He points out that the combination of generative artificial intelligence and data processing algorithms represents a real “game changer” in production. He explains that “the restart time following an incident is an important lever to increase the SRT of a line. As several of my customers have pointed out, this time is considerably longer when the outage occurs at night. Why? Because the manufacturer’s technical assistance is not reachable. To restart a line, two elements are required: a diagnosis and a procedure for resolving the problem. It is on this second element that conversational assistants are a valuable help as they provide the information needed by the maintenance team as clearly as a human could do over the phone and in several languages, if necessary. With these tools, it becomes possible to dialogue with the machine.”

Laurent Couillard believes that it makes perfect sense for AI assistants to be developed by manufacturers and integrated into machines. They have the best knowledge base as they register all their intervention tickets.” He added that these conversational assistants also benefit the technicians of the manufacturer’s after-sales service: “it changes their lives on the ground. And it gives them a real reason to document every intervention in the company’s information system.” Early observations from InUse customers indicate that the development of tools using generative AI could reduce calls for technical support by 30%. 

It encourages industrialists to integrate AI into their businesses, subject to appropriate control: “if you develop an application in a well-defined context, by limiting access to a closed knowledge base, there is no risk of hallucinating AI. If she doesn’t have the answer to a question, she will tell you. And you can ask the same question a hundred times, you’ll get the same answer a hundred times,” he says. The reliability of the troubleshooting instructions provided by AI is as crucial as the accuracy of the fault diagnosis.

Regarding the integration of conversational agents on production lines, Laurent Couillard insists on two aspects: it is important to be able to geolocate the question, because this allows you to go and find the answer in the right knowledge base. On a line, you can have five machines with an error code 42; you must be able to use the right procedure in the right place. I think we also need to provide end users with a prompt engineering function; this allows them to develop autonomy on their applications by building helpers that are adapted to each context and role.” He anticipates that intelligent agents will soon be introduced in factories: The AI agents will soon be able to analyse the question that is posed to them and chain the different operations to be carried out, both on the data side and in the search for information, in order to formulate the appropriate answer. Tomorrow an operator will even be able to use them proactively, for example when launching a batch, by asking them what problems they may encounter.”