Master the automation of document verifications

Master the automation of document verifications Informed

Over the years, there’s been a desire to automate underwriting – the ability to return credit decisions in seconds. Many Loan Origination Systems (LOS) providers have invested in capabilities to help their auto lending clients achieve automated decisioning. These include: pulling 3rd-party data, consumer scoring based on lender-specific credit models, and appropriate deal pricing. The challenge was around automating a re-structured deal, when the dealer requested a vehicle that doesn’t fit the lender’s credit risk policies. However, a handful of lenders and LOS providers have found innovative ways to solve this challenge. (Note – this is usually prevalent in indirect subprime auto lending)

However, one thing that most lenders have not been able to solve for is automated funding to instantly onboard loans from origination systems to servicing systems.

When auto lenders approve the loan application, they ask for a set of documents (referred to as stipulations). These stipulations range from consumer documents (identity, residence, income, employment) to dealer documents (retail installment contract, gap contract). In indirect auto financing the dealer obtains the consumer’s signature, electronically or manually, and sends the documents to the lender. The lender must verify these documents before they can fund the dealer for the vehicle. 

Before we dive into automated verifications, let’s take a look at the process. Lenders verify documents for 2 reasons:

  1. Verifying for compliance (missing signatures, name mismatch, misrepresentation of information etc.)
  2. Verifying for Fraud (intentionally forging income documents, synthetic or stolen identities etc.)

The line between verifying for compliance and fraud is blurry at times. When loan officers (verification agents/funders) verify a document, they focus on all aspects of the document. If they find inconsistencies in the data or if a signature is missing, they route it to the dealer or appropriate team to fix. Or, if they find the document is fabricated, they route it to the fraud/compliance department for further review. The probability of fraud is higher in direct lending as the buyer must be physically present at an indirect lender, 

Most lenders have traditionally relied on loan officers physically verifying the documents for compliance and fraud. But with advancements in technology, that is changing. With the latest improvements in AI including Large Language Models (LLMs) and LayoutLM Models, we can extract the data on the documents (OCR) and apply AI models on top of it to cleanse the data. Systematically extracting data turns documents into digital data that can used to run verification rules to check for compliance and consistency. 

Traditionally, the LOS houses the rules, calculations and funding policies and can write simpler rules that compare strings. But the art of verifications needs more sophistication. 

What You Need to Know

Here are a few things to consider as laid out in a previous blog:

  1. Name match – do you have the ability to understand Mike is the same as Michael and it can also be spelled as Micheal. 
  2. Employer name match – can your system identify that Envoy Air is a subsidiary of American Airlines. Can you use Bing search engine, powered by OpenAI or your own Microsoft Azure OpenAI models to power that search? 
  3. Dealer name match – Dealers often use their “doing business as” (dba) name on the ancillary product contracts. Can you invoke Google’s places API to understand if both dealer businesses have the same address?
  4. Missing ancillary product contract – can your system look at RISC and see that the dealer has a line item for Gap but the deal jacket is missing a Gap contract?
  5. Complicated income calculations – every lender calculates income their own way and there are subtle differences between lenders. Can your system meet your specific needs around paystubs and bank statements? Even after extracting data from bank statements, it’s hard to understand which entries are ATM deposits vs. salary. Do you have the sophistication to make those calculations?
  6. Flag for Fraud – can your system identify potentially fraudulent documents based on template matching, inconsistent fonts, forged calculations that don’t add up?
  7. Check for compliance – do you know if the TILA information on the Retail Installment Sales Contract checked out? Are your Amount Financed and APR calculations balanced? Are the fees within state limits?
  8. Check for document completeness – Dealers often scan documents with signatures or data but leave out the T&Cs. When there is litigation on the contract, the legal team wants to read through all the pages to understand the legal terms. Can your system identify incomplete docs?

Automating verifications to instantly board loans involves accurately extracting information from the documents and intelligently applying logic to verify all pertinent information. To reduce operational costs by automating verifications, consider a technology partner that can do document data extractions and intelligent verifications. 

author avatar
Kartheek Veeravalli Head of Product
Kartheek Veeravalli is the Head of Product, Auto. He has more than 15 years experience in building successful fintech products in auto and consumer lending at defi SOLUTIONS, Cox Automotive and FICO.

New: American Banker - Sharing information = best defense against AI fraud