.jpg)
Leaders In Payments
Leaders In Payments
How AI is Transforming Payments with John Minutaglio, CTO at Payarc | Episode 376
In the latest episode of Leaders in Payments, Greg Myers speaks with John Minutaglio, CTO of Payarc, about the role of artificial intelligence in payments. Payarc, a rapidly growing payment processor, has built a reputation for providing high-touch service to merchants and agents. Their focus has always been on fostering strong relationships, and AI is now playing a crucial role in enhancing that experience.
John explains that Payarc’s AI strategy is centered on augmenting human interactions rather than replacing them. Their AI framework, called PIE, assists customer support teams by providing insights and recommendations, allowing representatives to respond more efficiently and personally to merchant needs. One significant use case is churn prediction, where AI analyzes data to identify merchants at risk of leaving. By proactively alerting human representatives, Payarc can engage with customers before they churn, leading to increased retention.
Another powerful application of AI at Payarc is in developer support. Their AI tools help software providers integrate Payarc’s payment solutions more effectively by analyzing code and identifying errors, reducing the need for complex troubleshooting. AI also enhances fraud detection, particularly in mitigating ACH reject fraud. By analyzing behavioral patterns and external data sources, Payarc can identify suspicious activities before they escalate into financial losses.
John highlights that AI in payments is not just about automation but about using intelligence to drive better service, security, and efficiency. As AI continues to evolve, Payarc aims to remain at the forefront, leveraging advanced models to improve customer confidence, reduce fraud, and enhance the overall payment experience.
Welcome to the Leaders in Payments podcast, where we talk to C-level leaders from across the payments landscape. We'll be discussing the products and services that impact the payment space today, as well as trends and predictions for the future of payments. We will also hear stories from our guests about their journeys to the top.
Speaker 2:Hello everyone and welcome to the Leaders in Payments podcast. I'm your host, greg Myers, and on today's episode we have a very special guest, john Minitaglio, the CTO at PayArc. So, john, welcome to the show.
Speaker 3:Greg glad to be here. Thanks for having us.
Speaker 2:Well, why don't we start out by learning a little bit about you? So give us a little bit of your background, maybe where you're from, where you went to school, where you currently live, a few things like that.
Speaker 3:Sure, we'd be happy to do that. Greetings from Atlanta, georgia, where I've lived for the last 15, maybe 16 years. I'm in the Buckhead area of Atlanta, so that's where I am Grew up kind of in the United States. My dad worked for IBM, the esteemed typewriter company or computer company, and his career had us moving around. So Georgia is my 17th state and have been around the block from that perspective.
Speaker 3:Went to school at University of Florida, go Gators and have been in fintech pretty much my entire career and in payments at three stops. So I worked at Payark we're a payment processor, as you know. I'm excited to talk to you more about Payark. But back in the day worked for Vital Processing and was the CTO there. Vital was T-SYS and Visa USA's joint venture to get into payments and compete with First Data and some others in the US, and then also had the CTO role at Pfizer, the bank outsourcer, for a while. So that was a little bit more than payments but in and around payments. Outside of that worked for a couple banks and other fintech, mark tech, that kind of stuff. So fintech guy, fintech, marktech that kind of stuff.
Speaker 2:So FinTech guy Okay, great. Well, thanks for sharing that. So tell us a little bit about Payark. What does Payark do?
Speaker 3:Payark is an ISO with proprietary technology or a super ISO or a payment processor, so we are a young company. We were founded in 2016 with our first processing in 2017. And I've had kind of rocket ship ride growth, so I've enjoyed success in the market and growth in revenue and growth in client base In today's speeds and feeds. We have a little north of 150 employees. We're headquartered in Greenwich, connecticut, and have about 60 people there, and then have an office in Tijuana, mexico, and also an office in Sofia, bulgaria, so I've been growing rapidly with staff. And then, from a market standpoint, I have about 10,000 merchants that we provide services for and work with agents and provide services for agents as well. So we're an agent-focused processor with proprietary tech.
Speaker 2:So in this episode we're going to dive into a topic that comes up in almost every conversation that I have related to payments or fintech, and that's AI. So let's talk about AI and payments. So what areas of payments did PayArc excel at, maybe before AI? And then how will AI enhance those services to keep you guys being a leader in the industry?
