How Alternative Data Can Improve Credit Risk Decisioning
“If your data is more accurate, especially on financial metrics, you'll have better underwriting decisions, fewer delinquencies and loss ratios and be able to approve more,” said Charles Zhu, VP of Product at Enigma. Zhu joined Ziv Shabat, VP of Analytics at Noble in a conversation on the key challenges that credit teams face today when underwriting small-and-medium-sized businesses (SMBs). They discussed why accurate data is hard to find, how non-proprietary data can aid accuracy, and more.
Why accurate data is important and why it’s so hard to find
Zhu said having good data on financial metrics was critical for many onboarding steps for credit risk teams, from improving underwriting accuracy to decreasing delinquency. But he added that good data can be hard to find.
“A lot of folks are hesitant to share their data, there's not a lot of data on small businesses. It’s a much sparser world of data as compared to, say, the consumer side of things,” said Zhu.
Zhu pointed to the idea that many credit risk decision-makers are only given three-months of data when a SMB applies. This data doesn’t show that SMB’s change over time and, additionally, could be an inaccurate portrait of an SMB’s performance in other seasons. A fireworks stand, for example, is more likely to have higher revenues in the months surrounding July 4th, rather than in the fall where no seasonal holiday is boosting demand.
Even if you decide you want to onboard a new client, the data permissions and integration process can lead to drop off.
“The second you ask for someone's bank account login or some kind of QuickBooks login – especially for a small business – we see typically a huge drop off in SMB interest in the lending product right there,” said Zhu. “50% to even 90% of folks in an onboarding flow will just drop off when you ask for information like that.”
Zhu also addressed the fact that even after a company is onboarded, outdated data can make portfolio management trickier. You may have had accurate data when the company applied, but when you’ve been working with them for 1 to 2 years during portfolio management and monitoring, revenue and growth may have changed. This makes it difficult for you to adjust credit lines.
How non-proprietary data can aid accuracy
Third party or non-proprietary data, explained Zhu, can serve as an orthogonal signal – i.e. data that shows a different part of the picture than what businesses opt-in to showing you – and complement preexisting internal data sources. This is important, added Shabat, so that you can see beyond your data or a too-narrow common consensus built from a limited set of traditional big data sources or your internal sources.
“If you're built only on proprietary data, you have a sort of a tunnel vision of the world,” said Shabat. “You only see what you see, and you don't know what you don't know…seeking out non-proprietary data really expands that [knowledge].
Shabbat further explained this idea via an example: a customer may have never missed any payments to you, but has missed payments elsewhere. Having data about that business’s behavior elsewhere could serve as a bellwether to that customer’s future behavior with you, or prompt you to have a conversation with them for more confidence.
“It's important to diversify the kind of [data] sources that you're using, balancing between both your information – which is accurate and best fits your business – as well as the external indicators that help you grow your business and keep it healthy from macro economic events that you're not aware of,” said Shabat.
Zhu added that this sort of investment across a wider range of data sources can also help with fraud prevention.
“If you're only using proprietary data or really what the business is self reporting, we're seeing fraudsters get increasingly sophisticated, and showing you that they're very healthy,” added Zhu. “But the next day, all of a sudden, you're out half a million dollar loan.”
Data landscape today
Zhu and Shabat then dove into today’s data landscape – what data is out there for lenders and how can they access it?
Traditional data resources include:
Personal Information from businesses/owners themselves
Credit Bureau data
Small Business Financial Exchange data
Alternative data sources include:
Permissioned data, explained Shabbat, is where you as a lender need a sort of permission or action from your customer to receive that information. Plaid, Netsuite, Codat and Rutter operate on a permissioned data model.
Enigma, however, provides pre-permissioned data with the hope of transforming the lending landscape so healthy businesses can get access to credit. No data provider can currently truly capture total revenues and cash flow in a pre-permissioned way right now, but Enigma can help merchant cash advance lenders with card revenue transaction data that covers nearly every card-accepting business in the country. Before a business is onboarded to a card processor, we can give a better sense of what is an appropriate merchant cash advance (MCA).
“Pre-permissioned data also brings the value of objectivity,” said Shabbat. “You're not asking your client or your customer for that information. Once you ask for permission you are giving some power to your customer to show you what kind of image they want to show.”
If they have several bank accounts, for example, they can connect the one with the most cash or greatest cash flow and hide bad debts.
Challenges implementing new data sources
While alternative data can help lenders expand their SMB universe and answer more questions about current and future clients, lenders may worry about finding ways to add this data into pre-existing workflows to derive value from it.
“It’s critical to find a partner who can provide some kind of service, some kind of data science to help operationalize and implement the data for you.” said Zhu. “Every customer's portfolio looks different, every financial product is different, and you want some kind of ability to customize that data and develop some kind of score that's appropriate for you.”
Additionally, lenders may worry about finding value with non-proprietary data. All the data in the world isn’t useful if it’s overwhelming or too hard to understand. Zhu says the answer to this challenge is transparent data partners.
“Ideally, you understand why the score is working, and the underlying variables that are driving the score,” said Zhu. “Being able to map some of these model scores and model variables onto heuristics and rules becomes really important, because that enables organizations to understand why this is actually working.”
Opportunities in alternative data
While lenders may find challenges in the alternative data space, there are also opportunities. First off, alternative data can be used across your organization including in growth initiatives and marketing campaigns.
“You don't just have to use [pre-permission data] for risks, but you can actually use it for sales and marketing,” said Zhu. “We see a ton of applications like pre-qualifying the healthiest businesses or giving special offers for the healthiest businesses.”
Alternative data can also help you to better define your ideal customer profile (ICP). More information beefs up your knowledge of economic factors outside your immediate team, sector, or organization.
“When targeting high performing industries and verticals, I think a lot of lenders will sometimes look across all kinds of businesses,” said Zhu. “But with pre-permissioned data, you can start targeting the verticals that are performing in the stagflationary economy the best.”
This article is based on a webinar presented by Enigma