What Smart, Conversational AI Adds To Credit Scoring

Traditional credit-scoring models have not had it easy in the court of public opinion over the last few years. The complaints vary in their specifics, but all revolve around a basic premise: The old credit-scoring models are too backward-looking in a world where real-time data is available — and they are insufficient to the task of properly assessing risk.

It’s a proposition that can be — and has been — heavily debated, but no matter which side of the dispute one comes down on, it is clear that the world of credit scoring and analysis is changing. Traditional models, like the FICO, are undergoing major revisions, now factoring information like real-time cash flow when analyzing a customer’s score.

Those changes are happening because they must. Traditional credit-scoring models are just drawing criticism, as well as a generation of competitors like Aire, which offer alternative credit-scoring models advertised as better-able to offer a “three-dimensional” view of customers in real time.

Aire Founder and CEO Aneesh Varma first encountered this friction as a consumer seeking credit 13 or 14 years ago, finding he was unable to secure it despite being an up-and-coming, young professional with a solid salary and strong prospects. The problem was he had no credit history.

“How does a new borrower bypass the catch-22 problem of credit where it takes a while to get a history, but you need credit to start a history?” Varma asked. In the early 2000s, he found there was no good answer to that question. Thin-file credit applicants could go through a fairly painstaking process to build credit with low-value cards, but that wasn’t a good answer for him.

Since Varma was sure he was a good credit risk, it occurred to him that there might be others who experienced the same thing, which gave him the idea to build a better one: The idea of Aire was born — though, not the business itself. The actual business, Varma noted, was the product of a long research period, since building a better credit-scoring experience is complicated.

As far as traditional scoring methods go, they are an effective backward look at elements of a customer’s spending and financial life. The trouble, he said, particularly for thin-file applicants, is that this look doesn’t offer enough information to make a good decision. Aire, though, is a credit-assessment platform intended to fill in that extra data.

To date, Aire has aided in the underwriting of $10 billion of credit across various consumer credit categories. In addition, the company claims it has helped firms increase their credit approvals by as much as 19 percent without significantly changing their risk or default numbers. Furthermore, the firm recently raised $11 million in Series B funding, and has raised a total of $20 million.

“Our main product today steps in to engage with an applicant on a lender’s website when the existing decision engine is unable to reach a full decision,” he explained. “We enable the consumer to supply relevant financial data to us about their circumstances. This is beyond just transactional banking data, and, therefore, gives us a full picture … looking forward, not just the historical snapshot.”

The customer supplies that information through what Aire calls a “digital interview” via a chat interface, wherein the applicant is asked a series of tailored questions to establish a fuller view of their financial situation, lifestyle and profession. Those answers are then fed into Aire’s artificial intelligence (AI) on the back end for analysis and scoring.

The AI, Varma said, learns over time — as more performance data on outstanding loans becomes available — so that it can tailor its responses. That learning element is critical in the underlying offering because the biggest challenge Aire faces as an alternative model in credit scoring is to make sure it is an alternative. It is easy, he noted, to accidentally build the same hidden biases and assumptions that weaken the older system before pouring the whole flawed packaged into a newer, technologically sleeker package.

“Finding those biases is actually a lot of grunt work for real human checker[s], followed up by a lot of cross-calibrating [of] the models. You have to do all your own work, and check all your own work, or you aren’t really offering an alternative. A lot of tech firms are taking the easy road out and not starting at that foundational place — but if you don’t walk in that uncomfortable forest, you will make the same essential mistakes as what came before you,” he explained.

Today, the firm mostly serves credit card issuers and retail financing providers. As its models mature, Varma noted that longer term loans — mortgage, in particular — are on the firm’s radar.