Speaker 3:AI, ai it's like talk like a pirate or talk like a New Zealander. Ai that's a great question. Before AI, in Payark, years would be about a year ago. We really have been maybe 10 months on our AI journey and I know we'll talk more about that. But to your question, the Payark claim to fame is service. If you think about Payark and you talk to our customers, we love working with PayArk because the service is great. That generally is high-touch, relationship-oriented service. So our agents know us and they have relationships with the employees of PayArk. Our merchants know us and they talk to five PayArk employees, not just one. So it's not a nameless, faceless interface of service, but high-type service and a good way to explain that one of our corporate goals is actually to be the best part of a merchant's day. So we like to delight and we do that through service.
Speaker 2:Okay, and how have you used AI to enhance service so far?
Speaker 3:That's an interesting question as well. So what is near and dear to us is the human touch. So part of that service paradigm is things that AI by itself wouldn't necessarily do. Well, it's, you know, greg, how is your day, how's the weather, how are your kids? It's empathy and relationship tone and tenor. So our strategy with AI is not to put AI directly in front of the end customer, but to enable our humans to do better. So that's kind of the strategy.
Speaker 3:With that. It takes a couple different courses. So it's really assistance and advisory aids for customer support reps and relationship managers. So if I'm a customer support rep, when I'm giving you support, I'm using AI tools that advise me to get you the right answer quickly. Or if I'm a relationship manager, I use tools to help me guide the conversation around what you've been doing with your business. Think about a quarterly business check-in. Instead of doing my homework off spreadsheets and digging into the data, ai helps me accelerate that piece of the puzzle so that I can delight you at a human piece. So AI standing behind the human is really our take on the approach.
Speaker 2:Okay, and you recently launched Payark's AI program called PIE P-I-E, and you've stated you kind of talked about it there a little bit but it's not to replace humans, it's to enhance humans. So why do you think that's so important and how is it really helping your team members today? Yeah, so.
Speaker 3:Pi is our framework for all AI-enabled services. It runs the gamut from detailed, hardcore machine learning and traditional AI to chatbots and large language models and generative AI helping people talk. So that's really the wing-to-wing strategy and it does a couple different things. Success to date has really been on a series of focuses and, in no order, churn. So looking at our portfolio of merchants and predicting when they may leave us and then advising the human hey, John, you should call this merchant. They are having a problem and they seem likely to leave. Drive the outbound call, go talk to them, figure out what's going on and make it right. So you see that that advisory would prompt me as a human, to make a call that otherwise I would not be doing. In practice, we have a retention team of humans that take that advisory and dial out. So churn would be one example of how the humans are helped.
Speaker 3:Another good example would be development support. So we provide services to software as a service providers. So these are companies that develop software and they're integrating payments into their solution with PayArc using APIs or SDKs. So the questions of those types of merchants or ISVs is really coming from a developer. So I'm a developer and can you help me with my code.
Speaker 3:A frontline support person can say yes, give me your code. They plug it into an assessor and a language model that says sure enough, on line 36, you have a typo, you need a semicolon in that call and there you go. So that's another version where, instead of calling a level one and level one says well, I'm not a developer, I don't know how to talk to you, let me escalate it to level three, the developer. I don't know how to talk to you, let me escalate it to level three. The level one can actually provide service and help you right there on the spot. And that would be another example where, prior to AI, you know I'm level one, I don't know how to develop, so I can't really help you. In today's world, I can't.
Speaker 2:Okay, so how I'm going off script a little bit here, but it kind of makes me think about, you know, these tools. Are these like off-the-shelf AI tools that you've integrated into your platform, or are these things you guys have created using AI?
Speaker 3:It's a framework and a toolkit that is available commercially. Our technology stack and our hosting environment is in the cloud and we use AWS, amazon Web Services. That's our solution. Aws provides a plethora of AI-enabled tools that run the gamut from large language models to Gen AI tools to a tool like SageMaker, which is a machine learning engine that can be programmed. So think of them as Legos, right? So I have all these tools. I pay or wrap these tools and integrate them using development languages. So I write to SageM these tools and integrate them using development languages.
Speaker 3:So I write to SageMaker using Python, which is a development language, and that Python code that I write tells the engine how to predict an outcome like churn, and that proprietary tool or commercially available tool from AWS renders a result. I take that result. I develop a graphical user interface, a dashboard that allows the user to predict churn. So the user logs into a pretty screen. The screen has a work list of call Greg Co. You can click through that and it says well, the reason that Greg Co might leave is that support is terrible. He's had a number of issues. You click through it again. The screen says here's a script I would recommend that you use. So all of that is kind of a SageMaker did a prediction with a Python training. That training was rendered into a UI using traditional development technologies.
