Insuranceciooutlook

The Use of AI in Fraud Detection in SME Lending

By Sonal Jain, Chief Data Officer, Validus Capital

Sonal Jain, Chief Data Officer, Validus Capital

Artificial intelligence (AI, defined by FICO as the actual use or application of machine learning or ML techniques for certain tasks) is seeing wider adoption in the financial industry—for example, insurance and payments companies, and traditional banks. Supervised and unsupervised machine learning can predict potentially fraudulent and anomalous transactions and protect customer databases.

Alternative lenders—particularly those catering to small businesses—are becoming the latest target of fraudsters. LexisNexis Risk Solutions’ 2019 Small and Mid-sized Business Lending Fraud Study found that small banks and credit unions, and digital lenders saw an 8.2 percent increase in fraud levels over the past two years. They also had average revenue losses of 4.5 percent and 5.8 percent, respectively, to fraud, compared to 2.9 percent for banks with more than US$10 billion in assets.

This, therefore, merits investing in fraud capabilities to ensure protection against unforeseen losses.

Faking it until they make it

Fraud detection service DataVisor notes that there are three common kinds of fraud in the digital-lending industry: application fraud with synthetic identities (or identities where real and fake information are combined), loan stacking (where a fake applicant applies to multiple digital-lending services at the same time, with no intention to pay up), and account takeovers.

All three fraud types hinge on fraudsters accessing and stealing troves of the legitimate customer and business data from various financial institutions and executing these schemes at scale. Additionally, most fraud cases occur at loan origination—when someone first applies for a loan, and the lender reviews their eligibility and suitable loan amounts.

Among the tactics which fraudsters use to acquire said customer data for origination are phishing, SIM swapping, malware, and cold calls asking you to “verify” your accounts with your passwords. Fake loan applicants can also surface information through massive data breaches, and publicly accessible and customer-generated details (e.g., blogs, social media posts, photos) to create synthetic identities.

Other methods used in financial institutions include brute-force attacks and credential stuffing.

AI can help businesses cope with fast-developing and complex fraud schemes
As seen in payments fraud, classic rule-based systems—those leaning on user data such as geolocation, known addresses, and purchased items and amounts—can’t keep up with their speed and agility.

However, well-trained algorithms can spot irregular patterns in massive datasets much more quickly using both customer identity and behavior-based features, not to mention are likely to yield fewer false positives that ward off legitimate borrowers.

This can be leveraged in digital SME lending as well. The aforementioned LexisNexis Risk Solutions fraud study found that small-business lenders and digital lenders that poured the needed resources into fraud-detection tools, tech, and initiatives; and identified fraud sooner (e.g., at origination or within the first month) had only a 3.0 percent revenue loss compared to their counterparts.

It is critical for digital lenders to have access to credit bureau data, bank data, and fraud-data providers. Using data directly from the institutions themselves (and with the customer’s permission) reduces the digital lenders’ risk of dealing with fake or doctored documentation and synthetic identities.

"However, well-trained algorithms can spot irregular patterns in massive datasets much more quickly using both customer identity and behavior-based features, not to mention are likely to yield fewer false positives that ward off legitimate borrowers"

Non-traditional data such as psychometrics or behavioral and psychological information could also be used for borrowers with insufficient credit data, or are unbanked. Strong Know Your Customer (KYC) processes should also be in place and include newer steps such as image or video confirmation for new customers during verification and onboarding.

Having multiple data sources for verification and comparison will show customer patterns—and irregularities. For example, comparisons across databases will yield shared IP addresses or access points, bank accounts, and mobile numbers; or misspelled personal details that can be a red flag for fraud.

For submitted documents, optical character recognition (OCR) can extract the needed information, and pinpoint, which documents and digital files are legitimate or fake. Predictive and scoring models can also specify which borrowers are at risk of defaulting on their loans versus which borrowers are downright fraudulent, and speed up the underwriting process. Natural language processing (NLP) may be implemented to analyze unstructured data (for example, online posts) in different languages to detect anomalies.

Trust is vital in SME lending and fraud detection

Loans are built on trust. A borrower trusts they will receive timely monetary assistance. A lender trusts they will be paid back in full and on time. A lending platform trusts that both borrowers and lenders meet their terms and fulfill their respective obligations.

Trust in AI for digital lending is still evolving, given multiple participants may be involved. People still hesitate to hand over non-traditional data for service access, unsure of how this data is being utilized. Additionally, lenders may not communicate as often or as clearly as borrowers would like regarding how AI is used on their data.

It takes time for consumer and business trust to be earned. In the meantime, digital lenders must take the initiative to present both AI and short- to mid-term digital loans as helpful and safe—and build an excellent track record.

Weekly Brief

Read Also

Help employees learn the

Help employees learn the "how" for them to embrace the "why"

Robert Maksimuk, Business Intelligence Lead, West Bend Mutual Insurance
Claims Processing and Management: Optimizing the Complex Commercial Insurance Purchasing

Claims Processing and Management: Optimizing the Complex Commercial Insurance Purchasing

David Mathes, VP, Underwriting Systems, Ryan Specialty Group (RSG)
Data Can't Do It All: Why Advanced Claims Management Calls for Greater Harmonization of People and Data

Data Can't Do It All: Why Advanced Claims Management Calls for Greater Harmonization of People and Data

Antonio Rosa, Director of Strategy & Analytics, NSM Insurance Group David Gagliardi, Director of Corporate Claims, NSM Insurance Group
Innovation with Purpose: Delivering on the Promise with Claims Technology

Innovation with Purpose: Delivering on the Promise with Claims Technology

Jamie Lee, Senior Vice President and Chief Service Officer at Aflac
The Power of Digital Payment During a Crisis

The Power of Digital Payment During a Crisis

Elisa Logan, VP, Growth, VPay
It's a 2020 Trend: SimplifyYour Complex Documents

It's a 2020 Trend: SimplifyYour Complex Documents

Andi Dominguez, Principal, Insurance & Healthcare, Quadient