AI Makes Inroads into Mortgage Fraud Detection

Fraud is an issue in any business. In real estate, even consumer housing, the purchase numbers are large, and the potential impact makes risk management important.

Resistant.AI, a firm that applies machine learning to the detection and prevention of financial crime, recently announced that U.K. online mortgage site client Habito had claimed a “30% improvement in fraud detection.”

The use of computer systems to fight fraud in online commerce is mature. Originally, the work was done using statistical modeling and analysis systems. The software would review previous cases and then look for statistical relationships among many data variables. The intent was to find combinations that would suggest a greater likelihood of fraud. Then a seller—usually very large ones with sophisticated operations—would kick an order or application over to an internal anti-fraud unit, which could review the case and decide what action, if any, was needed.

Resistant AI does something similar, except using different computerized methods. Instead of people building statistical models, a company provides examples of results to train the application. The AI system will process information, receive ongoing corrections, and learn from them.

But these techniques have more flexibility than traditional statistical inference approaches.

“Habito’s first-line risk teams were facing challenges in assessing the authenticity of documents, and escalated cases were taking longer for financial crime investigators to resolve,” Resistant wrote. “To confirm authenticity, the teams were also relying on outreach to the institutions that issued the documents, which slowed down decisions.”

A Habito representative said that using the software could process the documents “far faster and far more accurately” than people could and “brings us to conclusions faster and with more confidence.”

“Resistant Al subjects every customer interaction to forensic analysis to detect document forgery, serial fraud, synthetic identities, bots, account takeovers, money laundering, and unknown financial threats operating at scale,” the company wrote. “Document Forensics was integrated as part of the Habito mortgage application process, with workflow triggers tied to the various verdicts provided to each document.”

Reportedly, the software could decline an application that demonstrated “clear attempts at fraudulent manipulation,” with murkier cases requiring more human review taking an average of 52 minutes less than previously.

The point to take is not that software will provide a magic solution, but that software offers powerful tools and that traditional tasks in CRE may be ripe for automation.