Speaker 3:That script on how to approach the customer is a language model in AWS Speak. It's a tool called Lex that has an input and produces an output that is English-based stuff. Some of the stuff is things that I think a layperson. It's a tool called Lex that has an input and produces an output that is English-based stuff. So some of the stuff is things that I think a layperson would have heard about. Everyone's heard about ChatGPT. Chatgpt is a language model. The company is OpenAI. In certain solutions we use OpenAI's technology, but it's not simply taking an off-the-shelf chatbot. It's not taking ChatGPT and giving it to every customer off-the-shelf chatbot. It's not taking chat GPT and giving it to every customer. It's using the embedded more detailed technical tools as a framework and wrapping them into an overall solution that our internal users to face customers and provide better service.
Speaker 2:Okay, great Thanks for explaining that. Sometimes we gloss over some of the details, and I think they're pretty important. One of the areas within payments that we always hear about when it comes to AI is fraud and security. So how are you using that PI framework to make your fraud and security tools better?
Speaker 3:Yeah. So there's a couple areas where you can attack that. With AI, our focus has been on ACH reject or chargeback fraud. So if you think about let's use ACH reject, that's not necessarily fraud. So I might reject an ACH reject as a processor. That would be Greg Coe is not paying his fees, so it's payment default. That might be because you're going out of business so you're not paying your bill because your cash flow is impaired and you can't pay. Ultimately the processor won't get paid. So we want to understand that. It might be nuances to your business. It was a holiday weekend and your retail establishment was closed so you have less funds. You can't pay your bill on the 15th because of that, but in the next business cycle you'll rebound and you will be able to pay. It could be that it could also be bad actors in the entry. So you frauded Greg Coe, you got through underwriting and your sole intent was to look like a good actor but then ultimately be a bad actor and take funds and leave, right. So that's kind of the ecosystem of what you see in ACH reject part, fraud, part not.
Speaker 3:So our approach with AI is a two-line of defense approach for ACH reject abatement. I guess that's a good way to call it. So the first line of defense would be in underwriting. So, prior to boarding you, can I see that you're a bad actor? Can I qualify you better? Can I understand that? And then say at the front door, no, you're not doing business with payer.
Speaker 3:And the AI version of that is looking at additional sources of information. Generally it's called ascent, so open source intelligence, which sounds geeked out, but that's what it's called. So think, in a traditional underwriting model, I look at you, the owner, greg Myers. I run a FICO on you. I understand your scoring. I look at history. I look at the demographics of your business. It's brick and mortar, yeah, you're great. In an OSNET model, I would also look at other publicly available information. So think of things like your social accounts, posts that you've made on Reddit, dark web activity, other things like that. To try to correlate, is Greg Co known for nefarious activity? And, if so, can I use that information to augment the underwriting data, not strictly FICO and that kind of stuff, but other types of parameters that would say it looks suspicious.
Speaker 3:So that would be one example open source intelligence and, more technically, ai, manner patterns. So, in the nature of your application and looking at all of the little nits and subfields, is there anything abnormal about the way you applied? Is the address abnormal? Is the order that you did it? Is the timeframe of being introduced to us to applying abnormal? Is the amount of times you reached out to us abnormal? Looking at that and try to pattern it against an abnormal behavior pattern where AI would understand that better than a human. It's so minuscule that a human couldn't see it to say, hey, there's some risk on this guy at the front end.
Speaker 3:So think about blocking at the front end.
Speaker 3:On the back end it's using AI to predict the ACH reject. So, looking at all of our in-production data elements hey, ai machine learning, what is the likelihood of this merchant to reject an ACH and, if so, signal a human to investigate the account? Oh, the prediction engine says Greco is going to reject. Let's look at it at a human level. You know, greco, I'm going to put a hold on your account because of this, that the other I'm going to call you. I'm going to explain what I'm doing and I'm going to put the hold on so that when the ACH rejects occur, it has no business impact because the funds already held. And then if the engine's wrong, we release the hold, etc. So general course of business, stuff wrapped in an empathetic customer service solution that doesn't alienate the merchant, but ultimately that prediction is driving human behavior that wouldn't otherwise go. So think about stopping at the front door, predicting it in the operations of the business, using AI on both legs of the equation to make the world a better place. As it relates to less ACH, reject.
