Lendbuzz blends its founders’ early credit experiences with artificial intelligence (AI) to disrupt traditional assessment methods and widen the pool of credit-worthy individuals. And with the founding team’s deep understanding of technology, they capitalized on the right technologies at the right time.
Co-founder and CEO Amitay Kalmar said the seeds for Lendbuzz were planted during his and co-founder Dan Raviv’s early experiences in the United States. Despite promising educational paths at top universities and good account balances, they struggled to get auto loans and basic credit cards. Family and friends shared similar stories.
The pair understood alternative data, machine learning and deep neural networks well. They knew they could be used to assess creditworthiness better and help underserved groups.
Kalmar and Raviv began by focusing on software solutions for auto loans, prioritizing folks with no credit file. But they soon saw that their system was better at scoring applicants in many areas. That was good news to the 45 million Americans who either had no credit file or one that was too thin to be effectively scored by FICO.
“Within these tens of millions of people, obviously, they’re not all bad consumers,” Kalmar said. “They’re not all not creditworthy. It’s just there is no system to evaluate them.”
Lendbuzz deployed the right technology at the right time
Timing is a crucial aspect of Lendbuzz’s story. Digital systems and cloud technology have matured enough to be useful for fintechs. Digital data and APIs abound. Storage and computational capacity have grown. Data models can now run thousands of data points. Those benefits didn’t exist a decade ago.
“You have the tools to analyze thousands of data points on an individual and run tens of millions of models to improve your modelling,” Kalmar said. “So why not use it? Why still be stuck with a credit file that has minimal data sets and serves a very specific consumer but excludes many others?”
Around 2015, Kalmar and Raviv began analyzing Lending Club loans and saw arbitrage opportunities. But to maximize those openings, they needed unique data sets. From their research, Kalmar and Raviv saw that augmented data enhanced credit assessments.
That was where they began to view the problem differently by looking at incremental data enhanced by machine learning and deep neural network algorithms through a localized system. That allows systems to consider each consumer as a vector that is better understood as more signals are collected.
Embracing alternative data and why the US lags behind
Lendbuzz technology is partially shaped by Raviv’s experiences studying how a Mexican bank considered alternative data sources. Kalmar said there is much to learn from how much of the world crunches alternative data.
The United States is late to that party because it is among the best in collecting credit bureau data. Few countries have such an evolved system.
“FICO, at the end of the day, works really well,” Kalmar said. “If you had a 750 FICO and a deep file, you’re unlikely to default. If you have a 450 FICO and a few delinquencies, you’ll likely default, creating a good separation. The fact that the data is so good and deep here, and there are three credit bureaus, and every payment is collected, is stopping innovation in underwriting.
“So if you’re in an emerging country, Asia, Latin America, or Africa, you don’t have that system. Whether you’re a small or large institution, you want to do better in your underwriting; you need to innovate and do something different. You need to use a different data set.”
Disruption is harder as you grow
It’s also much easier to innovate from the beginning than after you’re established. Kalmar sees that as Lendbuzz matures and appreciates how hard it is at a legacy institution. Some changes may also cost more than they benefit the company.
“Disruption usually comes from new entrants to a space,” Kalmar said. “I don’t think it’s very different than consumer credit. There are times when disruption comes from incumbents, and it’s very admirable because it’s much harder to do.”
How Lendbuzz meets a triangle of needs
Kalmar said Lendbuzz’s system is unique because it has three stakeholders: borrowers, lenders and dealers. Their system must deliver value to more sides. The software focuses on providing dealers with a streamlined loan application that simplifies origination. The machine learning-based system crunches 18 months of bank records within seconds of connecting an applicant’s bank information. That provides crucial insights on income, spending patterns, education, employment, and even vehicle usage.
“The thousands of features we collect on each individual go into deep neural networks, and then we use a localized system to score the individual,” Kalmar explained. “That creates a proprietary risk score, which we use to make a credit decision on whether this individual fits our credit profile or doesn’t.
“Every little thing is a feature. Then you have that profile of an individual, and now you only care about if they make a payment or not. Every month, you collect new signals on that individual. You compare that new profile to the most similar 500 other consumers that we originated in the last three years.”
Validation, thy name is a $219M AAA securitization
The proof of Lendbuzz’s success is in a very lucrative pudding. Earlier this month, Lendbuzz completed a $219 million securitization collateralized by a pool of auto loans. The transaction was rated AAA. Three years ago, the company’s first securitization was A-. That’s some validation.
“Our performance over time showed them that we significantly outperformed the market in a very challenging cycle of consumer credit,” Kalmar said. “The past 24 months have been a challenging cycle, and consumer credit delinquencies have shot through the roof, especially for non-prime consumers.
“We’ve been very consistent in our performance, performing better than the credit agencies-based models, whereas most traditional lenders have performed much worse than the base models. So over time, we were able to increase our credit rating.”
AI’s disruptive potential is generational
Lendbuzz isn’t sitting on its laurels. Kalmar said he’s watching generative AI, which he believes could have as much impact as the Internet, though it could take up to two decades for it to be fulfilled. Early adopters will benefit the most, though that comes with a caveat.
Kalmar said it’s possible to act too early. Key infrastructure pieces are not in place. Bet on the wrong pipes, and you miss the boat. Closely watch developments, have a team in place and be ready to pounce.
“We’re not going to build everything in generative AI,” he said. “We need to build on infrastructure like we’re building on Amazon cloud infrastructure. The fact that we can use it enables us to do things we couldn’t have done 10 years ago.
“The same thing with generative AI. So we’re looking deep into the infrastructure there. What do we want to base our solutions on? All of our time and new R&D will be focused on these areas. Big opportunity.”
Also read:
