How Ally uses AI to approve car loans

Ally Financial claims to have found a way to eliminate much of the manual work involved in processing car loan applications.

It uses artificial intelligence software developed by a San Francisco fintech called Informed.IQ that verifies documents and data in real time. The technology extracts data points from loan documents and compares them against numerous databases to confirm identity, employment, income and other vital information.

Ally Financial in Detroit, which has $182 billion in assets, isn’t alone among consumer lenders embracing AI.

According to a survey of nearly two dozen financial institutions by Leslie Parrish, senior analyst at Aite Group, 45% said they use AI for loan origination and 40% said they plan to do so. Additionally, 55% said they use AI to detect fraud and 25% plan to do so. The survey was conducted in the first quarter of this year and the fourth quarter of 2020 and has not yet been released.

“Banks are increasingly using AI in their lending business or have plans underway,” Parrish said. “They typically do this to create more personalized experiences for customers and increase the efficiency of their back-office operations.”

At Ally, the software rollout grew out of its work with fintechs, according to Dinesh Chopra, director of strategy and corporate development.

“We are always looking to find best-in-class solutions to provide to our customers, which include both consumers and resellers,” Chopra said. “As part of that, we’re still out there in the fintech ecosystem, evaluating companies.” Ally is part of several accelerator programs and incubates fintechs internally.

A few years ago, Ally bankers met Informed.IQ executives at a tech conference, saw a demo of its technology, tested it for a year in a proof of concept, and recently put the software into production.

“We are one of the largest auto lenders; we’re still trying to find a way to automate routine and manual tasks, Chopra said. “We seek to innovate in all links of our automotive value chain.”

Dinesh Chopra, Chief Strategy Officer, Ally Financial

“We’re always trying to find a way to automate” routine, manual tasks, says Dinesh Chopra, director of strategy and business development at Ally Financial. “We seek to innovate in all links of our automotive value chain.”

Ally uses Informed.IQ software in its contract processing centers. When dealerships submit car loans to Ally, most are approved automatically because loan data is simple and straightforward to verify.

But some loan applications trigger what the lender calls “stipulations.” For example, a loan may require proof of income or residency because the potential borrower has recently changed jobs or moved.

When a stipulation occurs, in the past manual intervention was required to clear the application and proceed to the next step.

The Informed.IQ software consults several databases to verify these data points. Now, far fewer loan documents require manual intervention, Chopra said, though he declined to share numbers.

He sees three main advantages to the software:

  • First, it’s faster. “It’s real-time verification instead of manually selecting each document and verifying things like the potential borrower’s income,” Chopra said.
  • Second, it’s more accurate than humans can be. AI can quickly compare many data sources and detect anomalies.
  • Third, it reduces fraud. For example, Informed.IQ’s software can read all relevant information about loan supporting documents such as pay stubs and it can compare them to all known fraudulent pay stubs that exist on the internet. “If you were to go to Google and search for fake pay stub generators, you would get 400,000 different results,” said Justin Wickett, Founder and CEO of Informed.IQ. “A bank like Ally relies on Informed.IQ’s AI to crawl the dark web, identify known fraudulent patterns that are constantly changing, and check for any similarities.”

Informed.IQ’s machine learning models learn and get smarter over time, he said.

“We are able to use this feedback loop to guide decision-making for predictions about fraudulent pay stubs in the future,” Wickett said.

One of the reasons Ally chose Informed.IQ over other vendors is its ability to integrate with banks’ core systems.

“It doesn’t require us to redo a procedure or change the registration system,” Chopra said.

It also works with bank policies, procedures and data models. “We found this to be a turnkey solution that doesn’t need to be reworked to generate insights,” Chopra said.

Informed.IQ’s software also interacts with humans. When the software cannot confidently verify data from an application, it allows a credit analyst to be part of that workflow.

Chopra also liked the machine learning aspect of the software.

“As more and more data is ingested, it is constantly learning,” Chopra said. “So it can provide checks and balances.”

For example, if a loan applicant claims to have a certain job, Informed.IQ’s model analyzes the person’s salary history and determines if it matches what people are normally paid in that type of job. .

Chopra is said to be open to using the software for other types of loans or digital applications.

“We’re always open to using technologies that enhance the experience of our work with humans and with dealerships, and so far we’ve seen good results,” he said.

Informed.IQ is looking to expand into other types of lending.

“We will definitely be evaluating them for other lending products and even broadly in other areas of finance,” Chopra said.

Wickett sees his company’s software as an audit modernization for financial institutions.

“The aim is to free up staff working in the back office of the financial institution, allowing staff to focus on higher-order functions such as customer service and loan servicing,” he said. -he declares.

Prior to launching Informed.IQ, Wickett spent a decade doing product management at companies like Credit Karma and Lyft.

“It was really at Lyft that I saw drivers struggle in the process of getting financing” for their vehicles, he said. Sometimes loan officers missed deposits on people’s bank statements. There were Native Americans with tribal incomes and hairdressers who worked outside of their salons, getting deposits into their bank through Venmo or Cash App.

“The loan officer was unaware that these were actual income deposits on the individual’s bank statement,” Wickett said. “It was a black box process.”

He and his team created a software product to turn documents into data and data into decisions in accordance with financial institution policies and procedures.

“We set out to democratize credit by enabling the largest financial institutions to give real-time feedback to customers,” Wickett said.

The software analyzes the documents loan applicants upload using different machine learning models. It uses image processing, adjusting skews and rotations, etc. It classifies documents, for example by identifying payslips. It performs entity extraction, extracting key data points from a document and assigning them their appropriate tag (name, company name, date of birth, social security number, etc.).

It reads information from documents and compares it to the app itself to make sure what the person is saying matches the documentation. Because he’s processed more than 20 million documents, Wickett said, he’s learned to recognize relationships, for example, between a pharmacy and the company that bought it.

“Think of Informed.IQ as this giant contributory database where we can cross-reference an application with a multitude of different data points,” Wickett said. “A big part of what we do is to enable seamless funding and straight-through processing.”

Lenders make credit decisions.

“What we do with 99% accuracy in the case of, for example, Ally Financial is calculate the claimant’s income on behalf of Ally Financial in accordance with Ally’s policies,” Wickett said. “Thus, Ally will define when, for example, data elements on a pay stub can and cannot be used as part of an income calculation formula. Informed.IQ automates this.

Informed.IQ started with auto loans because there are 35 million auto loans issued each year in the United States. The company works with five of the nation’s top 10 auto lenders, he said.