Speaker 2:And you mentioned customer churn earlier, so let's dive a little deeper into that. It's a major issue across a lot of different industries, but definitely within payments. I mean, I've been at organizations where that churn was 25%, so every couple of years, right, it's a whole new portfolio, which is crazy to me. But it seems like the way you guys have positioned Pi, that you can identify potential churn at a 85% accuracy rate, which I think is amazing. So what does it take to build a model like that? What information do you need and why is it really so important in this space?
Speaker 3:Yeah, let's start with why it's important. It is what you said. So 25% might be low. I think in Payark it's closer to 30. In other organizations I've been in it's closer to 30. So almost a third of the business leaks out the back door and there's various reasons for that. I, the merchant, could have developed a relationship with another agent. I could have been given better pricing. I might be frustrated with the support experience of you as a processor. There's a million reasons that you could go. But if you think about if we could bring that 30% loss at the back door down, even 1%, 5%, it becomes millions of dollars very quickly. So that's the importance of it. That's an industry step. If you're an ISO or a processor and you lose and churn 30%, that's probably right, plus or minus maybe 10%. So compelling cause big money. What it takes to do that is use AI, right. That's the simple part of it.
Speaker 3:The way we approached it is we looked in market and said, hey, are there tools that are payments oriented that fit the bill? Can we buy something? We didn't really find anything that drove a. Let's do it ourselves. Inter-ai and machine learning. What we did is we looked at big data. As a processor. We've been in business for years. We've had 10,000 merchants declining as we've grown back in time. They can look at all of that big data set, all processing data in the life of you being with Payr. We know exactly how you process what you did. We also understand non-processing data, like your application, who you are, what's your FICO, what's your birth date, what's your zip, other stuff like that. And then we also know your support velocity. So how many times have you called us for support? What were the issues, what was the average time of resolution? And then also, when you called, especially if it was a voice call, we can analyze the sentiment of that call. So if in the call you were saying you know I just called John and I love Payark and just had a minor issue and know you guys will help us, that's very different of a sentiment than this. Is the fifth time I've called, I'm getting tired of calling, this hasn't been resolved. Who do I have to talk to? One has frustration, the other doesn't. So we can look at that as well.
Speaker 3:And then in the machine learning engine we try to correlate all of these data points what is the average batch? What's the chargeback? What's your FICO score. How long have you been in business with us? Things like the zip. So maybe you're in a zip that is competitive, there's a lot of agents in your zip, so maybe you have a higher likelihood to churn because of the competitive aspect of your zip versus you are in Macon, georgia Nothing against Macon, but probably a less competitive environment, right?
Speaker 3:And then we look at all of that and have trained an engine. Training is important because we, as a processor, know the fatal event of churn. So if you look at any, we can look at a portfolio of our merchants and say these people continue to do business with us, these people have left us for whatever reason, and we understand all of that data associated with that set. What the engine can then do is model is hey, I'm seeing patterns in the data that would allow me to predict that a merchant that is with you today, what is the likelihood of them leaving you in the next 90 days? So that 90 days is important because it gives us a runway to remediate it and that runway is really the human in the middle saying hey, the engine told me you're probably going to leave Greco. Let me reach out to you and have a dialogue with you and see what we can do. The engine also advises on why you might be leaving. You're in a competitive environment. The pricing might not be competitive. Talk to him about pricing. He's had a number of tickets with frustration around his funding timeframe. Move him to same-day funding. It's not always price, but it's differing types of dialogues that ultimately become high-touch, that remediate it.
Speaker 3:But what does it take to do it? I mentioned at the preamble we're an ISO with technology, meaning that we have a development staff. I, as CTO of Payark, have a staff of about 35 people, mostly software developers. We use that resource pool to educate and learn about AI and machine learning technologies and then we built the training and built the model using this AWS toolkit to create this stuff, and that's the deal, and we've been excited about that. You mentioned the 85% accuracy of the prediction, which really is wonderful. So prior to that, I don't know if 30% are going to leave. How would I call them? I got 10,000, I can't call them when you get more of a focused Greco has a 99% confidence of leaving in the next 90 days. I know I need to call them, I know I need to get on it and that's how we do it.
Speaker 2:It seems like a lot of the AI solutions, because you talked about it when we talked about fraud. You talked about it with churn. Is developing that profile or that sort of patterns that you see? I mean, is that a fair assessment?
Speaker 3:Yeah, all the prediction stuff right. So anything that's going to use a big data set to make a prediction is you know. The joke is, nobody told me there'd be so much math, right, but it is looking at you know, training approaches and iterating. So you don't train a model and say here's some data, tell me how to predict churn. You do it hundreds of times and look at the mathematical likelihood of the success of the model and then train it up In our journey to the 85% accuracy.
