Beyond the ChatGPT hype: AI-driven value in the industrial sector

Artificial intelligence (AI) is commonplace in every sector – the industrial sector is no exception here. In recent months, there has been unprecedented hype about the topic of AI, particularly generative AI applications. These technologies have an extensive range of uses and, as such, can potentially revolutionise the industrial sector. However, in our experience, there is still a lack of a clear overall picture regarding the ubiquitous and overwhelming topic of AI. There is significant uncertainty about the areas along the industrial value chain where AI can create real added value to organisations.

We outline successfully implemented use cases here and highlight both the operational benefits, such as improved efficiency and cost optimisation, as well as the challenges, such as data protection.

In the Zühlke blog post ‘ChatGPT: generate revenue, not just text!’, we explained at an early stage what large language models such as ChatGPT are and how the technology works. We also discussed its potential initial uses and showed that large language models (LLMs) of this kind could offer businesses a broad range of benefits, from improving customer service and automating routine tasks to generating content.  

Generative AI can produce significant cost savings and boost your business’ efficiency, but it is not without its challenges. You need to ensure, for example, that technology is used responsibly and ethically and that the generated text is checked to ensure it is both accurate and appropriate.  

With this as our basis, we will now take a deeper dive and outline various concrete use cases, examining the advantages, challenges and technological issues involved in the use of generative AI models and large language models in the industrial sector.  

First, let us dive into the areas of product management and product development.

Smart documentation system for R&D teams

In R&D, efficient documentation of developments and innovation processes is crucial but also time-consuming. Traditional manual methods are prone to errors and consume valuable time of your specialists. A solution to this problem is an AI-based smart documentation system, which leverages artificial intelligence to automate the capture, storage, and retrieval of R&D data.

This not only ensures the accuracy of documentation but also significantly reduces the workload of engineers and technicians. By integrating into existing processes and software, the smart documentation system minimises human errors and accelerates the entire documentation process. This allows R&D teams to fully focus on their core tasks, enhancing efficiency, precision, and productivity.

Automated programming with AI-generated code

In the blog post ‘ChatGPT: generate revenue, not just text!’, we demonstrated that ChatGPT could be used for tasks such as generating service or marketing text. The exciting use case of using ChatGPT to generate code is likely to have a far bigger impact on the industrial sector, for example, in enabling machine controls to be automatically programmed. New approaches use large language models to automate the programming of control systems for plants and machinery. Here, the employee simply outlines what they want the machine to do, and the AI independently generates the relevant code.  

But the possibilities do not stop there: large language models can be used to generate code for automated testing. A description of the desired tests is enough to produce executable code here too. These approaches are also raising major ethical questions regarding the use of AI-generated code. At Zühlke, we are working with customers to draft guidelines for the use of this type of code.

In the area of internal processes and manufacturing operations, artificial intelligence offers also great potential:

Knowledge database for more efficient processes and better customer retention

‘Ask your document’, in particular, opens up a broad range of possibilities in the industrial sector, speeding up and simplifying both internal and external processes. For example, a large language model could respond to an error message from a machine or to a freely worded question from an employee by providing the correct troubleshooting instructions from a database, including the right replacement parts and step-by-step instructions.  

As well as error messages, this example can also be applied to the composition of materials for process manufacturing, hazard warnings for transport, instructions for assembly processes, technical drawings, replacement part identification, certificates for materials or products, certificates of origin, sustainability reporting and much more.  

The ‘Ask your document’ approach can also support external business processes, particularly in the area of self-service. Given the increasing shortage of service engineers, a large language model based on an intelligent knowledge database could respond to customer enquiries and problems, and help to remedy issues by providing information on usage, error messages and troubleshooting, initial setup, and so on. 

AI-based process monitoring, optimisation, and automation using computer vision

In industrial (production) environments, there are often values, results, or specific events that cannot be directly captured by traditional sensors. This limitation can hinder precise process control and automation. An AI-supported computer vision solution, such as the one Zühlke implemented for Martin GmbH, provides an innovative answer: it enables the precise detection and measurement of these elusive factors, reduces the need for human intervention, and significantly enhances overall process efficiency. These are substantial advantages over traditional methods. Furthermore, this technology simplifies the extraction of relevant information from images or videos and facilitates automatic context recognition, as demonstrated in our project with PDR.

Large language models as a single point of contact for employees

System dialogue is another compelling use case that significantly boosts efficiency and effectiveness in the industrial sector. It is conceivable that large language models such as ChatGPT can be used as a single point of contact within a business. Employees do no longer have to toggle between the different systems used for ERP, WMS or MES. From the users’ perspective, it doesn´t matter which system they are currently working with, and they can simply interrogate the system and enter data by inputting questions or commands. Defined, login-based access rights will ensure that data input and the system’s responses can be restricted to appropriately authorised users. 

Such interactions can help employees with workflow management and allow personnel to be used more flexibly. If a staff member is absent at short notice, for example on health grounds, it would enable employees from other areas to step in and take care of certain activities, like in intralogistics or in service, with iterative support from the system. This would also apply to new employees or temporary staff, who can be trained up and productively deployed in far less time than is currently possible. 

