Designing sustainable production
Strategically shaping sustainable production: Find out how companies are using site planning, circular economy, energy-efficient…
blog
Successfully mastering the entry into industrial AI
“Artificial intelligence (AI)” – there is probably no other buzzword you hear more often these days in presentations at business conferences or in conversations at countless trade fairs.
Even though the term was originally coined in 1955 by J. McCarthy and M.L. Minsky, for a long time it was only really a topic in academia(source). Although the topic became increasingly important in business circles in the 2010s thanks to technological progress, the breakthrough in the general public only came at the end of 2022, when generative AI tools such as ChatGPT developed by OpenAI or BARD (now Gemini) developed by Google made the possibilities of today’s AI models tangible. Since this development at the latest, all companies have been concerned with the question of how artificial intelligence can be used profitably and what requirements are necessary for this.
In the following article, I will discuss the areas of application for AI in the industrial environment and why generative AI tools are just one of the many possibilities. Finally, I will explain why a process-oriented approach to the topic using current process mining tools such as Celonis’ EMS is a promising approach.
How important is AI? Why is it so difficult to get started? A few facts to get you started
As the capabilities of the models and technologies were primarily demonstrated in playful environments or using Lego miniature models, especially at the beginning of industrial AI development, the topic of artificial intelligence struggled for a long time with the reputation of being a hype that brought little added value. However, this impression has changed dramatically in recent years.
In a survey conducted last year (2023) by Bitkom, the industry association for the German information and telecommunications sector, in which 605 companies were questioned, more than 2/3 of respondents stated that they see AI as the “most important technology of the future”. Only 29% said they still believed it was just hype.
Developments in recent years have probably played a large part in this. The many playful and experimental approaches from the 2010s have now become strong business-oriented applications. Whereas many companies used to simply develop an AI use case for the corporate shop window, today the focus is on the quantifiable added value of the applications. AI now delivers the results that senior management expect from it.
This is also shown by the results of the SME Digital Survey for small and medium-sized enterprises. In this survey, well over 90% of the experts questioned stated that artificial intelligence can increase process efficiency and improve distribution. However, the results are particularly interesting because of another point:
Almost 80% of respondents stated that they see AI as a key to reducing personnel costs. A comparison of this figure with the responses from previous years shows that it has improved significantly since 2019. One reason for this is likely to be the increasing maturity of the technologies. In times of skills shortages and demographic change, this characteristic is particularly crucial. This means that development will continue in a focused manner over the next few years.
However, if you look at the survey in more detail, a major discrepancy becomes clear. This is because the creators of the surveys also asked the company representatives to what extent the respective company has implemented the new possibilities. Only 15% stated that they were already using AI-supported applications. For 28%, deployment is at least planned in the near future. In contrast, more than half of the companies surveyed (52%) have no plans to invest in this area.
So while 2/3 of companies consider artificial intelligence to be the most important technology of the future, more than half of companies are not planning to invest in this area. Where does this discrepancy come from?
The survey results also at least provide an indication of this question. When asked what the biggest obstacles are before starting AI projects, 98% of the experts surveyed answered “Lack of know-how/specialists”, 86% a lack of data and 85% a “Low level of digital maturity”. These figures show that the German economy has a lot of catching up to do in terms of “AI basics”.
Fortunately, the barriers to entry for the industrial use of the various technologies have become smaller thanks to further developments in recent years. In the following, I would also like to show you a way in which we believe it is possible to master this entry.
Challenges of AI projects
The challenges described above are certainly not the only ones. If you take a closer look at the challenges relating to “AI solutions” in companies, you usually come up with 5 major areas of challenge, which I will briefly outline below:
- Data: The topic of data is certainly the biggest challenge – without the right data, at the right time, in the right place, no AI project can succeed. Therefore, data availability and data quality should be looked at more closely for AI projects. Data availability in particular is a major issue. In an industrial environment, many factors are often decisive for the desired result. Without the complete picture, even the most advanced AI models cannot provide any added value. Once these problems have been solved, it is often the issues of data protection and data security that lead to problems during ongoing projects.
