Podcast: Artificial Intelligence and Accounts Receivable with YayPay

Originally Recorded and published on JohnWright.ai Artificial Intelligence Podcast, July 27, 2018

You are listening to the JohnWright.ai Artificial Intelligence Podcast. Episode 12 is entitled Artificial Intelligence and Accounts Receivable with YayPay, featuring Eugene Vyborov. Eugene is a co-founder and Chief Technology Officer of YayPay. A cloud-based solution for accounts receivable that uses automation to make collecting money fast, easy and highly predictable. YayPay uses machines learning technology to predict risks for businesses — such as late payment of invoices — and suggests work-flow strategies such as how and when to follow-up with a customer about an overdue invoice.

Eugene is an engineering executive with performance-oriented and thoughtful about his work. At YayPay, Eugene is responsible for the company’s strategic technology vision and core product architecture in addition to other duties including product delivery and talent acquisition.

Here’s John’s conversation with Eugene.

JW: I’m here today with Eugene Vyborov from YayPay. Hi, Eugene!

EV: Hi, John! Thank you for having me here!

JW: Thank you for joining me. I do really appreciate it. I appreciate your time. You are with YayPay and I’m hoping you can tell me more about what YayPay is. And what do you do?

EV: Well, YayPay is an accounts receivable automation and management software that makes collecting money fast, easy and highly predictive. That’s a short version of the pitch.

JW: Sure, yeah. And before YayPay what were you doing? What’s your background like?

EV: Well, I have an engineering background at the beginning of my career I was working as a developer and a project lead for a few years. And then I founded and built two technology businesses in Ukraine both in software development space. I was also a Lead Technical Associate at TechStars Boston in 2016.

JW: Today, you are the CTO of YayPay. You are one of the co-founders. What can you tell me about your decision to join YayPay and how did you work up to it? What're the origins of your history with YayPay?

EV: That’s a great question. Let me step back a bit, so Antony offered me to join the company as a co-founder and CTO a few months after he started YayPay. So, I was not exactly next to Antony when the idea came along, but I know that his and my reasoning for this idea are pretty much the same. So, you see, as well as me, he built a few businesses before, while in his case, probably much larger ones. And in every business, he experienced these order-to-cash problems firsthand, and he realized that this is a real thing for many growing businesses and enterprises and I had exactly the same experience from my side. I would say that the main driver for my decision at least was due to the fact that the problem we are solving was actually pretty big issue on the market. And it’s a real thing, it’s something that takes people’s time, takes people’s lives. It was important for me to build something that makes the difference for people, something that can, you know, make the world the better place, if you want.

JW: Certainly, and you mentioned Anthony. I know that he first launched the company with a presentation at Disrupt London in 2015 and you joined sometime after that. I’m sure you are familiar with how everything at Disrupt London went. So, I’m wondering how you can tell me, you know, how YayPay has evolved since then, what has changed, what’s still the same, what’s the history like.

EV: Sure! Well, actually since inception the company changed quite dramatically because when we started we were focused mostly on B2C market and the lower at B2B market, different SMBs etc. We even had an MVP of the product that was designed to fit this particular market. After the TechStars program in Boston, we decided to weave it a little bit and focus on larger Mid-market companies in the United States and that’s what we do now. In our client base, we have got companies with a turnover anywhere between 10 million and a few billion dollars right now.

JW: So, for these companies, YayPay is using technology to predict when payments are likely to come in based on past payment behavior of customers, and I’m wondering if you can paint a picture at, you know, how accurate these predictions can be, and how far into the future are you able to predict things about a company’s finances.

EV: That’s a great question! Well, you know what, I’ll start with a little bit of history because we have gone through a number of stages before we came to the method, to the approach that we currently use. So, our first version of the prediction algorithm was attempting to predict an exact date of the payment of an open invoice based on the historical data and some other behavioral characteristics, just as you said. And, with this first version, we achieved an accuracy of around 80 percent, which means that our predicted invoice full payment date for an open invoice would be in the range of minus three — plus three days from the prediction with an 80 percent of the cases. Well, the problem with this approach was that this kind of predictions has a very significant dispersion, long tail, if you will, which means that it’s difficult with this type of algorithm to make sizable errors for some of the open invoices, and customers don’t really like that, obviously. On the other hands, it’s an interesting point. On the other hand, sometimes they don’t really need that much of a precision. They don’t really need to know a particular date when an invoice is going to be paid, but they rather need to know in which phase of the invoice life cycle it is going to be paid, whether it’s gonna be paid before due date or it’s gonna go overdue or it’s gonna go 60 plus etc. Knowing that, based on this information we changed the algorithm and we currently estimate whether the payment is going to be paid by the due date with an accuracy that is a way over 90 per cent right now. And, once the invoice goes past due, we estimate whether it will be paid in 30, 60, 90 or more than 90 days and the accuracy for those estimates vary by buckets, with the first 30 days being the most accurate, again at around 90 per cent followed by 80 per cent accuracy for the 90+ bucket. Well, we actually expect these models to improve significantly in performance, as we integrate more internal and external sources of data into our models.