Continuing along the industrial value chain, we would like to highlight a third focus area where we have already identified and implemented several beneficial use cases: the areas of marketing, sales, and service.

Automated quotation generation in marketing and sales

Sales teams in industrial companies, particularly in the machinery and equipment manufacturing sector, often face the challenge of creating detailed quotations for highly complex and comprehensive requirements within a short timeframe. This was the case for our client Burckhardt Compression: employees invest a considerable number of hours each year manually searching through product specification documents, which typically vary greatly in structure, content, and references—a tedious and time-consuming task.

Such manual processes are prone to errors, require significant time investment, and result in long wait times for your customers. Implementing an AI-driven automated quotation system can revolutionise this process. In the case of Burckhardt Compression, the solution promises not only to simplify internal processes but also to reduce the manual data extraction effort by approximately 10%.

Predictive maintenance and service support for maximum equipment availability

Maximising uptime is crucial for the efficiency and profitability of industrial companies. Unexpected downtime results in high costs. Compounding the issue, the increasing shortage of experienced personnel leads to significantly longer repair times. Outdated, reactive maintenance methods are thus becoming obsolete—enter Predictive Maintenance. However, even in this area, there is AI-based potential, as AI-driven Predictive Maintenance and so-called supportive repair guidance come into play.

Such systems can predict failures before they occur and provide step-by-step repair instructions. This minimises downtime and reduces maintenance costs by enabling proactive interventions based on data-driven insights. At the same time, repair efficiency is enhanced by guiding less experienced personnel through the repair process. This leads to a significant increase in productivity and a sustainable optimisation of your equipment’s operational time.

Efficient knowledge management, cost-optimised processes and better customer retention

The use cases described above clearly show that the industrial sector can benefit from generative AI. Use cases in the ‘Ask your document’ and ‘Dialogue’ categories not only enable more efficient searches of the business’ own data: in the process, they create a new dimension in knowledge management, too. Queries can be made in natural language with less dependency on keywords, and the models provide context-dependent results and allow iterative queries of the data via the chat feature.  

Large language models do not just boost efficiency and lower costs for internal and external processes, they also help businesses to respond to demographic change and the skills shortage, for example by making work easier for service staff. At the end of the day, when information in (self-)service is provided faster and with greater precision, this has the knock-on effect of improving the business’ customer retention and customer satisfaction. Code generation concepts already in place or in development enable businesses to accelerate, simplify and automate their processes, leading to increased efficiency and cost savings.

Data protection, costs and quality – the challenges of using generative AI 

It should be noted that using AI is not without its challenges. Below we examine and assess the most important issues. 

One key aspect here, without a doubt, is data protection. As a rule, businesses’ sources and documents tend to be highly specific, like service instructions for products, or even confidential, like process or function descriptions. It is crucial that these data and texts are not openly accessible, e.g. via ChatGPT. So, are businesses required to build their own AI model? 

In principle, it is possible for a company to develop their own AI model. But, at the current time, this will not be a cost-efficient course of action for the vast majority of businesses and, as such, neither an expedient nor necessary move. The use cases mentioned, however, can be effectively covered by an in-house system that augments the OpenAI and Microsoft AI models. In this scenario, a business can benefit from the impressive abilities of these models without making sensitive information publicly available. 

Internal company data can be indexed and input into the AI model with the question to improve or extend the existing large language model. In addition, traditional access control can ensure that sensitive data and documents are only searchable by authorised companies or employees, ensuring that technical and process-related data are protected.  

Our experience with personal data has also been positive, as the same standards apply here as with cloud storage. The processing for Microsoft’s version of ChatGPT takes place at Azure Cognitive Services in the Netherlands, within the EU. All models have been cloud-based so far, but rapid progress in terms of open source makes on-premises solutions a realistic prospect for the near future. 

Fees are charged for inputting your own documents and building a proprietary knowledge database, but such costs are much lower than the cost of building your own model. Large language models charge according to usage. In the case of ChatGPT, the fee is €0.0018 per 1,000 words for the initial one-off expense of establishing the knowledge database. On top of this, there is the cost of ongoing updates to the database and subscription costs, but these are not the main cost drivers. 

One final question that businesses often ask is about the accuracy and quality of the responses. As with human employees, there is no 100% guarantee that the responses provided are correct. However, it is easy to check the accuracy of the information found because the sources are provided. As with the introduction of any new technology, it is absolutely vital to adapt the underlying processes. 

Generative AI: ignore the hype – generate added value 

There’s no denying that ChatGPT has unleashed a lot of hype about the potential of generative AI – hype that is not always fully justified. But it is also clear that businesses that make optimum use of what it offers are benefiting from increased efficiency and effectiveness, saving time and money, and generating new sources of revenue.  

It is not always about creating new, generative use cases, existing use cases can just as well be improved too. (Generative) AI will have a lasting impact on industrial processes. But if the technology is really to become a game changer for the industrial sector, businesses need to start exploring the technology and its potential now.  And above all, companies should answer the question for themselves: Where do we see the greatest potential and why? Prioritise those use cases that clearly demonstrate a benefit and deprioritise those that don´t.

We are working on new ideas and developing use cases constantly to assist our customers with new technologies. If you have an idea for a specific use case or a question about using AI in your business, feel free to get in touch!