- Investments: Although the possibilities for using AI in companies have increased in recent years, so too has the amount of investment required to implement it. As a result, the view of the technology has changed, as described above. AI projects have a direct responsibility not only to realize added value, but also to quantify it. Quantifying the direct benefit is a particular challenge. For example, if orders are reprioritized as a result of an AI prediction, this changes the initial situation. As a result, it is no longer possible to determine the effect directly. A definition and possible quantification of the “success” should therefore already be available at the start of the project.
- People & Culture: Without employees, no company can be successful. Building acceptance for the new technology is therefore one of the biggest challenges. The key to this is building trust and involving employees in the development of solutions. The AI solutions must be perceived as support that takes care of the tedious tasks, while the increasingly valuable human resource takes care of the important exceptions. Of course, it is also a challenge to build up the necessary expertise for the development and operation of AI solutions in companies.
- Infrastructure: Ultimately, an AI solution must always be practical to use. The necessary infrastructure must be created for this, which is a challenge for smaller companies in particular. Training (some) AI models requires a lot of computing power and is therefore almost exclusively cloud-based. The source systems and data pipeline must run efficiently and reliably so that no high maintenance costs are incurred, which may offset the added value of using AI. Depending on the chosen implementation method and tools, this can be a major challenge.
- Ethics, accountability & traceability: The wider use of technologies has also sparked public and political debate about risks. Since the “EU AI Act” at the latest, there has also been a legal challenge to fulfill certain requirements. Both the assessment of the risk of an AI application and its transparency play a role here. Above all, this transparency is also critical within the company – as attention often has to be paid to traceability during implementation. A characteristic that not every AI method naturally has.
Process mining as a cornerstone and starting point
There are various ways to approach the topic of artificial intelligence in an industrial environment. These range from purchasing third-party solutions that specialize in solving a specific business problem (e.g. a tool for recording meetings or for visual quality control at the end of production) to setting up an AI department that natively implements and operates the AI technology. While the latter is certainly the ideal scenario, it is certainly also the most cost-intensive. This is only practical for large companies where the solutions produce the right economies of scale. The use of third-party solutions, on the other hand, can be implemented quickly. However, the diversity of challenges in German SMEs is so varied and the real core problems are often so individual that out-of-the-box solutions are either not available or only deliver inadequate results.
That’s why we often recommend using a framework as the cornerstone for the operational use of AI to get started with a data-driven decision-making culture: process mining. Before I explain the advantages of process mining technology in connection with the successful use of AI, I would first like to give a brief explanation of what process mining actually is.
What is process mining?
Nowadays, practically all company processes are supported by systems, be it in accounting, purchasing or order processing by an ERP system, in production by an MES system or in logistics by a warehouse management system. These systems store all changes made to a specific order, such as an invoice document. It is precisely this information, these digital traces that a process leaves behind in the system, that process mining makes use of. Process mining uses this change data to create a digital twin of the process with the help of additional process knowledge. This digital twin can be used to visualize and analyse all historical process runs. This enables problems to be understood and the process to be optimized. If you want to find out more about the technology itself, I recommend our white paper on the subject.
Process mining therefore offers the opportunity to take a process-driven approach to data analytics, which represents a more holistic approach to the otherwise common “root cause” approach. But why is process mining now also a good entry point into the industrial application of AI?
To explain this, I would like to go back to the challenges of AI projects described above. Based on this, I will outline the three main characteristics that a technology must have in order to lay a good foundation for the use of AI in companies:
- The technology must generate a comprehensive data basis that ensures data quality and provides a cross-source data structure for AI development.
- The employees who are to support the future AI applications in their daily work must trust the technology and, in particular, the measures derived from it.
