Mission Lane was founded in 2018 with the idea of expanding access to credit cards to users with limited or spotty credit history, and has acquired over 1 million customers since that time.

The San Francisco-based credit card issuer, which uses machine learning (ML) algorithms to supplement data available from credit bureaus, is looking to grow beyond being a credit card provider and expand into debit and digital banking services. Mission Lane offers its cards on the Visa network in partnership with the $959 million Transportation Alliance Bank.

But while ML can deliver compounded insights, Chief Data Officer Krishna Venkatraman told Bank Automation News it is equally important to test how those insights work for the customer segments they target, instead of just taking the recommendations at face value.

BAN caught up with Venkatraman, along with Mission Lane CEO Shane Holdaway and Zeenat Sidi, business head of digital banking, to learn how the company’s use of data drives its products and how the competition between legacy banks and challengers is likely to play out.

What follows is an edited version of the conversation:

Bank Automation News: What are some of the gaps in the existing credit scoring system? How do you think they can be patched?

Krishna Venkatraman: If you think about who the current suite of credit product serves, well, it usually serves people who have a lot of data already about them. I think that leaves about 100 million customers sort of looking for a home. And we want to be able to provide great products by really understanding that credit risk better by innovating on data, product and customer experience.

As we tap into newer sources of data and build up proprietary data sources, by the end, having those customers within our franchise and learning from them very actively, we think we will be able to truly understand risk and therefore offer products that are much more fair and more transparent.

BAN: Do you plan on continuing to work with banking partners in the future, or is the plan to pursue a banking license of your own?

Shane Holdaway: We work with banking partners; we don’t call ourselves a bank. We partner with great banks, and they’re important partners to us. But really, what we are trying to do is provide a path to better credit for a lot of people that don’t get access to that lending — being kind of the tip of the spear for that.

BAN: Over the next five years, do you think it’s going to be a neck-to-neck fight between challenger banks and the traditional institutions? Or is it going to be more of a banking-partner relationship?

Zeenat Sidi: I can tell you, having now worked with larger banks, and then most recently, spending almost three years with SoFi, I think one thing we’re seeing at companies appearing as challengers is that there is a real maniacal focus on what the customer needs are and solving for them.

A lot of larger companies that have, legacy infrastructure — operational technology or human — [and] are not able to be that nimble, are finding ways to partner. They bring scale; the smaller companies bring this nimbleness, agility.

KV: On the other hand, you have a lot of players that today you may not even think of as in financial services — like Uber, Google or Amazon— they are creating financial products. They are great at marketing, product innovation and experimentation, but they don’t have a long history of understanding credit risk and decisioning. A company that has the best of both — a very strong foundation in understanding the impact of decision making in a fintech world, coupled with this mindset of innovating and customer experience — I think those are the ones that are likely to have the best chance of success.

BAN: What do you make of regulatory curiosity and larger conversation around fairness impacts of artificial intelligence and machine learning use in financial services?

KV: What I think the challenge has always been for AI is: When you have data, you have to make sure that the insights you’re getting from data, you understand where they come from, and how they work for the different segments we’re trying to serve. The second piece is, at the end of the day, we do not let models make decisions. We use models to help us understand what data is saying but eventually, we have policies that make decisions.

BAN: Finally, for a customer, who is perhaps not a data scientist or not as acquainted with data science, how do you convey these elements without overwhelming them?

KV: Firstly, we want to be transparent about the data we use so that that customer benefit must be very easy for people to understand. Here is why we’re asking you for this data is to make this decision. Here’s why it helps you. The second is, when we decide, by regulation we have to give reasons for certain types of decisions on why we declined somebody. So, we do spend a lot of time in understanding and creating the mechanisms to really explain to a customer, what are the factors that went into a decision. We are very adamant about not just creating blackbox models.

[Editor’s Note: “Blackbox” models do not allow a third party to see the inputs and operations that determine system recommendations.]

SH: But we have well over a million customers now, and we can’t talk to each one of them individually. So, we must look for indicators of where we’re getting it right and where we’re getting it wrong. We are serving humans; we’re not just solving math problems. And using that information to inform judgment that we layer on top of what the models put out is a critical aspect of how we do underwrite.