PEP Episode 060 — AI in Payments: Game-Changer or Legal Nightmare? | The Truth About AI in Merchant Services

AI in Payments: Innovation, Illusions, and Legal Risk

Artificial intelligence is transforming the financial world—and the payments industry is no exception. In this episode of The Payments Experts Podcast, produced by Global Legal Law Firm, we examine both the breakthrough opportunities and critical blind spots AI presents for payment professionals.

The Hidden Risks of AI-Drafted Contracts

The discussion opens with a cautionary trend: merchants and agents using generative AI tools to draft payment processing agreements without legal oversight. While these contracts may appear polished, they often miss vital industry-specific nuances—especially in areas like residual compensation, liability allocation, and ISO-agent structure. The result? Agreements that fail to protect the interests of key parties and create unnecessary exposure to disputes and financial loss.

As the legal team points out, the payments ecosystem is highly specialized, and today’s AI tools lack the contextual understanding to navigate these complexities. Relying on AI for contract generation without expert review is a growing and dangerous misstep.

Where AI Shows Real Promise in Payment Operations

Shifting to the upside, we explore how AI is being leveraged by advanced payment organizations to enhance underwriting, risk modeling, and fraud detection. We also delve into AI’s potential role in residual reporting, highlighting a real-world case where an ISO transitioned from a transparent system to a limited reporting platform—coinciding with agent residuals dropping by 10 to 25 percent. This example underscores the opportunity for AI to bring much-needed transparency and accountability to agent compensation.

Understanding AI’s Limits and Legal Implications

Perhaps the most important insight from this conversation is understanding what AI can’t yet do. We address the well-documented issue of AI hallucination—where systems confidently generate false or misleading content. In an industry governed by compliance, regulation, and contractual accuracy, this presents a significant risk.

As one host notes, “AI is just a word predictor that often hallucinates to please you.” That sobering reminder drives home the central thesis of this episode: AI is a powerful tool, but not a replacement for domain expertise and legal due diligence.

Who Should Listen

Whether you’re an ISO building tech-enabled solutions, a merchant deploying AI across operations, or a payments attorney advising clients on risk management, this episode offers critical guidance for navigating AI’s growing role in our industry—intelligently, responsibly, and profitably.

Have you encountered AI-generated errors in your payment documents or agreements? The team would love to hear your experience.

*Matters discussed are all opinions and do not constitute legal advice. All events or likeness to real people and events is a coincidence.*

Leo Arzumanyan (00:00):

How about we touch on the future now, which I think is AI and how AI is going to intersect with the payments industry, maybe things we’re seeing or that we’re hearing about, because whether it’s Chad, GBT or open ai, whether it’s grok from X Gemini, from Google, these companies on a daily basis are improving their technology and people are using them. So one thing I’ve been seeing, actually, two things I’ve been seeing that’s very interesting in our industry and how it intersects with AI on the client side. I’ve seen examples of our clients clearly use AI to draft an agreement, and they come to us and say, Hey, I drafted this. Can you take a quick look? And again, the same exact issue because they don’t know what should be in the agreement, and they don’t know how to properly use ai. It’s a horrible agreement and it’s not an agreement that would in any way protect their interests. So that’s one danger of AI that I’m seeing.

Christopher Dryden (00:54):

And the other thing is that we are in such a unique industry. You cannot take a general independent contractor agreement and think that you’ve actually gotten what you need from it to have, it could be enforceable in certain ways, but it’s going to have a lot of ambiguity and it’s not going to speak to the detail of these relationships

Leo Arzumanyan (01:14):

That we’re talking. Yeah, there’s so many levels in what we do, as you can see from everything we talked about so far today that

Christopher Dryden (01:18):

We haven’t even talked about software.