- For an operational business, it must be ensured that the technology provides frameworks that enable monitoring of regular operation with as little maintenance as possible (especially for small and medium-sized companies with the smallest possible tech stack).
How can process mining tools be used?
Operationally designed process mining tools, such as the system from the current market leader Celonis, have all these features:
- Data: The aim of process mining projects is to create a digital twin of the process. Each historical process run and the changes it undergoes are reconstructed. This creates a holistic picture that translates the data into a language that the specialist department can understand. It does not matter which source system the data comes from. In the data model created, the information is already transformed into comprehensible activities. The process models generated in this way are verified directly with the specialist departments. Once set up, they can be used again and again in the long term without having to be validated again.
- People: It is precisely this close involvement of the specialist department in process mining projects (especially during validation) that has another major advantage. It creates trust. Thanks to the constant exchange between data and process experts, data errors are detected early on and the quality of the model is constantly improved. The goal of creating a digital image of the process plays a decisive role here. Of course, the data models are also validated in a direct AI project. These usually only show an incomplete picture of the overall process. By focusing on a specific use case, there is also a high risk of falling into AI-specific topics too quickly. As a result, the technical experts are quickly left behind.
- Infrastructure: This is probably where the current process mining tools differ the most. The process mining platform developed by Celonis certainly offers the most options at the moment. Data is extracted via connectors that are available for all common source systems. The cloud-based Celonis EMS offers its own environment for data transformation and the front end contains many process-related analysis templates. The current market leader’s offering is rounded off by its own action engine, monitoring and scheduling system, as well as the option to train AI models directly in Jupyter environments. Of course, there are other (possibly better and more comprehensive) tools for all these tasks, but for most small and medium-sized companies, the options are more than sufficient. If independent solutions are to be used in the future, the Celonis platform (both inbound and outbound) offers all the necessary interfaces.
Targeted use of AI – identifying the right use cases
The use of AI is primarily intended to improve existing business processes. Be it better scheduling of orders to prevent overstocking or shortages, prioritizing orders in production to improve the on-time delivery of the manufacturing process or optimizing stock allocation to speed up the delivery process. However, due to the high investment that an AI solution represents, it is crucial for companies to use the new technological possibilities precisely where they offer the greatest added value for the company.
This is where Process Mining shows its additional great strength, which we see as an important argument for strategic AI entry via this technology. Instead of deciding in advance which solution is decisive for the company, process mining generates complete process transparency. At the same time, the basis is created for any possible optimization (whether AI-supported or classic) and automation of the process. In this way, the process can first be analyzed holistically, simple adjustments can be made for process optimization and the decisive steps can be identified. Based on this, the best solution options for the company can be identified and implemented directly using the tools provided.
Dr. Clemens Wolf
Head of Digital Operations
Do you have any questions for me?
I look forward to an exchange on the topic and all aspects of AI and digital transformation. Please send me a message and I will get back to you as soon as possible.
Closing words
Both the development of process mining tools and the progress of AI methods is rapid, so this blog article is certainly only a snapshot. Nevertheless, I hope I have been able to give you an understanding of the benefits of combining process mining and AI. If you would like to find out more about the topic or current developments, I would be delighted to talk to you. You are welcome to use the contact form to send me your feedback or ideas on the topic.
Dr. Clemens Wolf
Manager Digital Operations, Frankfurt
He is responsible for the Digital Operations business area at Rothbaum. His goal is to drive digitalization forward with our customers from the manufacturing industry and to support operations with modern technologies and system solutions.
Johannes Mayr
Senior Consultant, Munich
As a senior consultant, he particularly supports projects in the area of digital operations. His focus is on analytics and process optimization using process mining.
Das könnte Sie auch interessieren
Process indicators: When data becomes meaningful
You don't know how good or bad you are in a particular area of the…
Improve on-time delivery and delivery time with process mining
From the customer's point of view, delivery time and adherence to delivery dates are the…