Most of the hype related to GenAI in the industrial environment centers around applications and use cases. I call this output-driven.
Users and internal sponsors find it easy to understand this approach because they can see the benefits and ROI of the solutions. There is a tool, clear CAPEX and OPEX, and an obvious result.
The approach is transparent and straightforward: Do you need an industrial co-pilot? You got it. Do you need a knowledge management tool? Here it is.
And it does not require a redesign of processes. You can simply create a new process or add the AI-powered app to a current process (e.g., as a recommender for a service desk operator).
But what about improving the entire process using GenAI, ML, and automation? In such a case, the journey itself can be regarded as the goal.
This, of course, requires process redesign — and your organization has likely undergone such an exercise several times over the last 10 years.
However, this time around, there is a technological leap provided by GenAI, which enriches a powerful tandem of AI and automation.
Take, for example, the production planning process in an engineer-to-order environment. Workflows include complicated order management, production planning, material management, logistics, and information flowing across stakeholders from different departments as well as sales and even customers. Software handles the situation by passing information smoothly among participants and by leveraging the data warehouse and real-time data from OT systems.
Using lean approaches and digital technology convergence, such transitions have delivered successful initial results including first-time-right outcomes, absolute transparency, and customer satisfaction.
Nevertheless, the software was, typically, heavily customized to fit the company’s needs, process owners, and other software.
The next step involves redesigning processes to apply robotic process automation (RPA) and enable the automation of repetitive and rule-based tasks to make them more efficient and fault-proof.
AI can now come into play.
Important: Do not consider a process optimized and efficient until it is super-optimized and super-efficient. All the people involved in the process, the data inputs and outputs, and the value obtained by the process’s customers, must be considered.
If process stakeholders need decision-making support, consider deploying AI chatbots and GenAI assistants to enable quick access to information and data analysis.
Collaboration is the new holy grail of time to value. Think about deploying AI-based collaboration tools to enhance communication and coordination among different departments involved in the process.
AI can optimize the allocation of resources (materials, labor, machines) in the planning process to maximize efficiency and minimize costs. Planners can benefit from AI-powered simulations of various “what-if” scenarios to understand the impact of different variables on production and customer delivery dates.
The entire process can be virtualized in a digital twin or an industrial metaverse. These will simulate different production scenarios to help find the best strategies without disrupting actual operations.
In addition, a trend to monitor closely is the emerging LLM agent technology, which serves as the “glue” between various process components. Dr. Michael May, Head of Technology Field Data Analytics & AI at Siemens Technology, advises that selecting appropriate use cases is crucial. This is due to the early stage of the agent approach (or any method integrating LLMs across a workflow), making it challenging to trace errors within a complex chain.
The Bottom Line
I believe organizations are too focused on single use cases and are missing the broader goal of becoming more efficient and resilient. They need to build AI-enabled processes that are still people-centric at some point and resilient.
Sparse data, data quality, trust issues, and transferring best practices across factories are some of the key challenges that must be faced. The good news is that there are solutions.
Sparse data can be addressed using synthetic data and reinforcement learning. Another crucial point is making AI accessible to non-experts. This can be achieved by using pre-trained models.
Ultimately, success requires close collaboration between process owners, hands-on users providing feedback, process optimization teams (focused on lean, efficient processes), technology experts (sharing their knowledge), and systems integrators (bringing the process to life).
The day is approaching when AI will design processes, automate tasks, and train both ML models and people. We’re not there yet, but it’s on the horizon!