JW: That’s interesting and I’m glad you brought that up how you thought or saw that you could predict with precision, you know, they gonna pay the invoice by Tuesday, but your client might not care whether it’s Tuesday or Wednesday. They want to know if it’s before the due date or after the due date. You are helping them to get that picture and catering to what they need to know, rather than precise date. I think that’s very interesting.

EV: Exactly, and it also really turns from the regression problem to the classification problem, which is just more precise by nature.

JW: Have you thought about larger scale analysis you can perform with all this date you are collecting? Because I’m thinking about, you know, large payroll companies. And we asked those big payroll companies, you know: “Where’s the statistics about the employment today? Could companies like Yay pay have good polls in different sectors, different industries, economies as a whole, by looking at the state of payment of invoices?” Is that something that you are looking into?

EV: Absolutely, we are actually looking at trends across multiple levels. For example, for given business we can look at how their sales are distributed across individual buyers and sectors, we can then track how the sectors are performing relative to each other and overtime. We will provide insights to the underlined trends, let’s say in sales, for example, in what sectors are growing or shrinking. As well as the results and risks of that, for example, what does the AR assets really look like from the expected cash flow perspective? When are we going to receive money? At the mental level that you just mentioned, we expect the changes in payment behavior, will not only allow us to predict payments when the invoice is not going to be paid but also would be an early indicator of softening economy in key sectors.

JW: That’s very interesting! And I have read that Yay pay is using an Artificial Intelligence technology not only to automate the collection processes but to make decisions like which method to follow-up on an invoice. For example, what I mean by that is — you might use the Artificial Intelligence technology to decide that we are going to call this customer about the invoice rather than send them an email. Because our AI system tells us that if we call, it’s more likely that they will pay when we want them to pay. And I thought that was really a neat example and I’m wondering if you have any other examples like that you’ve seen of the application of AI in this instance.

EV: Well, that’s interesting. You a kind of see exactly where we are going with all of that. We are working on the follow-up automation and other components of automating the communication between a supplier and a buyer. Right now, we are at the stage when we are collecting the data through a track of interactions between them. In the meantime, we are developing expert systems to provide early detection of issues that could disrupt to order the cash cycle as a whole. A couple examples here, we are, you know, maybe notifying salespeople when a buyer is approaching his credit limit, which is, you know, simply relative — a simple thing. We can flag at potential cyber fraud by detecting a certain drop in payments through a particular payment channel or, let’s say, another example: we can raise alerts on anomalies behavior such as an account that is highly likely to pay is going past due. That means that something is going on there which also, by the way, can be an indication of a bad economy.

JW: Right. I find it interesting that there are some customs in certain industries that also lead to late payment. You know, I don’t work in any of these large industries that you have probably worked closely with. But, you know I have a house and I have a mortgage and that’s an example, I think, of all the time where my mortgage payment is due on the first of the month. But, universally, I know of no mortgage company that will do anything of any consequence if I pay within two weeks of that due date. So, I know a lot of people who just pay a week after the first of the month of two weeks after the first of the month and that’s when they consider when the mortgage payment is due. But when you ask the mortgage company “When is that payment due?”. They say: “The first of the month”. So, it’s just an interesting custom that I have experienced. I am wondering how you deal with nuances like that and if you have to program them into the systems that you are developing.

EV: Yeah, well. Frist of all, I have to confirm that these kinds of nuances are everywhere in every industry.

JW: Oh, ok.

EV: And we are to handle such variation across different sectors to make sure that AI systems have as much access to contacts as possible. That’s why we start with the invoice but we also provide the system with the summary of fast payer behavior as well as the profiles of both the payer and the payee, including industry, the total revenue, and size, etc. Well, in addition to that, you need to be sure that you choose the appropriate algorithms that can actually leverage this contextual information by learning the interactions and the dependencies.

JW: I’m going to go a little outside of the Artificial Intelligence realm here. I’m also interested in blockchain technology.

EV: Oh, very interesting topic. Let’s do that!

JW: In particular, with your business, I read a lot about smart contracts where conceivably a payment might be due under an invoice but the smart contract might have set it up so that the payer account just pays it under certain conditions being satisfied onto that smart contract. So, I’m wondering how you think about that. And from your perspective, in your industry, where you are trying to automate the collections process, do you see the future where there is more automation between the payers and payees, where payment is set up to be automated from the outset?

EV: You know, I definitely see the future for this technology. For one reason or another, there is a very big hype around all the blockchain technologies and smart contracts and future currencies. Now, this hype is going down a little bit. It doesn’t mean that you know, technology is bad. It just means that technology is in this particular sector of the Gartner hype curve.

JW: Could you explain a hype curve a little bit for people who might not be familiar with that?

