimental design, reduce trial and error, and quantify risk before material is consumed. That shift is as much about trust and mindset as it is about math.

Risk Aversion and Regulatory Anxiety

Pharmaceutical organizations are, by necessity, risk averse. Patient safety, regulatory scrutiny, and stringent quality expectations create powerful incentives to avoid new approaches, especially anything perceived as a black box. In regulated environments, predictability and traceability count for more than novelty, and any new methodology must earn trust across technical, quality, and regulatory stakeholders.

Even though regulators have increasingly encouraged model informed development,^1,2 internal teams often hesitate to rely on M&S for primary decisions. There remains a lingering fear that if a result is not physically measured, it will not be accepted, or that simulation outputs will be difficult to defend during audits, filings, or inspections. Modeling therefore tends to be positioned as supporting evidence at best, rather than as a foundational decision-making tool.

The irony is that this cultural resistance persists even when traditional experimental approaches are statistically weak, incomplete, or poorly scalable. Limited batch counts, narrow operating ranges, and confounded experiments are routinely accepted as sufficient, simply because they are physical, while well-constructed models, especially mechanistic ones, are subjected to a much higher standard, regardless of the broader insight and predictive power they provide.

This is where Quality by Design (QbD), as described in ICH guidelines Q8 (R2) – Q11,³ fundamentally changes the conversation. Its core principles—understanding parameter relationships, establishing causality, identifying critical quality attributes (CQAs), and defining a scientifically justified design space—naturally align with mechanistic and hybrid modeling.

Mechanistic, physics-based models are uniquely suited to support QbD because they make assumptions visible, enforce mass and energy balances, and capture scale dependent phenomena that experiments alone often miss. Instead of producing opaque correlations, these models generate explainable knowledge: how inputs propagate through a process, why certain parameters matter more than others, and under what conditions failure modes emerge. Approached this way, M&S becomes a vehicle for demonstrating process understanding rather than regulatory liability.

The Silo Problem: Modeling as a Specialist Function

Organizational structure poses another cultural hurdle. In many companies, M&S lives in a small, specialized group set apart from R&D, process development, manufacturing, and tech transfer. This centralization may seem efficient, but it tends to reinforce the perception that modeling is someone else’s job. Process engineers focus on experiments, manufacturing focuses on execution, and modeling teams wait for requests. Siloed, in this way, M&S becomes transactional: a service function rather than an integrated capability. Requests arrive late, constraints are unclear, and results fail to shape real decisions because the teams responsible for execution were never part of the modeling journey. Adoption stalls not because the models fail, but because ownership is unclear.

Incentives That Work Against Modeling

Culture is shaped by incentives, and many organizations unwittingly discourage M&S adoption through the way they measure success. Experimental throughput, batch count, and on time milestone delivery are often rewarded more visibly than insight generation or risk reduction. Running “one more experiment” feels safer than trusting a model, especially when individual career incentives are tied to tangible lab outputs.

In that environment, modeling can be perceived as slowing things down, adding review steps, or introducing uncertainty, even when it ultimately saves time and cost at the program level. Without leadership explicitly rewarding model informed decisions, M&S struggles to gain traction beyond early adopters and internal champions.

Building a Decision Centered Culture

Successful M&S adoption does not begin with software or data. It begins with a cultural commitment to better decisions. In practice, that means:

• Applying modeling early, as guidance rather than validation

• Treating models as evolving knowledge assets, not final answers

• Rewarding teams for reduced uncertainty, not just experimental output

Mechanistic and hybrid models fit naturally into this cultural shift. They align with how engineers and scientists think, provide explainability, and scale beyond the data immediately available.

Why “Can the Model Run?” isn’t the Real Question

Even organizations that invest in infrastructure, assemble the right data, and voice cultural support for modeling tend to hesitate when the time comes to rely on models for real decisions. The question is rarely, “Can the model run?” It is almost always, “Can we trust the model?”

In pharmaceutical development and manufacturing, trust is not optional. Model-based decisions can affect product quality, patient safety, regulatory filings, and supply reliability. Skepticism is rational; the trouble is that the way validation is approached often turns healthy skepticism into paralysis.

---

---

The Validation Expectation Gap

A common but unspoken assumption is that models must reach a level of certainty comparable to physical experiments before they are allowed to influence decisions. In practice, that translates into unrealistic expectations: exhaustive validation across every operating condition, perfect agreement with limited and noisy datasets, and guarantees of performance in scenarios where experimental data does not, and often cannot, exist. Experimental data, tellingly, is rarely held to the same standard. Small sample sizes, narrow operating ranges, and confounding variables are accepted as a practical reality, yet models are frequently expected to outperform the very inputs that feed them.

