Procurement Risk Scoring via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current procurement monitoring systems fail to effectively identify high-risk procurements, particularly those below the typical dollar value threshold, leading to potential fraud, waste, and abuse, which can result in unnecessary expenditures and significant scrutiny.
Innovation Solution
A procurement analysis system that uses advanced analytic techniques, including data mining and machine learning, to develop scoring models that identify high-risk procurements by evaluating bid characteristics, supplier risks, and item risks, providing a dashboard for visualization and proactive risk management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional dollar value monitoring is used to identify high-risk procurements, then high dollar value procurements can be detected, but low dollar value high-risk procurements (e.g., $20 ice cube tray) are missed
Solution Approach 1:
The system transitions from monitoring based on a single parameter (dollar value) to monitoring based on multiple parameters including bid characteristics, supplier risks, and item risks. This is achieved through scoring models that evaluate various attributes such as bid price deviations, supplier performance history, and item category risks, enabling detection of high-risk procurements regardless of dollar value
Solution Approach 2:
The patent introduces scoring models as intermediary tools that process and analyze procurement data. These models act as mediators between raw procurement data and risk identification, transforming complex multi-dimensional data into actionable risk scores that flag high-risk procurements for further review
2Reliability
If all procurements are reviewed in detail, then high-risk procurements can be identified, but the time and resources required become prohibitively large
Solution Approach 1:
The system applies different levels of scrutiny to different procurements based on their risk profiles. Low-risk procurements receive minimal review, while high-risk procurements identified by the scoring models receive detailed examination. This localized quality approach ensures reliable risk identification without requiring uniform detailed review of all procurements
Solution Approach 2:
The system performs partial review actions on most procurements (automated scoring) and excessive/detailed review actions only on high-risk procurements (manual evaluation). This differentiated approach balances reliability with time efficiency by applying comprehensive review only where necessary
3Reliability
If post-procurement auditing is used to detect high-risk procurements, then fraud and waste can be identified, but the damage has already occurred and leadership time is consumed
Solution Approach 1:
The system performs risk evaluation during the bid evaluation phase, before procurement decisions are finalized. By identifying high-risk procurements preliminarily through automated scoring models, the system enables preventive actions to be taken before fraud or waste occurs, rather than detecting issues after damage has been done
4Measurement precision
If multiple scoring models (price risk, supplier risk, item risk) are implemented, then comprehensive risk assessment is achieved, but the system complexity increases
Solution Approach 1:
The risk assessment system is segmented into three distinct scoring models: price risk scoring model, supplier risk scoring model, and item risk scoring model. Each model focuses on specific risk dimensions and can be independently developed, validated, and maintained. This segmentation enables comprehensive risk assessment while managing complexity through modular design
Data Source
AI summary
Embodiments may include machine learning, including decision tree machine learning. Predictive variables may be selected for the machine learning through an iterative process. Predictive power of the predictive variables and collinearity between the predictive variables may be considered when selecting a set of the predictive variables for the machine learning.