Loan automation platform Upstart on Monday announced the launch of its Small Dollar loan, as well as a new product for small- and medium-sized business (SMB) lending.

Small Dollar loan is for personal, payroll-based loans of around $1,000 or less. That offering was launched at the end of first quarter, Dave Girouard, Upstart Chief Executive, said during Monday’s earnings call.

Girouard quickly pivoted to discuss the AI-powered platform’s new SBM offering. “Business lending is central to far more banks than is consumer lending,” he said on the call.

Despite that, FDIC data suggests 77% of large banks and almost 90% of small banks do not have an online application process for business loans, Girouard said.

“The first version of our SMB pricing model will include more than 500 variables about both the applicant and the business,” he said.

Model tests suggest that the SMB model “will deliver higher accuracy, as measured by area under the curve, or AUC, than peer models that have been in the market for years,” Girouard added. Plans are to begin testing the new product in the second half of this year.

The San Mateo, Calif.-based vendor revealed its bank and credit union partners originated 465,537 auto and personal loans in Q1 on Upstart’s platform, totaling $4.5 billion — an increase of 174% year over year, according to the company’s earnings release. Upstart reported first-quarter revenue of $310 million, up 156% YoY.

In addition to its growth in loans originated, the fintech added 500 dealerships as well as 57 banks and credit unions, which is up from the 42 announced during February’s earnings call.

“At this point, we’re adding about a lender per week. This is real progress, considering we had just 10 lenders on the platform when Upstart IPO-ed in December 2020,” Girouard said. “Additionally, we now have 11 lenders with no minimum FICO score in their credit policies, up from seven the last time we spoke.”

Revamping infrastructure to better support AI

Upstart leverages artificial intelligence models in its platform and is working on “fundamental upgrades” to the underlying infrastructure, Girouard said. Two factors contributed to the need for the upgrade.

First, Upstart is working on seven or eight unique models that focus on different aspects of credit targeting and origination, and those models work across five different credit products, he said.

Second, the amount and types of data used to train its models has grown exponentially, so the time and processing power required to retrain the models has also increased and will continue to do so, Girouard added.

This infrastructure revamp effort, internally called Machine Learning to Heaven (ML2H), will help improve the way Upstart builds, trains and deploys AI models, he said.

“Our goals with ML2H are to allow hundreds of research scientists to seamlessly and securely build new models and add data to existing models, train and test them in an automated fashion, and deploy them across the entire model ecosystem simultaneously,” Girouard said. “The system we’re working toward will provide maximum leverage to our research scientists, productizing and automating how new models are trained, tested and deployed.”

The fintech will also be incorporating more transparency about macro adjustments into its models with support for partners who want to add their own macro assumptions into the product, he said.

“This will provide significantly more transparency to our lending partners and will also put our focus squarely on risk ranking, which is the heart of what makes Upstart’s models unique,” Girouard said.

AI-based auto refinance model

Upstart is also using its own auto lending performance data to power its auto refinance model as the company continues to attract auto lenders and dealerships, contributing to a record in loan transactions in Q1.

“We continue to make rapid progress with our auto refinance product. In the first quarter, we transacted more than 11,000 auto refi loans on our platform, almost twice as many as we did in all of 2021,” Girouard said during the call. “We also launched our first AI model for auto refi that is partially trained by our own auto lending performance data.”

Upstart’s auto refinance platform historically pulled from a combination of the company’s personal unsecured loan data and third-party data, a spokesperson told our sister publication, Auto Finance News. Upstart now has enough auto loan repayment data to power the AI-based refinance model.

The fintech also expanded its dealership footprint to 525 locations as of the first quarter, up from 410 during the previous quarter and 162 compared with same period a year ago, according to the earnings presentation. Dealership partners span multiple OEMs, including Toyota, Subaru and VW, Girouard said.

The additions follow a year of growth in auto finance on Upstart’s platform in 2021, and a rebranding of its auto loan product to Upstart Auto Retail following the acquisition of Prodigy Software in the second quarter.

Upstart also is accelerating the launch of Upstart Auto Retail following success at a few dealerships in California, Girouard said. “Our auto teams are working quickly to smooth some of the product edges, filling in a few missing features and completing integrations with various legacy dealer systems, all in the interest of moving toward a broad-based rollout,” he said, noting a nationwide rollout is planned for Q3 with plans to bring lenders onto the product in early 2023.

Upstart funds auto refi loans from its own balance sheet, but eventually will funnel the loans to its bank partners and investors, freeing up capital for the auto retail product, Chief Financial Officer Sanjay Datta said on the call.

[stock_market_widget type=”inline” template=”generic” assets=”UPST” markup=”Shares of {name} ({symbol}) are trading at {price} ({change_pct}) as of {last_update}.” api=”yf”]

Editor’s note: Auto Finance News is a sister site of Bank Automation News.

Help shape our agenda for the Bank Automation Summit by applying to join the speaker roster here. Potential speakers will be contacted and confirmed directly by the editorial team, and only qualified submissions will receive a response.

Learn more about Bank Automation Summit Fall 2022.