Management accountants can help their organization to identify, mitigate, and eradicate fraud by urging leadership to invest in technology powered by AI and machine learning. Such an investment can bolster the organization’s finance, IT, compliance, and risk management functions’ efforts to pinpoint misconduct and bad actors early on. Ultimately, that’s likely to enhance the reputation of the organization’s personnel for conducting themselves ethically and protecting stakeholders from fraud.


Julian Shun, an associate professor of electrical engineering and computer science at the Massachusetts Institute of Technology (MIT) and a lead investigator in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), is the lead instructor of MIT Professional Education’s Graph Algorithms and Machine Learning course. His research focuses on the theory and practice of parallel algorithms and programming, with particular emphasis on designing algorithms and frameworks for large-scale graph processing and spatial data analysis.


Graphs are a way to model relationships in data. A graph has vertices and edges, where vertices model objects of interest and edges model relationships between these objects. Shun says that graphs can be used to model social networks, where the vertices are people and edges represent relationships or connections between them, or financial transaction networks, where vertices are buyers and sellers and edges connect individuals who made a transaction with each other.


“Graph algorithms are programs that one runs to find patterns or anomalies in graphs—they can be used to find communities of people with similar interests in social networks or detect anomalous behavior in financial transaction networks,” Shun says. “Data analytics, AI, and machine learning are all related to graph algorithms.”


To use graph algorithms in fraud controls, finance professionals need to first take the data set and determine whether there’s an appropriate graph representation of the data such as an adjacency list, matrix, or set. Then, they would need to convert their data into a graph format.


Graph algorithms are frequently used in machine learning pipelines, Shun says. They can be used to cluster data for classification tasks or to identify anomalous structures in a graph to detect fraud or spam, as well as build and train graph neural networks, which can then be used to predict a graph’s vertex attributes, missing edges, or global properties.




Finance professionals can use a graph to model financial transactions. Edges in this graph are directed—if person A made a payment to person B, then the edge would point from person A to person B.


“Oftentimes, in money laundering, some person X sends money to other people, but the money eventually ends up back at person X,” Shun says. “By using graph algorithms, one can find a sequence of vertices, connected by directed edges, that start at person X and eventually end up at the same person X.”


This is called a directed cycle, which is a red flag. It can be a candidate for further investigation into potential money laundering (although not all instances of this pattern indicate money laundering).


“Graphs are usually large enough that humans can’t manually check for such cycles, so graph algorithms can significantly narrow down the number of cases that financial analysts need to focus on, thereby increasing productivity,” Shun says.

Graph algorithms can also help detect insurance fraud. One can model the customers in an insurance company using a graph, where edges connect vertices of people involved in the same claim.


“If one finds a group of people that are always involved in the same claims across multiple lines of businesses in the company, then there could potentially be insurance fraud,” Shun says. “Using graph algorithms in this scenario will significantly increase productivity, as these patterns are difficult to detect manually.”




Data privacy is an important concern when graphs are used to represent people and their relationships. Using algorithms to find patterns or anomalies in graphs often requires a global view of the data, because many algorithms take into account both local and global features, Shun says. Therefore, analysts studying these graphs will have access to a lot of confidential data, and it’s important that they abide by data privacy rules.


Choosing which data set to analyze must be done by a human, and algorithms could be biased by what data sets the human believes may have relevant information. Choosing which graph algorithms and setting their parameters to use on a data set must again be determined by a human, and, according to Shun, such a decision could be biased by what they think they should be finding in the data set.


The output of an algorithm will eventually lead to some decision being taken, and the outcome may not benefit everyone—and may potentially harm some people. Management accountants can help to gauge the potential impact on each group of stakeholders.


“Recently, there has been a whole lot of research on how to design ethical AI and machine learning algorithms and frameworks, and this body of work is important to look at when deploying these solutions in the real world,” Shun says.


Training in the space of graph algorithms would help to improve the productivity of finance professionals who analyze data to find patterns or anomalies, he says. It would also be beneficial to managers of various teams, as they will learn about new ways to analyze data and be able to make better decisions about what kind of data, algorithms, and tools would be the most effective for their team to use.


Financial organizations are now collecting and managing more data than ever before, and that volume is only going to keep increasing each year. Graph algorithms can help management accountants position themselves and their organizations to create value from that data, better serve their customers, and maintain a competitive edge.

About the Authors