In auto finance, the concept of risk management boils down to creating profitable ― not perfect ― predictions about whether consumers will repay their car loans. With data availability increasing at a rapid pace, and advancements in decisioning and pricing algorithms constantly being developed, lenders must prepare for a highly analytical decade in 2020, said Preston Cecil, vice president of risk management at Innovate Auto Finance, at the Auto Finance Risk & Compliance Summit last month.
Human capital, technology, and data are the trio of risk-management tools that lenders must employ to hone scoring mechanisms, Cecil said. Here are some specific ideas to implement within each area:
Invest in human capital.
- Offer training courses so that employees can bolster their knowledge. Courses may be company-funded or based on available openware.
- Encourage creativity by challenging the status quo and setting aside time for brainstorming.
- Pay for performance. Define and track key performance indicators, then reward employees appropriately.
Allow technology to aid workflow.
- Rely on automation. “Let the machines do the busy work,” Cecil said. Their speed and accuracy allow for deeper dives into non-quantifiable data.
- Employ the appropriate statistical software, and tools that integrate easily into your loan-origination system.
- When it comes to data architecture, use an adaptive and robust infrastructure.
- Develop the ability to quantify unstructured data ― the information that doesn’t reside in a traditional row-column database ― like email messages, call center notes, and survey responses.
Harness data, don’t abuse it.
- Tap into non-bureau data to enhance tailored scorecards, particularly for thin-file applicants.
- Identify leading indicators, which are ever-changing. Historically, consumers paid for groceries first, then for utilities, rent, and cars, Cecil said. Nowadays, phone bills commonly trump utilities, and car payments trump rent.
- Aim for meta-analysis. Use statistical methods to contrast and combine results from different data sets in the hope of identifying patterns or other interesting relationships.
