Inside Arivo Acceptance’s new risk model

At Arivo Acceptance , measured growth is the name of the game as the lender looks to expand its geographic footprint and increase its access to the capital markets. [caption id="attachment_82092" align="alignright" width="230"] Landon Starr, chief risk officer, Arivo Acceptance [/caption] Driving that growth is data, interpreted by the subprime lender’s risk platform, which is currently undergoing its fourth transformation to include more than 40 machine learning algorithms, Chief Risk Officer Landon Starr told Auto Finance News . "The sheer number of models is one of the backbones the data expansion," Starr said, noting that the vast amounts additional information the lender has been gathering served as a "catalyst to [building] an even more sophisticated platform as we continue to grow as an organization.” Starr sat down with Auto Finance News to discuss the new iteration of the lender’s risk model and how it’s evolved over the lender’s lifetime. What follows is an edited version of the conversation. Auto Finance News: How long has Risk 4.0 been in production? Landon Starr: Roughly speaking, we’ve been observing and starting to build a variety of different elements that will culminate into Risk 4.0 for the last year. When we went live with what we refer to kind of arbitrarily as Risk 3.0 back at the beginning of 2021, we introduced a whole new set of pieces of data that we were capturing that we knew were going to becomes assets for the Risk 4.0 model. We have a whole host of new data sources that we’ve been tapping into and have been gathering that data over the course of the last year, sourcing as many different sources of information to help us build out and differentiate even more powerful algorithms. That’s all then the foundation. The actual fitment and development of the model we move pretty quickly on once we have the data that we’re going to be applying and fitting and evolving. AFN: How often does the risk model get updated? LS: We have a few different versions of how we update the model. If we're revisiting the entire construct of a piece of it, that's where we ultimately will facilitate kind of a new version of redevelopment will catalyze a risk 3.0 to risk 4.0. But we are always doing model updates or parameter tuning and additional tweaks to evolve those algorithms as new data feeds in, but in more of a minor update, tweak, refine and polish type of manner, rather than a full reassessment of the construct and introduction of brand-new algorithms that will help facilitate with some of the ensembled interactions. AFN: Is there anything in particular that’s going on in the market right now that you could point to as significant drivers in the choice to roll out a new iteration of your model? LS: There’s a lot of interesting things in the market and a lot of new behaviors that we're certainly looking at and trying to capture and make sure that we're considering as it relates to risk assessment. For example, there's been a massive behavioral shift in consumers and how they interact with their money, how they spend their disposable income. That in and of itself has been massive. And then layer on the reporting to the credit bureaus where there were a lot of extensions, and a lot of government stimulus tax factors that are coming into play — there’s a lot to take into consideration. If you don’t qualify your model on some of those dynamics and find ways to contextualize the algorithms from how the data coming in has been different versus some of the prior years, you can get yourself into a situation where you’re dramatically under predicting default risk just on the premise that a lot of things that happened over the last couple of years aren’t going to remain in place in the next couple of years. There are a lot of external factors that impacted consumers from a behavioral standpoint and from a financial and monetary standpoint that are important to try and contextualize and take into account. That couldn’t be more real over the course of the last year. We haven’t seen any of those vehicle value inflationary pressures subside — we’ve seen them do nothing but expand and intensify over the last year. Supply constraints didn’t magically solve themselves, and we still have massively inflated vehicle values relative to where we were beforehand. Government stimulus was a big part of it. We saw a very significant amount of that coming into play in both 2020 and 2021, and then there’s the lender responses, like I mentioned, and the reporting to the credit bureaus. Its just a multitude of different things that are layered on to of each other that really provides a lot of different unique risk challenges to risk modeling for 2022. AFN: What sort of human capital or manpower does it take to launch a new iteration of your risk model? LS: It's honestly the whole organization. When you think of all the different pressures that we're experiencing, the parts of our organization that are ultimately influencing the new risk framework are very widescale. For example, we have to understand the challenges coming from the collections department. I have a number of different strategic conversations with what's been happening on the collections front, asking the team what are the subjective elements of what we hear from our customers? And how do those ultimately affect their willingness to pay and, and their ability to pay? On the origination side, we talk about some of the contextualized frameworks of dealerships’ supply constraints and what vehicles they have on the lot, and how that translates into the kind of applications that we're going to see as a result of that. So, as part of how we define and capture the data that becomes part of Risk 4.0, it’s absolutely been a massive team effort. We’re very technology driven, we’re very risk driven as an organization, and having stood up our own loan origination system and building our own software, the kind of data we capture allows us to really tap into individual intelligence and insights and knowledge of employees at every operational touch point with customers both internally and externally to really facilitate the right kind of data capture that will facilitate the best predictions we can get. AFN: You mentioned that you have 40 algorithms in Risk 4.0, and you had maybe four in 2018 in the first iteration of Risk. About how many algorithms were in Risk 3.0, and who writes those algorithms? LS: For Risk 3.0, we were sitting mid-20s, I want to say 27 or 28 machine learning algorithms that were live in production when we rolled it out. I am the one that primarily is fitting those algorithms and building them. And the interaction and assessing and kind of monitoring and things to that effect, I’m actually individually doing a lot of that myself. So, building those models, capturing the data, gathering the data, having the context of what our historical data means — and being able to qualify the data in that kind of way — there’s a lot of tribal knowledge that is important to make sure the data is properly qualified. Coding and understanding how to translate the algorithms and the interaction between them and assess the abstract complexities of how these things ultimately culminate to decisions for a business has been my passion for a lot of years and is something that I still get really nerdy about and excited about to this day. Auto Finance Risk Summit, the premier event for risk and compliance management in auto finance, returns April 25-26 in San Diego. This event is the central resource for education and networking focused on best practices in risk management and will prepare your team for the dynamics of the evolving auto finance market. To learn more about the 2022 event and register, visit www.AutoFinanceRiskSummit.com .

Inside Arivo Acceptance’s new risk model

Risk 4 should go live in about a month, CRO Landon Starr said

Stellantis unveils the all-new Alfa Romeo Tonale

Rental car companies shake up used-vehicle market as demand resumes

Related Posts

Chase Auto head of private label captive finance on growth, competition

CAC former CEO Ken Booth says tech, disciplined approach are key

Volkswagen Financial CEO prioritizes leasing, EV strategy, Wells Fargo transition in 2026

Chase Auto CEO talks OEM relationships, AI investment

Rental car companies shake up used-vehicle market as demand resumes

Stay Informed with Our Newsletters

The Roadmap Podcast

SPONSORED

Why credit unions give dealers an edge in today’s auto market

Driving better decision-making across auto finance operations with SAS

Auto finance’s first line of defense: Raising the standard in integrated software partnerships and data strategy

Retrieve your password