It takes lots of data to make AI work
Artificial Intelligence

AI in Fraud detection

Ayasdi offers big data analytics and artificial intelligence services through its software Ayasdi’s Model Accelerator(AMA,) which it claims can help enterprises in financial services predict and model regulatory risk using machine learning. Ayasdi claims their software can help banks with applications such as regulatory compliance for anti-money laundering (AML),  automatically monitoring customer transaction data to identify anomalies and reduce the false positive rates in fraud detection as compared to traditional rule-based methods. Ayasdi also reports that it’s platform uses Topological Data Analysis (TDA), which was developed for a project funded by DARPA.

Ayasdi claims banks and financial institutions can integrate the software into their enterprise data networks. The user could then upload customer transaction data or sales revenue records into the AMA. The algorithm behind the software would then be able to comb through the data to test and compare several different risk-models, such as loss-given default (LGD), probability of default (PD) and other regulatory models. The system then provides users with options of viewing insights from the data on a dashboard that allows them to search, discover and predict risks.

Below is a short 2-minute video demonstrating how AMA software can be used for fraud detection:

Ayasdi claims in a 2017 case study that they were chosen by Citigroup to help create justifiable models of Citi’s revenue and capital reserve forecast to pass the Federal Reserve’s Comprehensive Capital Analysis and Review (CCAR) process.

The CCAR process was initiated after the 2008 financial recession to assess the financial standing of banks and Citi had failed the first two out of three annual CCAR stress tests conducted by the Federal Reserve. A team of developers from Ayasdi worked alongside subject matter experts in the bank’s business units to understand and collect data on macroeconomic variables such as revenue and capital reserves as stipulated Federal Reserve.

Ayasdi’s machine learning platform was then used to correlate the impact of the increase or decrease in these variables on each business unit’s monthly revenue performance over a six-month period.

The company claims to have developed several models to predict the future performance of these business units under different market conditions using the AMA. The feedback learning component of the project was in the form of insights from the business unit heads, who were once again roped in to evaluate the predictive model’s final performance.

According to Ayasdi, before the integration, the regulatory methodology that Citi followed was a  nine-month process involving hundreds of employees. After the project, this time was cut down to three-months utilizing less than 100 employees.

Ayasdi also names HSBC as a client for an anti-money laundering application.

Ayasdi was co-founded by Gunnar Carlsson, Professor Emeritus in the Department of Mathematics at Stanford University, and CEO Gurjeet Singh who previously earned a PhD from Stanford in computational and mathematical engineering.