“Wherever you are on your financial journey, it’s our mission to be by your side.”
This is the mission of the aptly named financial technologies (or “fintech”) firm, Mission Lane. Primarily, Mission Lane’s financial solutions are focused on helping people who aren’t banked (or are underbanked, or don’t have access to credit) and working to give them viable financial options in today’s world.
Speaking of our world, a topic on millions of people’s minds in the past six months or so has been AI (specifically, ChatGPT). I write a lot about AI nowadays but, at times, there’s only so much I can glean while combing press releases or discovering relevant and engaging topics about which great stories can be crafted.
Fortunately, I was recently given the opportunity to have a full conversation with Mike Lempner, Head of Engineering and Technology at Mission Lane, as the firm has made a lot of headway with AI-powered resources.
Here is our conversation, lightly edited and condensed for clarity:
Alex Passett: What has drawn Mission Lane to incorporating AI into some of your workflows?
Mike Lempner: So, a big part of what we do is regulated. We’ve got to be very careful about how we do what we do; that goes for both the cloud and AI.
AP: Can you elaborate a bit?
ML: We grew up in the cloud. We never had a data center, and we started out on AWS. Now, in fact, we’re actually mostly on Google. We’ve built our applications with a modern tech stack, and one of the things we do with these applications – which can be really hard – is making financial decisions.
If you’re underwriting somebody with prior bankruptcy, for instance, traditional banks may basically ignore that person. We don’t want to do that. We want to figure out a way that we can say “Yes.” and support them, whereas lenders elsewhere might shake their heads as a default and say “No.”
AP: And through AI comes that support, you’re saying?
ML: In part, yes. We’ve always been a very heavy adopter of data-centric technologies that allow us to better help others. Advanced modeling techniques, including AI, machine learning (ML) and other statistical algorithms, help with the decisions necessary to make that happen.
That said, there’s a nuance to what we do; we can’t just throw AI at a challenge and say “Hey, based upon this person’s credit bureau and other information, should we accept them or not?” because we ultimately have to be able to justify the decision we made. There has to be true auditability of our decisions and transparency as to how we arrived at that decision, whether that’s approving someone or not.
AP: So, AI for Mission Lane is, in a sense, like it is for a lot of us; a tool, not the end-all, be-all problem-solver.
ML: Of course. It isn’t our go-to for everything, but it is a big part of credit decision-making and underwriting. We use advanced modeling techniques, but we’ve got to make sure that we can be held accountable at every step, which with AI you can’t always strictly do.
With fraud, on the other hand, that’s different. We can use AI a bit more. We just need to make sure that we’re protected.
AP: That’s a really good point. With AI, it isn’t the most crystal-clear process in terms of determining how it arrives at conclusions based on “X, Y or Z” prompts. So, for nuanced financial decision-making, caution with AI is a must. I mean, you’ve always got to be cautious with it and double-check the outputs, but especially here.
ML: At a minimum, yes. It’s a big topic in financial services, as is making sure that we aren’t introducing bias into the decision-making. When you’ve got something like AI, double-checking the inputs is a huge part of our models, too. We have to take a pure credit decisioning approach with unbiased standpoints, and we have to balance that with what AI is able to optimize. The decisions need to be justified by humans, and we still need data scientists to explain what their models do and how they have arrived at “X, Y or Z” conclusions.
AP: Balance is key. I’m with you on that.
ML: And another part of that balance is ensuring repeatability for our clients. If we’re responsibly using AI, we have to be able to achieve the same decision every single time. We can’t have requests come back with dissimilar results; that has potential to cause undue variations in our financial services. We can’t afford that, and we wouldn’t remain compliant if that were allowed to happen.
AP: Another really great point. For instance, in the editorial world, we ensure whatever content we utilize via ChatGPT’s assistance is free of “hallucinations” and is accurate, so our journalistic duties are met. But in fintech, it’s inherently more stringent, it sounds like.
ML: Yes, it has to be.
AP: In terms of Mission Lane, any specific AI use cases you’d be able to share?
ML: Definitely. The team I lead at Mission Lane – with our technologists and software engineers – builds out a lot of the technological solutions we use for credit decisioning, for fraud, for marketing, mobile apps, all those things that support our business.
One of the great ways we’ve been able to use AI is to generate code; it can be super productive and can give you something to start with. Even if we don’t get a finished product from the AI, that’s okay. It generates ideas.
I have a few examples. The first involves a tool we use called Jira. Jira tracks software delivery, how many tasks are active, how long did completed tasks take and so on. Well, we wanted to get Jira data from Jira and load it to another system so we could analyze it. So, a team member went to ChatGPT and asked it to generate a Python script that extracts Jira data and puts it into a file.
That was done in minutes, if not seconds, and it was largely accurate. We still had to rework it, but for the most part this was a one-time task that we largely automated via AI. Beyond the spot-checking, this was a real time-savings moment.
Another good example involves an organizational change we did a few months ago. We have an org chart software, but it’s largely driven by HR. So, when we wanted to play around with a couple of scenarios, one of our developers – he wasn’t a front-end developer who could normally develop a website, as his skills were more on the platform side of things – he took a list of everyone in the organization and asked ChatGPT to generate a site. That site also needed specific colors from our brand kit, a stylized chart with the names from the spreadsheet, and the data broken out by department and by manager.
Within minutes, that was done too. It was honestly a really nice UI, and that would’ve taken significantly longer if not for AI.
AP: Those are as practical as you’ll get for use cases, I’d say.
ML: They were. Another practical place we use it is on databases. We asked ChatGPT to write an SQL query that could identify specific types of customers and arrange them in a table. It did that. It helped us restructure part of that database so all we had to do, then, was copy and paste it.
Our marketing and design teams can use AI to generate email ideas for prospective customers about reducing fees or whatever it may be. We go in after and incorporate our mission and our values; again, these practical examples aren’t indicative of totally finished products. But still, they’re great starting points and generators for effective ideas, and we stay mindful of the outputs as we go.
There are actually more examples, even. AI can summarize our longer internal chats and tell us what’s been discussed and how we can improve. It can transcribe calls, as opposed to hiring someone to record the meeting’s minutes. Overall, there are a lot of real-life AI applications and capabilities in fintech with a lot of great potential.
Edited by
Greg Tavarez
