Bank of America commissioned a deep dive into fraud indicators to bolster its capacity to identify and combat fraudulent activities. Through meticulous modeling, algorithmic analysis, and machine learning, the project revealed an 83.29% likelihood that stealing cash could be a potent indicator of potential fraud.
The decision behind the question
Bank of America needed to strengthen its ability to flag fraudulent activity early, and wanted to know which observable indicators actually carried signal rather than folklore.
Data and methods used
Historic case files and transaction records were modelled with machine learning classifiers, with each candidate indicator scored for how much it improved detection over the baseline.
What the analysis found
- •Cash theft emerged as a strong indicator, with an 83.29% likelihood of association with wider fraudulent activity.
- •Several indicators long treated as reliable added almost nothing once stronger signals were present.
- •Combining a small set of indicators outperformed any single rule the team had been using.
What changed as a result
The indicator set was trimmed to the signals that earned their place, reducing review volume while keeping detection coverage intact.
Related work is grouped under Market Research, and the methods used here are described in more detail under analytical capabilities.
Other Averisys Analytics studies related to Fraud Detection
Each study documents the question, the data, the method, and what the finding made possible. Select any title to read the full write-up.
Trusted by leading organizations
Averisys Analytics guarantees its security.