Speaker 3:The first version of the model had 13% accuracy. At least it was something. But then what do we want to do? How do we tweak that? How do we iterate it through to 85? 85 sounds great, but we're not done. Part of what we're doing is every day I guess more literally twice a week we are working on new models and we shoot those models out. Are they better or worse than what we're doing in production? If the new model predicts at 86% and the production model presents at 85, we switch them out. So over time we get better at this and our training gets better and our proprietary prediction model has a higher confidence, which ultimately allows us to have more of a laser-like focus on that human outreach to really abate the churn. But yeah, I mean, there's math and data science and you have to have developers do this, sure, sure.
Speaker 2:Okay, well, looking ahead, how do you see Pi helping to evolve payments, and what role do you think PayArc will play in that?
Speaker 3:I mean, some of it is have you watched TV or read a magazine? So I think every industry is going to be disrupted to some extent by AI. Payments is no exception. I think everyone in payments is thinking about AI. You can find a lot of white papers. Visa has great white papers. Boston Consulting Group has white papers on the impact of payments to AI. It's prediction and business intelligence and language models.
Speaker 3:At the top of the house level, it's all the same stuff. Ai will make everything more productive or cheaper. With stats that are interesting. Productivity will probably go up 30%. Cost will probably reduce 10%. That's what gets the board excited about this stuff. It's real improvements in efficiency, effectiveness and cost. That's great. I think the next step is a little harder. We have the white paper, and then what do we want to do about it? I think that's where we're excited at PayArc that we feel like we're on the leading edge, cutting edge I don't know what the right word is, but something like that where we have our white paper and we're talking about it, we're actually doing it and we also actually have the models in production and we're using it, and that is an advantage that we think we have in our business relative to the competitive landscape. Because when I talk about this iteration model.
Speaker 3:You have to start to finish. So if I'm eight months into it and you haven't started, you're eight months behind. To some extent. You can't bring more people to it to catch up. You just have to do the work and go through it.
Speaker 3:I think what happens in payments it's higher confidence for customers and consumers. I think ultimately it is I use my payment and it works and there's no fraud. I don't have to worry about anything. For a merchant, it's lower charge backs, it's higher profitability For an agent, it's a larger book of business for merchants and for a processor it's lower churn and greater service. That's what's going to happen in payments. But probably the biggest thing is going to be consumer confidence. That's going to drive more unique payment types, more advanced payment types, more acceptance of alternate payment schemas and stuff like that, I think is what's going to happen.
Speaker 3:And how do we want to be there at PayArk? Certainly in a leadership role. I think we're trying to figure out what that means. We have thought about the tools that we're using at PayArk have been developed for PayArk as the customer, meaning we're developing them and we're using them. Said another way is that they're built to be multi-tenant right. So these might be tools at some point in time that we may want to offer to the market. So, for example, if our churn engine is smoking, we might say, hey, greg Iso, we offer this churn engine as a service for you as a processor. There's a lot of stuff involved in that, because you may view Payark as a competitor, but we've seen others in the industry kind of go from being a competitor to being a service provider and that's probably something we want to explore if Wynn is still in flight a little bit. But yeah, we would love to take our learnings and our success and realize some market value in them as well. So that might be something that we would see Payark doing in the future.
Speaker 2:I hear a lot of companies talk about they're AI driven and when you really kind of and I'm not saying this is necessarily just payments, I think it's kind of across the board and you peel the onion back and they're putting a chat bot on their website and it's, you know, it's answering some questions and they're an AI company. But I think you said something interesting there, like you're eight months ahead and if you haven't started, you're definitely eight months behind and you can't catch up overnight. And to me, that's kind of the fascinating part about companies like you guys, who are out there not just talking about it but doing it. So, john, we've covered a lot of ground about you and the company and AI. Is there anything else you'd like to add before we wrap up the show?
Speaker 3:No again, thank you for having us. It was great talking to you and great meeting you, and I know we've had a relationship with you going on at a company level, so it's nice to meet you individually and I really appreciate the time this morning.
Speaker 2:Yeah, absolutely so, john. Thank you so much for being here. I know your time is very valuable, so again, thanks for being on the show today. Great thank you, and to all your listeners out there, I thank you. Time as well, and until the next story.
Speaker 1:Thank you for joining us this week on the Leaders in Payments podcast. Make sure you visit our website at leadersinpaymentscom, where you can subscribe to the show and where you'll find our show notes. If you enjoyed listening, please share on your social channels as well.