Eyal Shinar: Your distinction between Kabbage and Ondeck, and Amex and Intuit is accurate. But I think it’s missing the big picture. You don’t need credit as a seller if you’re getting paid immediately. You need much less working capital than before. That’s what we’re trying to solve. We’re allowing the seller to get paid instantly.
Sramana Mitra: Let me get one thing sorted out here. Are you talking about your focus being receivable financing?
Eyal Shinar: No, I’m not talking about that. Our focus is to facilitate a seller to get paid now just like Amex and Visa did back in the 50s on the B2C side. When a restaurant is selling to a consumer, instead of waiting, they get paid instantly. On the B2B side, when you’re providing wood for a furniture store, you have to wait for 90 days.
Sramana Mitra: They are parallels, right? One is business credit cards. People do order with business credit cards. The second is receivable financing. I understand the positioning. I’m trying to understand how you do what you do.
Eyal Shinar: I would say it’s more than just positioning. The way we do what we do is, we’re trying to go into much bigger trends. You mentioned political and demographic macro-trends that affect the decision of investments around FinTech in India.
We attach to much bigger technological and data trends. If you look at what happened in the last eight years or so, there’s this big trend of digitalization. We’re moving from offline to digital form. Even if you’re a government agency, you’re going to have sets of digital data.
The second trend that is slightly more recent is not only migration to digital but migration from desktop to the cloud. This phenomenon is relatively new in the grand scheme of financial history. What is created is a few things.
First is, you have all of those apps that are very use-case specific. You mentioned some of those. You have accounting software apps. There are ERP systems, CRM systems, and inventory management. It’s super granular. It’s also highly accessible because all of those apps have a few sets of APIs. Some of them are private and some are public. All of these data are accessible with the permission of the user.
The third trend around technology, which is beyond the shift in data culture and usage, is breakthroughs around deep learning and machine learning technologies that really allow you to draw the data in real-time, come to very accurate conclusions, and determine the probability of risk and default.
We tap into those bigger secular trends and say, “In the first few years, we’re going to use all of this data and signals in a true unsupervised machine learning way.”
Let’s say you want to build an app that recognizes a cat’s face on Instagram. The right scientific way to do it is to build a few machine learning models and without having any set of rules, run those models on a very large set of samples.
One of the models would be a little bit more accurate than the others. You’re killing the model that is not accurate. The model that is accurate, you improve. After many iterations, you’re going to have a model that, by itself, is as accurate or more accurate than a human.
The cost associated with that is, mostly, the cost of data processing and data storage. When you try to do that method in payments and credit, there’s an additional set of cost because every sample is either a successful payment or an unsuccessful payment.
Unsuccessful payment actually costs you money. It’s very tempting not to use a true machine learning model and use the rule-based form, but you cannot build your machine learning model based on that. You need to let them boil the ocean, so to speak.
In the early days, we just let almost everybody in. Over time, the models just taught themselves default rates. In the US, we have the most accurate model that is 100% machine learning-based and fully automated. We still have zero underwriters in the business.