Explanation-Based Supply Chain Planning
Almost all supply chain planning solutions deploy some types of heuristic and/or optimization algorithms to come up with a plan that meets the material and capacity constraints as well as management objectives. Some use primitive techniques such as scenario planning to examine a handful of scenarios, amongst thousands of others, to see which one they like better. Some systems adopt mathematical algorithms to come up with optimal results that need tweaking by the users to make it work, since mathematical techniques are not capable of handling many complicated constraints in the supply chain. And there are some systems that produce results by using a combination of AI and OR techniques. An example of this is Multi-Echelon Inventory Optimization (MEIO).
In all these cases, the system makes decisions in a “black box” based on the input from the users and the available data. However, once the results are produced, the user would have to spend an inordinate amount of time to understand why certain decisions were made. For example, the user may want to figure out why certain customers’ orders were delayed and not some others, or why a more expensive part was used or why inventory for a part number was increased. In all these cases, users need to explore and investigate many screens and try and find out why a certain decision was made and if it was the right decision and what other choices may have been available that the system could have made.
AI can help to explain to the users how and why certain decisions were made if there is full pegging capability in the solution. As the application solves for a solution, there are many decision points and branches that are considered based on objectives of the users. The system considers the management objectives to ensure the right decision is made at every decision point. In some cases, it may have to backtrack and choose another branch. These decision points are essentially the path to the solution and why it was produced. Using Generative AI, we are now able to explain, in as much detail as needed, to the user why a certain decision was made. It is as simple as a user asking “Order P123 is planned for late delivery, tell me why and what can we do to deliver on time.?” Given the system’s full pegging capability and traceability, we know what the cause is and what other options we might have. In this case it could be that we use a more expensive substitute part number to avoid lateness.
Furthermore, such interaction with the system can further make the system more intelligent via explanation-based learning, where each user interaction would be a learning point for the system from the user. Thus, the system would make “better” decisions in the future. To learn more about the use of Gen AI in supply chain planning visit Adexa.
