Artificial intelligence is becoming a key topic in production optimization. We spoke with Dr. Rok Prešeren, an experienced data science expert at Metronik, about its practical impact and challenges.
How would you describe the current role of AI in optimizing production processes?
AI encompasses a wide range of technologies and solutions, which is also true for its use in manufacturing. Its role therefore varies depending on the application area. A good illustration is Gartner’s Hype Cycle, which shows where specific solutions currently stand in terms of maturity and adoption.
Among the mature and widely adopted technologies are machine vision solutions, which significantly boost productivity in many manufacturing environments. Slightly lagging behind but rapidly developing are solutions for autonomous guided vehicles (AGVs) in logistics and robotic handling (e.g., bin picking). These are approaching the so-called productivity plateau, where they begin to see widespread real-world implementation
At the peak of the Hype Cycle are currently solutions related to language models. We’re seeing a growing number of both general-purpose and highly specialized language agents, operating in the cloud or locally. These agents already significantly improve the efficiency of support processes such as maintenance or energy management. Metronik also offers a language-based co-pilot for production, an AI assistant that helps users quickly find the information they need, such as how to resolve an issue on the production line. A key feature of our solution is that it runs autonomously on the client’s premises, with no data being shared externally.
At the early stage of the cycle are highly targeted solutions for optimizing specific production steps. These are often based on proven machine learning technologies but require a customized approach, which means higher costs and slower implementation.
What are the most common AI use cases in manufacturing that you see with your clients?
Most use cases fall into three main categories: classification solutions, predictive models and optimization solutions. Manufacturing environments face many classification challenges, such as automatic defect detection, defect type categorization, anomaly detection in machine operation or identifying employee competencies. These solutions often serve as the basis for further optimization
The second group includes forecasting future system states. This includes examples such as predicting future energy consumption, product quality or early detection of potential failures. Due to their predictive capabilities, these solutions are a valuable foundation for the third group: optimization problems. One particularly interesting solution we developed is predicting employee competencies based on past performance, allowing for optimal workforce allocation to boost production efficiency.
Recently, due to rising energy prices, there has been growing demand for energy consumption management solutions. Here, reliable consumption forecasting is key. This is an area where AI can be directly translated into a tangible product. A good example is our product MePIS Energy, which uses predictive models to enable energy optimization at the production-wide level.
What conditions must be met for a company to effectively implement AI in its production process?
Successful implementation of AI in the manufacturing process requires certain key conditions to be met, which vary depending on whether the problem is a general one, such as predicting energy consumption, or a customer’s individual challenges. For the latter, in our experience, one of the most important success factors is the content manager on the client side as someone who is enthusiastic, committed and knowledgeable about their production process. This person is able to clearly define the problem and expectations, and often provides a one-page summary of the challenge before the first meeting, which is a great sign that the foundations for the project are solid. This usually means that the data is already available in digital form, that the client understands its purpose and scope, and that they can quickly access missing information. It is also important that there has been ‘evangelisation’ within the organisation, i.e. an understanding of why AI is being deployed, with participants focusing on how to deploy it in a way that makes substantive and economic sense. The content manager thus also becomes the main guarantor that the solution will actually come to life in practice.
To successfully implement AI, several key conditions must be met. These vary depending on whether the challenge is general, such as forecasting energy use, or highly specific to the client. In the latter case, one of the most important success factors, based on our experience, is having a dedicated project owner on the client’s side, someone who is enthusiastic, committed and deeply familiar with their production process. Such a person can clearly define the problem and expectations and often prepares a one-page summary of the challenge even before the first meeting. This is a strong indicator that the project has solid foundations.
This usually means the data is already available in digital form, the client understands its scope and purpose, and can quickly fill in any missing information. It’s also crucial that the organization has already gone through internal “evangelization”, meaning that they understand why they’re implementing AI and how to do so in a meaningful, cost-effective way. The project owner also becomes the main guarantor that the solution will be adopted and used in practice
Another important condition is the client’s willingness to conduct a pilot project or feasibility study. AI implementation is not yet a routine task. Each case requires a research phase known as “problem discovery”. This phase does not yield the final solution but reliably indicates whether the goals are achievable, what’s missing and what needs to be built. It’s important to understand that even this information has value – and a price.
The third element, which is often mistakenly considered the first, is the technological infrastructure. This involves whether processes are digitalized, which business systems are in place, how data is stored and accessed and how difficult it is to extract information from various sources. Diverse data is essential, as it enables AI models to learn. Again, the project owner plays a key role here, since they understand the digital landscape and can ensure the necessary data becomes accessible and usable.
How does Metronik integrate AI into production and energy management systems (MES and Energy)?
Metronik thoughtfully integrates AI into its MePIS MES and MePIS Energy platforms as functional modules addressing specific challenges in manufacturing and energy efficiency.
In MePIS Energy, AI solutions tackle general challenges such as energy consumption forecasting. This functionality works as a standalone module that can easily be added to the existing system. Implementation has no special limitations. Metronik can install energy meters, integrate with the appropriate databases and provide everything needed for model training and operation. A key advantage is that the model training process is fully productized, meaning clients can perform it themselves without additional developer involvement.
In contrast, AI integration in MePIS MES often involves custom solution development. Notable exceptions are the modules for predicting employee competencies and process stabilization. The former allows for optimal workforce allocation, while the latter detects and predicts critical process deviations, enabling preventive action. In most other cases, the AI solutions are project-based and tailored to the client’s specific needs. These run as separate applications connected to digital data sources via software or data interfaces.
This type of integration is more complex because it requires defining what data is needed, where it comes from, who manages it and how it will be integrated. Projects usually involve multiple systems and stakeholders, requiring extensive coordination, clear communication and a shared understanding of technical details. However, this individualized approach provides the most added value, as it tailors solutions to the client’s environment and specific challenges.
What advice would you give to companies considering their first steps into AI in production?
We advise companies to move from ideas to action as soon as possible. AI solutions are becoming essential, not only for increasing productivity but also for developing new products and services that ensure future competitiveness.
The first step should be choosing the right problem, one that the company can clearly define and summarize, and where there’s a capable and engaged project owner who understands the process and can shape expectations.
Next, the company should find the right provider to collaborate on a feasibility study or a smaller preparatory project if a full pilot seems too ambitious. When choosing a partner, look beyond technical skills – process expertise is just as important. The provider should understand the industry environment, standards and operations. Their long-term stability also matters. AI is not a one-off project but a lasting solution that must evolve over time. Your partner must be able to provide reliable support and continued development.
One common barrier is the perceived high cost of implementation, which still discourages many companies. But the EU offers targeted funding for AI deployment. Metronik has extensive experience in this area and can showcase many successful projects supported by EU funding.
With the right approach, entering the world of artificial intelligence is achievable, feasible and highly profitable in the long run.
The interview was originally published in Časnik Finance.
