From the connected factory to intelligent decision-making

Industrial digitalization has significantly expanded the ability to monitor what happens across production systems. Sensors, controllers, connected equipment, MES, ERP, IIoT, and different layers of automation continuously generate information on production, availability, quality, consumption, maintenance, materials, and performance. However, as this infrastructure matures, a fundamental distinction becomes increasingly evident: a factory capable of generating and transmitting data is not necessarily capable of using it to make better decisions.

For years, a significant share of digital transformation investments has focused on connectivity, data acquisition, and process digitalization. These initiatives remain essential, but they provide the infrastructure for a more complex challenge: converting industrial information into knowledge that can effectively support decision-making. A study published in Production Engineering in 2025 identifies a disconnect between data collection and its application in use cases capable of generating value. The authors propose a “data-to-value” approach in which the intended application and expected value help determine which data and technologies should be used.

This evolution is already reflected in industry priorities. The 2025 Smart Manufacturing and Operations Survey, conducted with 600 executives from large manufacturing companies, found that 92% consider smart manufacturing one of the main drivers of competitiveness over the next three years. Among respondents that have already implemented these initiatives, reported average improvements ranged from 10% to 20% in production output, 7% to 20% in employee productivity, and 10% to 15% in unlocked production capacity. These figures indicate that the discussion around digitalization is beginning to move beyond technology adoption toward the effective generation of measurable results.

The question, therefore, is no longer simply how much data an industrial company can collect, but which decisions need to be improved and what information is required to support them. In industrial systems, equipment, people, materials, inventory, maintenance, sequencing, and capacity are interconnected, meaning that a localized change can affect multiple stages of the process. This complexity is what distinguishes visibility from decision-making capability and requires data to evolve from a predominantly descriptive function toward analytical, predictive, and, progressively, prescriptive capabilities.

Connectivity is the infrastructure, not the ultimate goal

The first requirement for this evolution is the integration of information generated within the industrial environment. In this context, MES systems play an important role by connecting data related to production, equipment, resources, quality, and traceability with other enterprise systems, creating a more structured view of the production system. A study published in Computers in Industry in 2023 identified MES/MOM as important elements for end-to-end integration in Industry 4.0 architectures and highlighted data extraction from legacy equipment, including older machines, controllers, and systems that were not designed for current connectivity standards, as one of the challenges associated with this process.

Proprietary protocols, outdated interfaces, and different architectures can make it difficult to integrate these assets with newer systems. A systematic review published in the Journal of Industrial Information Integration in 2024 reinforces this issue by highlighting interoperability and standardization as relevant components for integrating data and processes in smart manufacturing environments.

Even with the advancement of Artificial Intelligence, this infrastructure continues to receive significant investment. According to the 2025 Smart Manufacturing and Operations Survey, 78% of the executives surveyed reported allocating more than 20% of their improvement budgets to smart manufacturing initiatives, including technologies related to analytics, sensors, automation, and data infrastructure. The survey also found that 88% expected to maintain or increase these investments in the following fiscal year, demonstrating the importance placed on building a technological foundation capable of supporting more advanced applications.

For this reason, increasing the number of sensors or connected systems does not, by itself, create greater intelligence. Fragmented or decontextualized data can increase informational complexity rather than improve decision-making. MES contributes to consolidating and contextualizing this information, creating a more consistent foundation for visibility. The next step is to use this foundation to identify patterns, anticipate behaviors, and evaluate alternatives before implementation.

When data moves beyond explaining the past

The growing availability of industrial data has created the conditions for expanding Machine Learning and Artificial Intelligence applications in manufacturing. A scientific review published in 2024 analyzed 114 journal articles dedicated to the use of Machine Learning in manufacturing functions, classifying the research according to algorithms, input and output data, supported functions, and application domains. Despite advances in the field, the study also identifies fragmentation across applications, which are frequently focused on specific problems and objectives.

A review published in CIRP Annals in 2024 demonstrates the breadth these technologies have already achieved. The study examines Artificial Intelligence applications in production system planning, process modeling, optimization, quality control, maintenance, and assembly, while also identifying challenges related to data quality, interpretability, transparency, integration between physical knowledge and data-driven methods, and the reliability of results.

The contribution of these technologies lies in their ability to extract new information from available data. Maintenance histories can reveal patterns preceding failures, quality information can identify relationships between process parameters and nonconformities, while time series and predictive models can anticipate equipment, demand, or production behavior. Analysis therefore moves beyond answering what happened and begins to provide evidence about what is likely to happen.

However, anticipating an event and determining how to respond to it are different challenges. If a model predicts demand growth, for example, it is still necessary to evaluate how that demand can be accommodated: expanding resources, modifying shifts, changing sequencing, reallocating production, or investing in new equipment. Addressing this type of question requires understanding not only the predicted trend, but also how different alternatives may affect the behavior of the system.