EV: Yeh, absolutely, it’s just basically a life cycle of the technologies that are changing the market. It shows the significant rise of the interest in technology and then a certain pressure or disappointment when people are not meeting the expectations they had towards this technology and then a gradual levelling up at the medium-high level with the actual applications of technologies in different industries. So, I think that right now blockchain might be on a decline of this Gartner curve but it’s not final. So, getting back to the original questions, you are right, smart contracts and blockchain is definitely something that might be useful for our customers. Assuming again that they will be working out of the same shared ledger with all of their customers. And in this case, they would not need to spend, you know, time and effort on the reconciliation process. That makes it very easy and simple. Actually, I mean, just to think broader, any kind of industry that depends on the ledger one way or another might make use of blockchain technologies and smart contracts. I mean, it actually seems like a no brainer that you can just take this and use that smart contract technology to control and automatically execute business transactions between different agents. But it seems like it, unfortunately, is not that simple. And the reason for that is that technology needs to overcome the limitations of the existing businesses processes as well as people participating in these processes. Many people need to be educated on this topic. And, to give you an example, last year I visited a conference on the West Coast. I don’t want to give a specific name of the conference but I can tell you that there were a lot of finance people there. And in three days of this conference that had three different streams, as far as I remember, there was only one talk about blockchain technology — only one. And I was also chatting with some financial folks over there from different industries, and I can tell you that a many of them have no idea what is blockchain or smart contracts — what they do at all. They do know a little more about a cryptocurrency. Well, based on that I believe that even though, one day smart contracts will become the thing. This day it is somewhere further down the road than we expect.

JW: I can see that. And, I think you would know because you are so deep into this fintech, financial technology space. Within that space along with terms like blockchain and cryptocurrency, Artificial Intelligence is a term. I know I can throw it around, it’s a buzz word, it’s used in a lot of headlines to collect clicks and attention whether it’s correctly used or not. But it’s related to other concepts like machine learning and deep learning. And I’m wondering if you have any thoughts or have seen any interesting examples of any of these technologies in its application in your space of fintech.

EV: Well, first of all, I’m pretty sure that what we’re doing at YayPay is going to be a great example of the application of the AI in fintech.

JW: Sure.

EV: Well, the reason I’m saying that is that, you know, the back office guys including finance teams quite often are somewhat forgotten by the process. People over there still have to spend a lot of time and effort in doing manual and mundane things. And in a way, Artificial Intelligence is the tool to let those people break free, switch to useful, creative, value-generating things in their day-to-day job. And of course, there are many more examples of, you know, successful Artificial Intelligence applications in the industry. And actually it’s probably everywhere now, you can look at fraud detection, credit decision making, risk management, trading, all kinds of conversational services, insurance underwriting, etc. It seems to me that many of the Artificial Intelligence start-ups are taking the same path as we are: they are applying machine learning techniques to already well-known inefficient processes and making them work better rather than trying a sort of reinvent an entire thing. Basically, any area that requires decision making and has large amounts of structure data accessible, which is very important in this case, can and will become handled by AI at some point of the near future. One big challenge was all the different applications of the AI in the industry right now is explainability, though. What I mean by that is, that consumers of the decisions, that these AI systems are making, need to understand why those decisions were made. And with narrow networks it’s virtually impossible at least right now. Many groups are working on solving this issues but I have doubts that it’s even possible to solve in a general case. Probably because of a human brain may not even be able to comprehend the explanation of the decision made by this narrow network, even if the explanation is available.

JW: So, no matter how hard we may try, there might be an outcome that a narrow network reached, that we just might not be able to get to an answer.

EV: We’ll get to an answer but what this system gives us, we won’t be able to understand the explanation why it gave this answer.

JW: Hmm, interesting.

EV: Actually, you know, some of the researches, even if we get back to a human brain… Some of the researchers say, that there is one system, that is making decisions and there is a completely another system in the brain that is explaining those decisions. And those two are not necessarily connected in a causality way. One of them is explaining the decision not necessarily is using the same factors that make this decision happen.

JW: That’s very interesting. I find it fascinating how much we can learn about AI technology by going back to research, that is about how our own minds work. So I appreciate that example. But it sounds like, how you’ve described YayPay, that you are the heroes of the folks in the back office and working with them to help them do their job better. Are there any other kinds of issues that might not be the same as the account receivable or the collection process that your technology can also address to? What do you think?

EV: Well, as I said at the beginning, the mission of the company is to make collecting money fast, easy and highly predictable. Obviously, we are striving to fulfill this mission, which for us means building an entire order to cash workflow and making it automated as much as possible. That by itself is a very substantial mission to deliver on and it will take us some time to accomplish that, change mindsets of the finance teams and revolutionize market in a way. But ones we do it, you know, anything can happen, there are a lot of other not exactly efficient back-office processes that are sitting there waiting to be automated. From the machine learning perspective, we are collecting a lot of very interesting data that could and should actually be leveraged, for example, for credits, for invoice factoring etc. But as I mentioned we are in the efficiency game right now, which means making finance teams run on autopilot is our main priority.

JW: Oh, that’s very exciting. I am really excited to learn more about YayPay and see how your company develops. So good luck to you. And I hope to stay in touch and maybe we can have this conversation in the future to talk about more about what YayPay is doing down the line.

EV: Thank you very much for having me here, John.