Mortgages are important to both consumers and financial institutions — they’re often the largest loans that people take out in their lives and they account for the majority of outstanding consumer debt in the US. Unfortunately, the mortgage lending process is very complicated — it involves multiple parties and many separate documents, and it’s governed by a broad set of regulations. As a result, mortgage lending can be extremely tedious, time consuming and frustrating for both consumers and financial institutions. However, advanced technologies such as machine learning and predictive analytics are changing this by streamlining workflows, reducing required human intervention and automating critical decisions.
This post discusses a few critical ways machine learning is transforming mortgage lending to benefit both consumers and lenders. If you’re unfamiliar with machine learning, take a look at this post for an introduction.
1. Improved Application Verification
A key responsibility of mortgage lenders is to verify and confirm information about both the applicant and the collateral property. A good example is applicant income verification, which lenders complete to help ensure they’re lending to consumers that can afford to repay their mortgage, both to reduce defaults and ensure compliance with “ability to pay” regulations. Income verification often involves collecting supporting evidence such as employment information, proof of stated income (such as pay stubs) and income estimates from third parties. This takes time and often results in imperfect information and a judgmental human decision.
Machine learning can greatly help with verification to the benefit of both lenders and applicants. For example, lenders can use historical data to train predictive models that accurately estimate the applicant’s income, based on a wide variety of potential factors. This estimate can then be compared to the applicant’s stated income level to help confirm that it was reasonable and accurate. This type of automated assessment can accelerate income verification while also flagging high-risk cases for manual review, and machine learning can be used as either a decision-making tool or a decision-support tool (depending on the specific situation).
2. Application Queue Prioritization
For many lenders, having the assistance of effective machine learning models is almost like having more employees. This is because such decision-making systems can often amplify the productivity of workers by better directing their efforts.
The mortgage lending process involves back and forth between the lender and the borrower, and loan officers need to contact applicants frequently to communicate next steps and move the process forward. Historically, this has involved manual processing applications one-by-one in a simple queue, which can result in sub-optimal allocation of the loan officer’s time.
As lenders gather data on how consumers move through the process, respond to reach outs, and take actions, they can train machine learning models that optimize the loan officer’s workflow and recommend their “next best action”. This lets loan officers work through their queues dynamically (based on the opportunities that are most in need of attention), reduces the time loan officers spend deciding what to do next and creates more consistent processing across the organization.
3. Better Document Validation
Mortgage lending is document intensive and many documents are required to execute a loan. One important job of lenders is to validate that information matches from form to form.
Loan officers have historically had to do this work themselves, but now machine learning models can be trained to do it for them. Models can automatically identify text within documents and compare it to ensure that the forms have been filled out correctly, letting the loan officers focus on more important tasks. When issues are found, the exception is flagged for human review and a signal is sent to the loan officer to manually resolve the problem.
4. Underwriting & Loan Pricing
Mortgage underwriting involves assessing the credit risk of an applicant and whether they meet the standards established by the lender and others such Fannie Mae and Freddie Mac. Traditional mortgage underwriting has been rules-based, where application information — usually in the form of metrics such as loan-to-value — is run through a series of pass-or-fail rules that determine loan eligibility. Increasingly, model-based approaches are being used, wherein applications are scored by predictive models that determine eligibility. While some types of mortgages, such as those insured by the federal government, must follow certain specific rules and have limited underwriting flexibility, other mortgage products afford lenders more leeway. In these cases, lenders are increasingly looking to machine learning as an effective way to create predictive models for assessing credit risk.
Machine learning is most effective when many data points inform a given decision, and mortgage lenders generally have a large amount of data on the applicant at the time of underwriting. This data typically comes from a credit report on the consumer from a major credit bureau, such as Transunion, Experian or Equifax, as well as property valuation information from a range of real estate data providers. Machine learning can effectively discover relationships within this information that can more accurately assess credit risk than traditional rules-based underwriting.
Machine learning is a significant development that will continue to have increasing applications within the mortgage lending industry. With machine learning, mortgage lenders and underwriters are going to be more efficient, more profitable, and capable of delivering a much better customer experience.
Originally published at digifi.io on September 21, 2018.