From prediction to virtual experimentation

Digital Twins add a new dimension to analysis by enabling the evaluation of how industrial systems may respond to different conditions and alternatives. In complex environments, capacity, variability, availability, inventory, material movements, labor, sequencing, and demand continuously interact. Virtual models make it possible to represent these relationships and test scenarios before each alternative needs to be implemented in the real environment.

However, there are different levels of integration between the physical environment and its virtual representation. In a Digital Model, there is no automated exchange of data with the physical system, and updates are performed manually. In a Digital Shadow, data flows automatically from the physical system to the virtual model, allowing it to reflect changes and conditions in the real environment. In a Digital Twin, integration is bidirectional: information flows automatically between the physical and virtual environments, enabling a broader interaction between them. This classification is used in manufacturing literature to distinguish these three forms of digital representation according to their level of integration.

Regardless of the level of integration, the ability to use these models to support decision-making depends on their reliability. Data, assumptions, calibration, verification, and validation are fundamental to ensuring that results adequately reflect the behavior of the system being analyzed. The value lies not only in the visual representation of the industrial environment, but in the ability to build a quantitatively consistent virtual environment in which alternatives can be compared, capacity can be evaluated, changes can be tested, and potential impacts can be understood before implementation.

Intelligent decision-making emerges from the combination of different capabilities

The convergence of connectivity, integration, Artificial Intelligence, Digital Twins, and optimization is beginning to change the nature of the decision-making process itself. When an unexpected variation in demand occurs, for example, data can identify the deviation, algorithms can estimate its progression, a Digital Twin can evaluate different strategies, and optimization methods can explore combinations of resources, sequencing, and capacity. The value does not necessarily lie in any single technology, but in the ability to combine them according to the specific challenge being addressed.

A study published in Results in Engineering in 2025 provides a quantitative example of this approach. In an application within the steel industry, researchers combined Digital Twin, neural networks, Reinforcement Learning, Machine Learning, and metaheuristic algorithms to adjust strategies in response to conditions such as fluctuations in electricity prices, component degradation, and increasing demand. In the application studied, the methodology achieved a 5% reduction in cost per tonne produced, a 5% reduction in CO₂ emissions, and a 30% improvement in aligning production with demand.

More recent research also points toward increasing integration between models and human expertise. In 2026, a study published in the International Journal of Production Economics presented an Intelligent Digital Twin applied to production scheduling, combining Digital Twin and Machine Learning to generate decision policies and using four Large Language Models to explain them in natural language. The architecture also allows expert preferences to be incorporated into the proposed policies, bringing computational capabilities and technical expertise closer together.

The direction indicated by these studies is not necessarily the replacement of specialists, but rather an expansion of their ability to understand alternatives, consequences, and trade-offs. Algorithms can explore a large number of combinations, while virtual models enable their potential impacts to be assessed before implementation. The final course of action remains linked to business objectives, constraints, and priorities, but is supported by a broader quantitative foundation.

Better decision-making becomes the objective of digitalization

The digital maturity of an industrial company should not be assessed solely by the number of connected devices, installed sensors, or available dashboards. A company may have thousands of real-time variables and still make capacity decisions based on static averages, invest without evaluating systemic impacts, or accumulate years of historical data without using it to anticipate failures. The transition from a connected factory to intelligent decision-making occurs when digital infrastructure effectively begins to change how these decisions are evaluated.

This combination also changes the logic behind industrial investments. Digital Twins can support analyses of capacity expansion, equipment, layouts, and process changes before CAPEX is committed. In automation, the same principle applies to Virtual Commissioning, in which virtual models can be connected to control systems to validate PLC logic, communication, sequences, and interactions between equipment before physical startup. In both cases, the underlying principle is similar: obtaining evidence about expected behavior before implementation.

Historically, many important insights only became available after a decision had already been made: the capacity effectively achieved following an investment, automation issues identified during commissioning, or the behavior of a production line after a change. Advanced digitalization seeks to move part of this knowledge to before implementation, using data, models, and intelligence to evaluate alternatives and their potential impacts in advance.

This is one of the main differences between a factory that is simply connected and one oriented toward intelligent decision-making. The next stage of industrial digital transformation is not necessarily about producing more data, but about reducing the distance between data and decision-making. When connectivity, context, analytical intelligence, virtual experimentation, optimization, and human expertise work together, industrial organizations strengthen their ability to evaluate alternatives in advance and with quantitative evidence, reducing uncertainty and supporting more consistent choices for the future of the production system.

From technology to application: results across major industries

The application of these technologies is already part of projects developed by EYF Solutions for major industrial companies, supporting decisions related to capacity, productivity, logistics, maintenance, automation, and investments. Through the integration of data, Digital Models, Artificial Intelligence, and optimization, these studies enable companies to evaluate scenarios, quantify impacts, and compare alternatives before implementation, with projects delivering an average ROI of approximately 40%. Want to understand how this approach can be applied to your company’s challenges? Talk to our specialists.