In our earlier post, 4 Ways Machine Learning Makes Small Business Lending More Profitable, we discussed how machine learning can help traditional lenders make smarter lending decisions and better control risk. But Mirador’s connection to machine learning runs much deeper than this. In fact, machine learning was, and is, central to why we founded Mirador as a company.
As a non-engineer working for Mirador (I’m the company’s recruiter and company curator), I recently sat down with Co-Founder and Chief Technology Officer William Beaver to learn more about machine learning and how Mirador wants to harness its potential to collaborate with banks, not disrupt them — setting the company apart from many other fintech startups.
How did you get interested in machine learning?
I spent about 10 years working in financial services, software engineering and product development. Towards the end of that time period, I started going to class part-time at Columbia University [in New York City]. I wanted to conduct research, so I transitioned to being a student full-time and on my first day on campus, I met the person who became my graduate advisor. His background was in machine learning theory, and I was really interested in becoming more aware of the capabilities of machine learning and what that meant from a pragmatic perspective. We would go talk to an academic department or a business to learn about a problem they faced, and then I would work on it for a quarter or a semester. For about 10 years, we did a lot of machine learning work in a lot of areas that gave me exposure to a variety of problems that people were facing.
Which of your projects stand out as particularly interesting?
I really enjoyed the process of learning new problems in new areas and domains. My background in consulting let me easily adapt to new environments and understand what problems businesses faced without being an expert in any of them. My degree is technically in machine learning and applied statistics, but my Ph.D. project involved building a machine learning system so developmental biologists studying the fruit fly didn’t have to spend as much time measuring information in their microscopy images. They would take a fruit fly embryo, chemically treat it to highlight certain genes, and then take these pretty pictures when they ran the experiment. From there, they would use editing software to manually count the number of cell nuclei or the number of genes expressed and store all the information in ledgers and spreadsheets.
Essentially, I used machine learning to automate all the analysis for them. They taught the machine what they were looking for in those images. They would annotate directly on an image, and the machine would come back and say, “I think this is what you’re annotating. Tell me if I’m right or wrong.” They would give feedback, and the machine would learn and adapt based upon what they told it over time. Because of this, the biologists were able to start using the information they had collected over the years.
How did you go from that project to founding Mirador?
If you have a problem that you can define and enough data to tackle it with machine learning, pretty much any graduate student will be able to use machine learning to get you a decent answer within a semester. The problem is, if you ask a different question next semester, he or she has to start from scratch — either because the problem changed enough that the approach is no longer applicable, or [because] the technical underpinnings can’t be adapted to a new environment. So I learned that it’s really important to create systems that allow you to build machine learning classifiers quickly and efficiently.
When I met my co-founder, Trevor Dryer, he described this problem of how to help lending institutions make decisions about small business loans application quickly. I realized that if you could describe how an underwriter thinks about a problem, I could draw parallels to all these other domains I had solved problems in. If I’m an underwriter for a certain type of business loan, for instance, I have a pretty clear idea of what I’m looking for in a loan application that tells me if it’s a good deal or a bad deal. The less time I spend coming to that decision, the more opportunity there is for profit in that loan. If we can wring out as much time and energy from the lender’s experience and make the borrower’s experience enjoyable, machine learning itself can provide a really clear picture of what that particular small business might look like from a risk perspective.
What role do members of the Mirador engineering team play in the machine learning building process?
Because our emphasis is on incorporating machine learning into everyday decisions, we’re fortunate that we try to bring machine learning engineering into the engineering organization itself. Our engineers implement the solutions we use. So there isn’t a real difference between the team working on the application that puts the borrower through the system easily and quickly and the system itself. Everyone is involved in everything here — which is a real benefit, especially if you’re not trained specifically in machine learning. It gives those engineers a great opportunity to gain exposure to machine learning without needing to have an advanced degree in math or machine learning itself. The job itself is a continual learning opportunity, and people have the opportunity to grow into the areas that they’re interested in.
What new features are the engineering team excited about?
We’re about to roll out the ability for our lending partners to drive part of the application process. So if you’re a small business owner, and you want to apply for a loan, you can work with a lender right on the premise to start filling out the application. If you need to attach a copy of, say, your tax return, the lender can electronically send you a link that will allow you to go back and finish the application later at home. This allows lenders to drive most of the applications coming into the system instead of relying on an online marketing channel. And it makes the process as painless as possible to borrowers, too.
What’s the most exciting thing about working at a fintech startup?
At [any] startup, when an employee walks through the door, he or she will have an impact on the business every single day. At a fintech startup, we have the extra benefit in that our business model is rooted in sensible business that has a positive impact on communities by improving the loan application process.
What makes Mirador exceptional?
Rather than disrupting banks, we really strive to partner with existing financial institutions to help them do a better job. Our mantra from day one has been that banks are the pillars of their communities and of the financial system. While there’s certainly room for them to improve in their efficiencies and their customer interactions and online experience, our job is to help them get there instead of displace them.
In fintech, the companies that get all the press are the disrupters. But there are a lot [of fintech players] that are successful by providing efficient process replacement for an existing domain. At Mirador, it’s not an “us versus them” mentality. It’s all about how we can collaborate together.