Public Contract Law Journal

When Artificial Intelligence Met Public Procurement : Improving the World Bank’s Suspension and Debarment System with Machine Learning

by Kunmi Ageh

Kunmi Ageh (Kageh@law.gwu.edu) is a J.D. candidate at The George Washington University Law School and served as Senior  Articles Editor of the Public Contract Law Journal during the 2018–2019 academic year. She wishes to thank Professor Bryan Byrd for his guidance. She also would like to thank her friends (Kim, Liz, and Ana) and her family (Victoria, Funmbi, Tosin, and Bola) for their support.

I.  Introduction

On February 1, 2017, the Sanctions Board of the World Bank Group1 (the Bank) debarred an undisclosed respondent and that entity’s affiliates2 for fourteen years due to corrupt and obstructive practices.3 This case arose from a Health Section Reform Project in Romania. Romania (borrower), the Bank (lender), and the European Investment Bank (additional lender) entered into a loan agreement worth $80 million to promote emergency medical care.4 Romania’s Ministry of Health issued a bid for medical, laboratory, and emergency care equipment.5 The undisclosed respondent was awarded nine out of the twenty-one bid lots available, totaling over $7 million in contracts.6

The Integrity Vice Presidency (INT)7 argued, among other things, that the undisclosed respondent engaged in obstructive practices when it refused to adhere to INT’s requested audit.8 The undisclosed respondent contended that it refused the request because of an ongoing criminal case in the national court and argued a right against self-incrimination.9 The Sanctions Board ruled that the undisclosed entity engaged in corrupt practices when it offered to bribe the Procurement Advisors with personal trips and a percentage of each bid won.10 Corruption and fraud is a universal problem that countries, contractors, governments, and banks seek to eliminate.

Governments and banks establish suspension and debarment systems (S&D system) to regulate and sanction contractors or entities who engage in sanction-able activities.11 Within the Bank’s sanctions system is the Sanctions Board,12 whose role is to determine whether the INT’s evidence meets the burden of proof, more likely than not, that a respondent engaged in sanction-able practices.13 The INT conducts investigations on Bank-financed contracts, and submits cases, with a likely possibility of misconduct, to the Evaluation Officer of the Sanctions Board (EO) or Suspension and Debarment Official.14

Concurrently, domestic banks continue to fight the war against corruption with various techniques.15 With the increasing use of credit and debit cards, companies look to techniques that are more efficient, such as machine learning,16 to determine unusual consumer spending and fraud.17 In machine learning techniques the computer system is first given information on a pattern of normal spending habits, which are tested through algorithms that determine the probability of fraud in future spending.18 When transactions that are outside the consumer’s normal spending habits occur, the system receives an alert and the consumer may receive a text message confirming the transaction.19

Machine learning is a concept that stems from the evolving wave of artificial intelligence (AI).20 AI improves accuracy, increases speed, and minimizes human error. Similar to the use of a calculator, AI is the quicker way to a right answer. Machine learning is a process that can occur in the realm of AI and Big Data.21 Big data, specifically individual data, is the computerized ability to collect and store large amounts of data daily.22 Through the internet, companies can collect information based on our patterns and behaviors. The system uses the collected data to support aspects of everyday routines, including shopping needs.23 The system also can use the collected information to target consumers with personalized advertisements.24

Specifically, machine learning is the method of collecting or tracking data through the process of data mining.25 Machine learning also can be defined as “the study of computer algorithms that improve automatically through experience.”26 Through this method of data collection, the system detects and organizes the data into patterns.27 These patterns can be predictive, meaning used to review the pattern for future use, or descriptive, meaning used to understand data in present time.28

This Note considers the use of machine learning in suspension and debarment as a continued effort to collect data, promote efficiency, and combat fraud. This Note begins by introducing the Bank’s current sanctions procedure and domestic banks’ fraud prevention techniques. Part II provides a comparison of suspension and debarment by reviewing the systems of the United States, Nigeria, and the Bank. Part III elaborates on the Bank’s sanctions procedure and discusses the issues and setbacks in the sanctions process, specifically lack of data collection, borrowing countries’ acquiescence to corruption, and contractor resistance. Part IV provides an overview to machine learning, and discusses the modern uses, advantages, and disadvantages. Part V introduces the use of machine learning in the Bank’s sanctions process, specifically using mandated invoices and receipts collection to promote transparency and increase data collection, resulting in efficiency and transparency.

Premium Content For:
  • Public Contract Law Section
Join - Now