Data Science & AI

3 Tips to Consider Before Using Alternative Data in Underwriting

Jay Budzik

March 13, 2018

More data is available for use in lending than ever before. In recent years, alternative data providers have proliferated. But it can be daunting to determine which data sources are worth evaluating. Here are some tips to consider before you begin your search for credit data providers.

Lenders often turn to credit data providers with the intention of reaching new market segments, for example, borrowers with thin or no credit bureau files. These applicants have historically been very hard to underwrite. Here are a few tips to help you identify the data sources that have the highest potential to help you reach new customers and market segments.

Many early adopters of big data in credit realized the biggest gains by applying machine learning techniques to their existing bureau and internal data. For example, a major U.S. credit card issuer increased its loan approval rate by 10% without any increase in risk simply through the application of better math to the data it was already using for underwriting. Before doing the hard work of evaluating data vendors, ask yourself whether you are using the data you already have to its full potential.

Understanding credit data providers in underwriting.

Tip 1. Are you fully leveraging the data you already have?

The best credit underwriters are constantly evaluating data providers but it’s important to remember that you probably have your own internal data assets too. And the proprietary data you already have can often provide more value than third party data sources, especially when applied in a machine learning model.
alternative data stacked up:

Case Study

A large auto lender used Zest’s Automated Machine Learning (ZAML) software to develop a machine learning credit model. The ZAML model reduced losses over 20%, saving this auto lender tens of millions of dollars annually. ZAML also empowered the auto lender to explore the value of alternative data in underwriting. ZAML’s explainability tools determined that the the alternative data examined did not contribute much to the model’s performance. This was especially evident when compared with the gains driven by using data the customer already had but wasn’t using due to the inherent limitations of traditional credit risk modeling methods.

Here’s a chart that shows how the alternative data stacked up:

Contribution to ZAML model

It is important to understand whether the data provider has proprietary data or if it’s a reseller. For example, you might find that one vendor generates a score using Data Provider A’s data. Another vendor resells some of Data Provider B’s data. It is recommended that you source data from the original data provider, so the data isn’t subject to translation errors that may make data processing difficult.

Tip 2. Identify overlaps.

Early on when evaluating a data provider, it is important to determine whether the data is redundant with existing sources. Usually you can uncover this by asking the vendor where their data comes from and how they calculate their scores. For example, many vendors carry small dollar loan data (bankruptcy, lien judgment data, consumer’s identity, etc.) derived from the same public sources. Beware that, the pricing of similar data from each vendors may be very different.

Tip 3. Identify resellers.

It can also be less expensive to buy from the original source, though this isn’t always the case. By interviewing the data provider, it is possible to learn about a vendor’s partnerships and reseller agreements as well as to compare pricing.

Zest’s clients, which include lenders from all over the world, have found a handful of alternative data sources to be extremely useful in underwriting thin and no-file borrowers. The incremental cost of alternative data from these sources could be justified by the increased profitability of the loans issued by machine learning models trained on that data.
Ultimately, the use of alternative data will depend on the goals you are trying to achieve. There certainly are times when alternative data is useful, however, many lenders find substantial increases in credit performance by simply leveraging machine learning to make better use of existing bureau data and other unused data already in-house.

Deciding whether to use alternative data.

ZestFinance can help you determine the value of alternative data and machine learning for your credit business.

Thank you for subscribing!
Something went wrong while submitting the form.