Lenders know the importance of scoring models for those with thin or no credit files, but the problem persists, Credolab’s CSO Michele Tucci said.
Credolab tackles the problem using privacy-consented and permissioned data through smartphones and web pages.
Their proprietary technology is embedded with clients so anonymized data can be accessed during application.
The Credolab recipe: Focus on the ‘bads’
Tucci explained that user behavior reveals valuable insights into an applicant’s ability and desire to repay (more on this later). How do they interact with the user interface? Do they frequently retype their income levels on applications?
Credolab combines that information with the design philosophy of identifying the probability that an applicant misses the first payment. That is crucial because it saves clients the costs of credit and verification checks.
They do not process personal data, only permissioned, anonymized metadata sets such as the frequency of added contacts or if someone has a personalized ringtone. And they do not use AI.
“We don’t do any AI,” Tucci said. “We can easily explain why a customer should be rejected based on net logistic regression. And we train our models by using repayment data of the lenders.”
Our behavior reveals so much valuable information that is valuable to lenders. What percentage of selfies do we take? How many apps do we download? Did an applicant download several finance apps in the weeks before seeking a loan?
Also, read:
How Credolab removed onboarding friction
Credolab’s technology is embedded in the onboarding process. It is designed so that no friction is introduced, Tucci said. He contrasts that with some popular banking apps requiring users to connect their bank accounts during onboarding. Credolab leverages the permissions of the app integrating its technology. Applicants will grant access because they want to apply.
“…the end users, if they want to improve their chances of being approved, they need to connect their bank account with the application,” Tucci said. “The moment they do that, they need to abandon that application… They need to remember the login credentials entered into the bank account and then grant access to the lender.
“Now, that adds friction. It’s not about cost per se, but friction to the onboarding journey, which reduces conversions.”
Tucci added that the process is completed in real-time, with total privacy compliance. It complements the lender’s model, improving it with behavioral data that correlates poorly with transactional, credit bureau, and telco data. That widens the net.
The pros and cons of ‘alternative’ data
The United States differs in its interpretation of alternative data. The country takes a more narrow approach, Tucci said. Contrast that with emerging markets where they use telco, mobile money, and the frequency and timing of topping up prepaid cards.
Tucci questions the validity of some data sources that others are lauding for their predictive capability. Netflix and Spotify payment histories?
“Because you pay $10, $11 a month, what does it tell me about your capacity?” he asked. “To repay tells me that you spend a lot of time on Netflix; it doesn’t tell me that you get you can repay a $1,000 line of credit.”
The single valuable alternative data point is bank account data, Tucci said. Look for patterns such as paycheck history. Are there any changes in patterns? That’s the goldmine.
Assessing both the desire and ability to pay
For Credolab, using alternative and biometric data isn’t new – they’ve been working with those sources since 2016. Tucci said it brings a behavioral component to onboarding by assessing an applicant’s willingness to pay. Combine it with credit bureau data on past behavior, and you have greater predictability.
Assessing an applicant’s willingness to pay differs from many scoring models that evaluate their capacity to pay. Tucci cautioned that just because someone has more money doesn’t mean they’ll use it to pay you. Credolab begins by training its models on identifying the bad bets instead of the good bets. If someone doesn’t fit the “good” model, that may be because they have a thin file for many reasons. Perhaps they are in a career that brings significant pay bumps.
“Observing the ‘goods’ has no value for anybody because they are good and will pay,” Tucci said. “The only question then becomes how much money they can afford to repay, and then we go back to credit history income and the assignment of a line of credit.
“The other consideration is that if a customer is bad, they will deny access to data. That information in itself is already valuable.”
Expanding reach through partnerships with Provenir, TransUnion, and credit card firms
Credolab has expanded its reach through several partnerships. The company recently announced a union with Provenir, which provides AI-powered data and decisioning software. Credolab’s CredoSDK is available on Provenir’s Data Marketplace, where digital lenders, neobanks, and BNPL firms can access it. The marketplace provides organizations with open banking, KYC/KYB, fraud, credit risk, verifications, social media, collections, and affordability solutions.
TransUnion is another Credolab partner. Credolab’s technology is embedded in a consumer lending product sold to banks and lenders. It benefits Transunion because they can score 100% of applicants. While their reach is strong in the US, it may only be 40% elsewhere.
It adds to the capability of other alternative data interpretation methods lenders have in place. Perhaps they have access to data from the top telco services provider. They won’t have data for everyone, even though those others use mobile phones. With Credolab using mobile user metadata, the coverage expands. Tucci said that helps with inclusion in underserved areas like LATAM, where less than half of adults have access to financial services.
Credolab is helping a major credit card provider expand its services to SMEs in Mexico. That firm wants to increase the number of cards in circulation. To do that, they must serve the bottom of the pyramid. Credolab is helping them build an alternative scoring model.
SMEs as growth opportunities
Tucci said serving SMEs is a growth opportunity, especially post-pandemic, as many started their businesses. Like its other efforts, Credolab takes a different approach here too.
“We assess the business owner, and if the business owner is delinquent, then the business will be delinquent,” Tucci explained. “So the business behaves the same way as the owner does.”
Improving service with behavioral psychology
Credolab is improving its behavioral analytic capacity by working with a behavioral psychologist. The psychologist mapped one million metadata sets against the OCEAN personality traits (openness, conscientiousness, extroversion, agreeableness, and neuroticism). Together, they designed a framework to map how people use their smartphones against their personality types. The necessary information is available in the device’s digital footprint.
“The idea is that now with the same sort of data and the same technology, we can also help marketing teams to improve their marketing campaigns with personality-targeted messaging,” Tucci said. “This can improve conversions by 50%.”
The future is bright
These developments have Credolab excited about improving financial inclusivity around the world. It’s coming at the ideal time, as many banks seek to expand their client bases.
Tucci conceded there are some challenges, such as adapting models to serve the good bets who work in the gig economy. More work can also be done in the developed world, where an estimated 54 million Americans and 5.5 million Brits are credit invisible.
“It’s not surprising that they cannot include more people because they keep using obsolete models to assess people new to banks,” Tucci said.
To address financial inclusion, lenders must meet people in their natural habitat, Tucci concluded.
“It’s also a priority for them because they want to assess how people behave in today’s world. And what do they do in today’s world? They use their smartphones.”
