<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=228149342841576&amp;ev=PageView&amp;noscript=1">
Blog Banner - Fraud (1)

Written by Jeff Keltner, SVP Business Development

As banks and credit unions look to bring more products online, they face numerous challenges: from developing automated instantaneous credit models, deploying new digital workflows and interaction tools, all while training their employees on a new way of operating. But one of their largest concerns center around fraud, and how to handle the influx of fraudulent applications that many institutions see when they bring their processes online. This is especially true for new customers. That is why many financial institutions restrict their online products (particularly online lending products) to either current customers, or potential customers that they've already contacted with pre-screened offers.


AI-enabled fraud detection for digital lending 

Traditional Approaches of Fraud

Applying traditional approaches of fraud mitigation to the digital channel adds friction for borrowers, as well as increased costs for the institution.

  • Excess Documentation: Institutions have a standard set of documents, many of which can be minimized using AI-enabled analytics. This expedites the borrower’s application process without increasing fraud risk.

  • Operational Constraints: using an AI-enabled approach to fraud, combined with world-class credit analyst team enables banks and credit unions to originate loans more quickly and with less operational overhead than with legacy verification systems. This frees up teams to focus more on supporting customers.

  • In-Person Process: Multiple verification steps often require several documents and even a branch visit. This adds friction and frustration to the user experience, and increases application processing time. Given that personal loan customers often need cash quickly, this can significantly cut conversion rates and increase the likelihood that qualified borrowers will turn to faster alternatives driving adverse selection in credit performance.

AI-enabled fraud detection for digital lending can dramatically improve accuracy, speed, and efficiency - meaning an easier process for the applicant and lower costs for the lender.


Upstart uses AI based fraud detection that has limited fraud rates to <0.3%. 

Achieve Near Zero Fraud Rates with Upstart

Upstart's AI-based approach to fraud detection reduces friction for the borrower, lowers the cost of origination for the financial institution, while also achieving lower fraud rates than traditional methods. Upstart fraud detection has limited fraud to <0.3%1.

  1. Using Alternative Data: Beyond the standard branch process of looking at photo IDs and paystubs, it is valuable to look at what sources of data are uniquely available. There are opportunities to plug in APIs for all sorts of use cases in order to analyze bank transaction data, pinpoint specific lending patterns, or tap into a variety of different fraud services. There is a lot of information available, and leveraging that information creates a safer environment for the bank or credit union - and an easier process for the borrower. 

  2. Create a Holistic View of the Applicant: No single piece of information is going to solve the puzzle. We need to look at all the data that we have about an applicant. This is really a case where the whole will be greater than the sum of the parts. By aggregating all of the applicant data and augmenting it with additional data sources, our models are able to detect fraud signals that are stronger than those from the individual application data points.

  3. Targeted Review: Even with the best automated systems, some applications will still require manual review. The key is to use a scalpel and not a sledgehammer when determining what to review. For example, how few documents can you ask for safely? How much can you reduce the friction in the process, but still feel secure in your decision? Consider what's triggering warnings of fraud, and address those concerns. Whether it be identity fraud, income fraud, synthetic fraud or first-party concerns, prioritizing which documents you are putting through the review process will allow you to target the specific concern, and not have an overly broad process.

The Upstart platform can automate about 70%2 of loan originations end-to-end, with no human review and no manual document upload. This is achieved while limiting fraud losses to less than 30 basis points of originations. 

Delivering instant decisioning alongside an automated fraud and verification process that leverages our experience with personal loans removes manual documentation requests and reviews, so that borrowers can finish their application in one sitting. This leads to faster close times, and a larger portfolio of net new customers. 

Achieving Near Zero Fraud Rates

Watch the LinkedIn Lending Video


A safe digital process starts with building a system that brings in new sources of data, combines them in interesting ways so that each data source helps piece together the larger picture, and minimizes the friction when you do have to introduce manual review by asking only for the documents that address the specific fraud concern. Combining these key elements will help banks and credit unions build a great digital process that can increase consumer throughput, increase pull through, and reduce the incidence of online fraud.

 1Based on an internal Upstart data analysis of personal loans first payment defaults as of 9/30/2020.

2As of 9/30/2020. Fully automated loans are defined as loans originated end-to-end (from initial application to final funding) with no human involvement.