This photo was taken during France’s Bastille Day celebrations on July 14, along the Rhone River in the Isere department in the Auvergne-Rhone-Alpes region of Europe in the town of Montalieu-Vercieu.
getty
As the world’s leading provider of fireworks, the French pyrotechnics company Etienne Lacroix ships fireworks globally and has supported events such as the Burj Khalifa Grand Opening Fireworks Display in Dubai and Bastille Day, France’s national holiday celebrated on July 14 each year.
Logistics plays a major role for the global player in the pyrotechnics sector, as transporting hazardous materials such as fireworks requires special precautions, especially when crossing international borders. Labels usually include warning pictograms and complex regulatory information in various formats. Consignees like customs officers need to be able to tell at a glance what kind of goods are inside, what the required shipping conditions are and who is authorized to open the cargo.
“In addition, customers have their own specific requirements,” said Eric Marini, director of Information Systems at Etienne Lacroix. “There are different color codes for hazmat, depending on the country of destination. Red may mean danger in European countries, but in China, for example, it means celebration. Green has a different meaning in the Middle East than it does in the US, and so on.”
With so many different regulations in place around the globe, labeling shipments at the end of the production line is not only time-consuming, but also leaves a lot of potential for human error. Now, with the support of SAP partner STMS and SAP, the prototype for a generative AI solution that has the potential to reduce human involvement to a minimum was presented.
AI Business Co-Innovation Project
In early 2024, Sébastien Faure, general manager at STMS SOLUTIONS, a longtime SAP partner, and Etienne Lacroix, attended a Hack2Build event where SAP partners used SAP technology to address their customers’ pain points. The idea for a generative AI use case came up and STMS approached its client Etienne Lacroix.
At the time, the IT and business experts at Etienne Lacroix were very interested in how generative AI could help them streamline their processes.
“But I have to admit, I was a little skeptical,” says Marini. “Everyone is currently talking about AI, how to introduce it to the industry and the huge benefits we will reap from its use. But suggestions are few and far between when it comes to implementing it in a way that guarantees the company will benefit from it.”
“Together, we looked at the company’s pain points and discussed potential use cases with the SAP Co-Innovation Lab,” Faure said. “We got the demand from business that AI should help avoid human tasks that are error-prone and don’t actually add value.”
The generative AI supply chain use case
The team quickly identified that the administrative process of creating shipping labels was a task currently performed by human workers that required a large amount of time and focus. The huge potential for using generative AI was evident.
“With guidance from experts from SAP partner organizations, we were able to build a specific SAP application in SAP Business Technology Platform (SAP BTP) and SAP S/4HANA to organize and use AI,” says Faure. “We then gathered feedback from the business experts at Etienne Lacroix and, in the next step, brought in the generative aspect.”
To meet the regulations from the customer, the developers from STMS collected all the necessary data and created a model in the vector engine SAP HANA Cloud.
“Technically, this prototype uses everything SAP has to offer in terms of artificial intelligence right now,” Faure said.
The remaining human task is to verify that everything on the label is correct. “That was the most important request,” says Marini. “Human intervention must be guaranteed, as with all AI use cases.”
“It’s the generative aspect that makes all the difference,” said Miliau Pape of the SAP Co-Innovation Lab. LLM suggests what should be put on the specific label – such as warning pictograms – based on legal regulations, historical customer requirements, cultural standards, and so on.
Return Augmented Generation (RAG) provides context to the LLM and makes the output relevant and reliable.
“Simply put, when AI is trying to be as creative as possible, RAG provides guardrails, so it potentially doesn’t go wild and get stupid,” Pape said.
For each shipment, a notification is based on the destination, shipping route, customer-specific data, such as the storage and language this customer last requested, or the colors or markings used to indicate that the shipment contains hazardous materials and can only be opened by experts with a certain certificate.
“This generative AI use case, at this point, may still only be a prototype, but it’s a viable idea, an actual use case for a real pain point in our company,” Marini said. “It’s a very crucial first step in our AI journey.”