Model validation does not mean proving a model is “right.” It means demonstrating that it is fit-for-purpose. The Context of Use (COU) defines that purpose: the scope of a computational model used to address a particular question of interest, the choice of model, what the model will be used for, how its results will be applied, and how heavily the decision depends on it.^4,5 The governing question is always the same: is the model sufficiently accurate and credible for this specific COU?

A model built to rank the sensitivity of process parameters, inform scale-up risk, or guide experimental design does not require the same validation depth as one used for real-time control or product release. Yet many organizations apply a one-size-fits-all mindset, which makes early stage and strategic models nearly impossible to justify. Trustworthy models are not defined by perfection; they are defined by:

• Clear scope and intended use

• Transparent assumptions

• Quantified uncertainty

• Documented limitations

• Consistent behavior with known physics and data

Mechanistic and physics-based models support this framework naturally. Their structure is explainable, their extrapolations are grounded in first principles, and their failure modes are often easier to diagnose than those of purely empirical models.

The “Black Box” Problem

Trust erodes quickly when models are perceived as mysterious black boxes. This perception is not limited to AI or ML; even complex mechanistic models can feel opaque if not properly built, documented, and communicated. To have confidence, stakeholders need to understand:

• Why certain assumptions were made

• Which data informed calibration

• Where predictions are strong, and where they are not

• How uncertainty propagates into decisions

Without this transparency, even technically sound models struggle to gain acceptance from engineers, quality teams, and regulatory stakeholders. Validation, moreover, is not a one-time event, and organizations that treat it as such rarely build the sustained confidence that M&S demands.

Trust is Built Through Repeated Success

Trust in modeling does not come from one perfect model. It accrues through the consistent delivery of value. As models reduce experimental burden, reveal non-obvious risks, explain scale-up behavior, and prevent late-stage surprises, confidence builds organically. Leaders become more willing to rely on simulations, teams bring modeling in earlier, and validation discussions shift from “Why should we use this?” to “How can we use this more effectively?” and “Where else can we apply similar models?”

Staying Focused on the Core Mission: A Case for the M&S Contract Organization

In an industry where the cost of wrong decisions is high, trust is the currency that determines whether modeling remains theoretical or becomes transformative. The commonality among the barriers explored here is the assumption that M&S adoption must be fully resolved internally before it can deliver value. That assumption, not the technology, is what stalls programs.

The industry already has a proven template for solving exactly this kind of problem. Pharmaceutical companies long ago accepted that they do not need to own every capability required to bring a product to market. Discovery chemistry, toxicology, and clinical development are routinely entrusted to CROs, and manufacturing to CDMOs and CMOs. Treating M&S as a highly specialized service – an M&S CRO – offers a pragmatic alternative to navigating these hurdles alone. It represents a disciplined allocation of capital, providing expertise on demand without the burden of ownership.

The value of M&S is available now. Partnering with experienced modeling teams is a legitimate, practical path to capture that value without waiting for ideal conditions that seldom materialize.

References

1. U.S. Food and Drug Administration (content current as of July 7, 2026). Model-informed drug development paired meeting program. https://www.fda.gov/drugs/development-resources/model-informed-drug-development-paired-meeting-program

2. International Council for Harmonization of Technical Requirements for Pharmaceuticals for Human Use. (2026). ICH M15 general principles for model-informed drug development (final guideline). https://database.ich.org/sites/default/files/ICH_M15_Step4_Final_Guideline_2026_0129.pdf

3. International Council for Harmonization of Technical Requirements for Pharmaceuticals for Human Use. Quality Guidelines. https://www.ich.org/page/quality-guidelines

4. Viceconti, M., Pappalardo, F., Rodriguez, B., Horner, M., Bischoff, J., & Musuamba Tshinanu, F. (2021) In silico trials: Verification, validation and uncertainty quantification of predictive models used in the regulatory evaluation of biomedical products. Methods, 185, 120-127.  https://pmc.ncbi.nlm.nih.gov/articles/PMC7883933/

5. U.S. Food and Drug Administration, Center for Devices and Radiological Health. (content current as of November 16, 2023). Assessing the credibility of computational modeling and simulation in medical device submissions: Guidance for industry and Food and Drug Administration staff.  https://www.fda.gov/regulatory-information/search-fda-guidance-documents/assessing-credibility-computational-modeling-and-simulation-medical-device-submissions

---

---

---