March 31, 2022
Join Infor scientists Dr. Ulas Cakmak and Dr. Inderjeet Singh at the 2022 INFORMS business analytics conference April 3-5 in Houston to learn how machine learning (ML) enabled Infor customers handled supply chain disruptions during the COVID-19 pandemic.
Supply chain disruptions have always been a part of business, and recent tumultuous times emphasized the importance of the right way of handling them. Frequently occurring minor disruptions are often ignored in operational planning, likely because these are not expected to cause serious issues as long as they are not widespread.
Most modern machine learning methods can, indeed, mitigate the risk caused by randomly spread minor data anomalies, however, even the best forecasting models are susceptible to invalid data points caused by large and persistent supply chain disruptions.
At INFORMS, we will present a forecasting approach combining machine learning and traditional time series-based techniques. This flexible approach, ornamented with outlier detection and lost sales estimation, allows us to build forecasting systems that are robust against supply chain disruptions. Machine learning provides the ability to estimate the impact of various causal factors from a holistic viewpoint, while traditional methods ensure the lower-level characteristics are not lost in the process. Moreover, these systems provide critical input needed in making, pricing, promotional, and markdown decisions.
The pandemic caused various changes in consumer behavior, some ephemeral and some here to stay, at least to a certain degree. In particular, lockdowns and store closures implemented in early phases led to increased demand for essential goods and the opposite for nonessential products. This was accompanied by a transition to same-day store pickup of online orders. Excessive buying of essential goods resulted in unexpected peaks followed by stockouts. Frequent use of same-day pickup and quick home delivery options implied using store inventory to fulfill an important part of the online demand, instead of warehouse inventory used traditionally for this operation. This created an extra layer of complexity in managing inventory. For nonessential goods, demand decreased and some of it shifted online. Moreover, store closures caused lost sales without stockouts.
Data anomalies like the ones caused by the Covid-19 pandemic are by no means unusual, however they tend to be localized and short lived. The pandemic simply exacerbated the disruptions in both magnitude and duration. Left untreated, these issues are bound to cause degradation in forecast accuracy.
Almost a decade ago, Infor designed customized solutions incorporating outlier detection and lost sales estimation into specific forecasting applications. These proved quite efficient in handling major disruptions.