IBM’s Institute for Business Value found that 75% of CEOs believe generative AI will be key to competitive advantage. However, in the supply chain realm, despite the rhetoric, the current use cases for generative AI are limited. At supply chain and enterprise application conferences, software companies have talked about using generative AI to create user manuals. That is not exactly a game changer. They have also talked about the use of generative AI to improve supply chain system user interfaces. The idea here is an Alexa-type experience where a user asks a question and the technology searches through an application to find the answer. So, a user might ask, “who is the supplier on this order.”
But Tom Sorgie, the senior vice president of technology at Infor Nexus, says if providing an Alexa type experience is the best that you can do with generative AI, “then we are not going to do it.”
Infor Nexus is a supply chain network that connects businesses to their key partners and makes coordinating the flow of materials and information much more seamless. The solution is particularly strong in creating visibility and coordination of international transport, collaboration with supplier networks and automating trade finance processes.
Infor Nexus is prototyping deeper capabilities. Mr. Sorgie explained their goal: “We’re seeing the ability to communicate in outcomes, to communicate the way you would to a person about the thing you want to stay on top of.” A good example is saying “What are my demurrage issues at the Port of Long Beach?” Demurrage is a fine paid when a shipper can’t move a shipping container during the allotted period. For a big shipper, the fines can add up to hundreds of thousands of dollars per year.
The way a manager might talk to a transportation analyst would be to say, “I want to stay on top of any risk of demurrage at this port. Let me know a day before my free days are about to expire and which containers will cost me the most in demurrage if I can’t move them all.” This check involves connecting carrier contract data and shipment dwell times. The analyst then checks every day and then informs the manager if there is a risk. The digital assistant becomes that analyst.
“This kind of communication is more outcome-oriented,” Mr. Sorgie explained. “If you think about how people interact with our system, or any system, they have a bunch of mental checklists. Things that they check on every day or even throughout the day.” They look at the data and ask themselves, “is this a problem?” This kind of reasoning absorbs a lot of time and effort.
Then if they see that there is a problem, they must do something about it. With demurrage, would need to follow up with their carrier.
Infor Nexus’s approach is not to just give a specific answer to a specific question, but to provide the right data visual, in this example a matrix type view of several days of shipments with demurrage risk. Mr. Sorgie calls this “rich visual controls.” These visual controls present the information in an intuitive and verifiable way and enable the users to dive right into the data.” With an Alexa type question and answer user interface, if you don’t ask the right question, you don’t get an answer.
“By jumping to visuals, we are lowering the bar for occasional users,” Mr. Sorgie said. Supply chain systems, including Infor Nexus, can be intimidating. “There are a million places you can look. There’s often a lot of ways to interpret information. There’s a learning curve and it’s deep.”
Overcoming the Accuracy Issues Associated with Generative AI
Generative AI has been known to just make answers up. Will that be a problem?
Mr. Sorgie says the opportunity for that in their application than is much less than in applications like ChatGPT. There are not vast amounts of external data that are part of the training model as is the case for ChatGPT. “We’re trying to teach the learning model to be a Nexus power user, so that all the actual information it communicates is data from the platform. We’re really training the system how to work with the data in our platform in a very advanced manner, like a sophisticated user. And so, the surface area for a solution comes way down because it just needs to know how you get to information that exists in Nexus.”
And what is being communicated back is not just an exact answer. It is a visual control. It is data in context. Getting to context, and providing a human-type answer, is difficult for machines. But there is a powerful technology that can help with this.
Infor Nexus’s Knowledge Engine is built on a graph database. Graph databases go a long way toward providing context or what Infor Nexus refers to as “knowledge context.” Graph databases work best when the data that is being worked with is highly connected, with complex many-to-many relationships. A few supply chain software companies are starting to build their solutions with this technology at the heart of the application.
Mr. Sorgie says that the combination of generative AI and a graph database is powerful. The nodes in the graph are things people think of as master data – like the partners a company works with, the products, and the physical geographic locations in a supply chain. But that data is then enriched through all the interconnections between the nodes. Data is contained in the nodes and in the lines between nodes – the interconnections. “And so, when we want to drive knowledge, like language pathways, into our data, what we’re really saying is that all that this rich transactional data now has hooks into the knowledge context.” Those hooks then allow rich business data to be accessed.
A relational database has highly structured hierarchies of data. But a graph database operates more like the way a human mind operates. It can flexibly make connections and move from the beginning to end of a query in a very nonlinear manner.
