Editor's note: The SCM paper, “Procurement Control Tower: A Proof of Concept with Machine Learning and Natural Language Processing,” was written by Bishwajit Kumar and Pablo Barros Gomez and presented by Dr. Elena Dugundi.[email protected]) and Dr. Thomas Koch ([email protected]For more information about the research, please contact your thesis supervisor.
Our capstone project sponsor, a global pharmaceutical company, faces significant challenges in its procurement processes due to high procurement spend, diverse product needs, an extensive supplier base, and diverse software insights. The company recognizes that gaining insights faster, enhancing decision-making capabilities, and optimizing exception management are essential to staying competitive in today's volatile, uncertain, complex, and ambiguous (VUCA) market. As a potential solution, the company wants to explore the value proposition of implementing a procurement “control tower” – a centralized platform that provides end-to-end visibility and control over the procurement process.
To address sponsors’ objectives, our research sought to answer two key questions:
- Does a Procurement Control Tower create measurable value to the sponsor’s procurement function?
- Can you prototype a use case for Procurement Control Tower to demonstrate its value?
Two studies: qualitative and quantitative
Our research was a two-stage process. First, we conducted qualitative research to define the scope, value proposition, and implementation strategy of the Control Tower. We interviewed various procurement process experts to understand their existing challenges. Additionally, we researched industry best practices and aligned with sponsors on specific use cases that the Procurement Control Tower would cover, such as spend analysis, contract management, risk management, and supplier management.
The qualitative study proposed a comprehensive architecture for the Procurement Control Tower and outlined its value proposition to sponsors. The first critical step in implementing a Control Tower is to consolidate data from disparate data sources into a common data layer. This consolidation ensures a single version of the data (SVOT) and serves as the foundation for the Control Tower. The consolidated data and information becomes easier to find, eliminating inconsistencies that arise from differences in source data. The consolidated data enables enhanced data analysis from a single source, enabling the Procurement Control Tower to generate valuable business insights.
We then conducted quantitative research to develop a prototype (proof of concept) for a spend analytics use case, specifically focusing on materials spend classification. This use case has significant value to sponsors because approximately $250 million worth of spend data remains unclassified in terms of precise category or subcategory classification. Lack of precise classification of spend hampers business analysis based on spend categories and increases the chance of inaccuracy.
In our quantitative study, we compared multiple machine learning algorithms, including logistic regression, decision trees, random forest, and XGBoost, to use the data to predict the appropriate material classification of unmapped spend. After careful evaluation, we selected Random Forest as the best performing algorithm in terms of accuracy. To further enhance the predictive power of the algorithm, we preprocessed the data using natural language processing (NLP), a computational technique designed to mimic human-like understanding of text. The final algorithm achieved a classification accuracy of 94% at the category level and 90% at the subcategory level on the uncategorized spend data.
Benefit from advanced insights and implementation
Implementing a Procurement Control Tower provides sponsors with advanced insights resulting in end-to-end visibility, enhanced exception management, improved decision making, improved risk management, and cost savings. Classification of unmapped material spend by machine learning algorithms positively impacts sponsors’ businesses in multiple ways. Specifically, it creates opportunities for renegotiation with suppliers, improves budget accuracy, and reduces the man-hours required for manual classification. Considering that sponsors add thousands of new SKUs every year, which leads to tens of thousands of spend data records, our proposed solution is highly valuable as it allows for ongoing and regular classification of spend data. Our proposed solution has been accepted by sponsors and implementation is underway. This is a big step towards optimizing procurement processes and gaining competitive advantage in a VUCA marketplace.
Each year, approximately 80 students in the MIT Center for Transportation and Logistics’ (MIT CTL) Supply Chain Management (SCM) master’s program complete approximately 45 one-year research projects.
These students are early career business professionals from multiple countries with 2-10 years of industry experience. Most of the research projects are selected, sponsored, and carried out in collaboration with multinational corporations. Collaborative teams, including MIT SCM students and MIT CTL faculty, address real-world problems. This series brings together a selection of the latest SCM research.
