Transforming Supply Chain Data Quality with Machine Learning
Machine Learning (ML) based algorithms are slowly percolating into a wide variety of supply chain planning solutions. However, they play a key role in building foundational building blocks of supply chain planning as well, particularly in data quality management. Data quality has emerged as one of the key focus areas SAPinsiders have highlighted in their responses to the survey for our upcoming research report, Supply Chain Planning in the Cloud. As Ranjan Bakshi, CEO of
Prospecta, a provider of AI-enabled data quality and integrity platform, highlighted in a
conversation with SAPinsider:
“If your core data, like your spare parts or your heavy assets in manufacturing or your information around your suppliers, are not accurate or complete, it’s very hard to enable supply chain transformational projects.” In this article, we share a few ways Machine Learning (ML) algorithms are already being used for supply chain data quality management. We will specifically use vendor or supplier data management for these examples. Note that these suggestions are illustrative examples and are the ones that are currently being used in many solutions. Remember that potential use cases for leveraging Machine Learning for supply chain data quality management extend beyond these examples and beyond vendor data management.
Data cleaning: Unsupervised Machine Learning (ML) algorithms can be used to clean bad data. An example is vendor name and product description rationalization. This is a very significant application in my opinion since the world of analytics is pivoted on Garbage-In-Garbage-out (GIGO). Improving data quality enhances every other application area listed here. While I have mentioned only a couple of attributes here (vendor name and product description), if you have worked on spend analytics projects, you know that there are several attributes/fields that can leverage this ML-enabled rationalization.
Data categorization: AI algorithms can help categorize, clean and classify the data automatically. Using the example of vendor data management again, a supervised Machine Learning (ML) classification algorithm can automatically categorize transactions into spend categories, even if the user does not enter the correct category or enters the wrong category. While this is just one example, multiple types of classifications can be done on transaction data.
Anomaly detection: “Fat finger” errors are common when data is manually entered in spend systems. ML-enabled anomaly detection works at two levels. The first level is when the user is entering the information. The ML algorithm, based on other features of the transaction, like vendor, material ID, material type, material description, supplier, quantity, etc. can flag any “fat finger” errors since it would know from its “training” that the spend amount should be within a certain range. The second level is where the user still decides to go ahead and enter the amount, the algorithm will flag it as an anomaly and will trigger a notification to the concerned person.
Premium
