Research project leveraging collaboration to better detect corruption

Integrity Distributed (InDi) is developing a platform designed to find patterns of improper payments in corporate third-party payment systems without those companies having to share their raw payment data.

InDi is comprised of researchers from Massachusetts Institute of Technology (MIT) and Harvard Business School, technology companies Kona AI and Integra Ledger, five corporate law firms, and eight global Fortune 500 companies. The project received its initial round of funding from the AB InBev Foundation.

Here’s how it works: The platform compiles corporate third-party payment systems data in emerging markets, anonymizes it, and then analyzes the data company by company to measure corruption risk across the corporation’s supply side spending patterns. Each company risk scores flagged payments in its own data, across a library of hundreds of tests and behavioral algorithms. The highest-risk transactions are reviewed by each company’s representatives or outside counsel.

A predictive model is then created for each company, “designed to proactively identify a potentially improper payment based on the attributes of each transaction,” according to a white paper on the project. The characteristics of problematic payments are identified without including the company’s raw data.

Finally, the company’s model is combined into one “super-model” that allows all the project’s participants to share insights while also protecting data privacy and anonymity.

According to InDi’s white paper, released in November, the predictive value of identifying a potentially improper payment is 25 percent greater when companies collaborate compared to results when each company’s model is performed in isolation.

The aim, said Vincent Walden, chief executive of Kona AI, is to identify key features of an improper payment—perhaps some of the keywords they have in common or what tests they consistently hit on.

“The goal is to profile the attributes of a corrupt or fraudulent third-party payment,” he said.

Praneeth Vepakomma, a PhD student at MIT who led anti-corruption research for InDi, said the project builds upon recent advances in a decentralized machine learning technique called split learning, which allows for participating entities to train machine learning models without sharing any raw data.

“We integrate split learning technology from our MIT research with the know-how of anti-corruption experts and workflows to form a distributed model to detect vendor fraud, corruption, and circumvention of controls,” Vepakomma said.

For corporations, the collaboration allows participants to “train machine learning models to identify potentially improper or corrupt payments,” said Bryan Judice, global head of compliance data analytics and monitoring at Panasonic North America, one of the eight companies to participate in the first stage of the research. “Participants benefit from the models’ ability to learn across all participant data sets without having to share their data with each other.”

Jeannine Lemker, an InDi board member and former senior compliance executive at Meta and Microsoft, said governments have made great strides in collaborating to fight corruption. InDi uses technology so corporations can do the same.

“There are immense possibilities for private companies to change the world together,” she said. “InDi is a way to share collective wisdom on what corruption risk looks like and tackle it early.”

For the results to continue to improve and be more robust, the platform requires more companies to contribute their data and their analysis of that data to the project, Walden said.

“As more companies are added to the cohort, it is our expectation that the super-model results will continue to improve, thus fueling InDi’s mission to help organizations fight global corruption,” the nonprofit stated.