Leo Arzumanyan (01:20):

Yeah,

Christopher Dryden (01:20):

Exactly. Right. I mean, I don’t even want to get into that on this, but there’s a lot of things that the agent could be selling here,

Leo Arzumanyan (01:25):

Right? Agreements are not just agreements. There’s a lot that goes into it. So one side of the coin that I’ve been seeing is clients are using AI and thinking they can just get away with it. And you can’t. This industry is far too specialized. There’s too much behind the scenes that a general language model cannot at least yet encompass. And then the other side of things that I want to talk to you about is where do you see the payments industry as a whole going forward when it comes to ai? How are they using it? Just for example, I’ve been reading articles where firms are using AI for underwriting, for credit review and things of that nature. So have you seen anything that’s come across your radar?

Christopher Dryden (02:05):

So it’s difficult. I think if you build a formula and you teach the AI what the criteria are, I think it could be a really good tool. But that’s more for transaction risk management, underwriting somebody from emergent account maybe to prevent fraud, maybe to prevent some kind of chargeback, automatic detection. The problem is, is that it doesn’t interface with public databases to understand, like here for underwriting, right? I’m going to go and I’m going to try to underwrite somebody for an account related to a business. Unless you’ve shoved into the AI portal, something about the business, potentially the public database for the formation of the LLC, how much have you scraped the internet and public databases to shove in what you need to be able to actually spit out with what you have on an application is accurate to look at it and utilize it. Now, I think for sophisticated ISOs, FSPs and upstream vendors from even that, I think they’re probably getting into this.

(03:17):

I think from where we look at things, what the agents really need to do is work with somebody who understands residual reporting and how revenue and expense is allocated to transactions to put that into a system that can then take data. I mean, we have this dispute going on right now. We’re on the other side of an ISO that went from priority to IRIS when they were with priority. Priority uses a reporting system called mx iso. Mx ISO is very detailed. It has a lot of data fields that on the residual report don’t even populate because some merchants don’t use them, but they’re there because it’s a comprehensive reporting system. So back in the day, some of our clients, when they worked for this iso, when they got the reporting, they may not have understood it, but they got complete reporting about how the residuals were calculated For the rev share, this ISO went from MX iso, and now they have that same data put into iris.

(04:28):

Iris is an agent management system for really working with your agents and doing reporting, and it’s got all sorts of functionality. But for agent reporting, it’s now taken a residual report with a hundred fields of data, shoved it into seven fields of data, and you now don’t know from the information going in how it unpacks from while it’s being shown to you. The other thing about IRIS is, is that you can manipulate the data and exclude some fields. So you could exclude some fields of revenue, include some fields of expense that shouldn’t be there, and now it’s just this nebulous profit expense or revenue expense to get profit. And when they switched over, almost every single one of the agents had a reduction in residual revenue between 10 to 25%. And when the agents went to ask ’em about it, they were like, oh, well, we were paying you out of the wrong column, blah, blah, blah, blah, blah, blah, blah. So there’s manipulation that goes on with this data. That’s why the reporting’s really important. But I think if you were able to, I think somebody who does consulting in this industry that has knowledge about how residual reporting from a very detailed level and the Schedule A can be reconciled with one another, I think AI could be used for that to make it so that it’s less manual and laborious for the consultant so that they can offer the service to agents.

(06:02):

And I think it would legitimize the industry. I mean, look, the one thing that I’ve seen about this industry is that as it relates to ISO agent relationship, there’s less abuse of the agent on a widespread scale. I think agents have become more sophisticated, and I think ISOs have been willing to create better partnerships with agents. Now, there still are the outliers, like this ISO that I was mentioning that get people that aren’t familiar with the industry promise ’em a whole bunch of stuff. They get people who aren’t used to making the amount of money that they’re now making, and they don’t even question what they’re being paid because they’re just like, well, this is better than it was. And then when they do question, that’s when they find a way to terminate you. I feel like those players are less and less in the system because technology is legitimizing what’s happening and there’s good relationships to be had. But I think AI could really help the agents from that perspective.

Leo Arzumanyan (06:59):

I think those are all great points. One thing I would issue, kind a word of caution. I’m a bit of, I guess, a nerd when it comes to the AI topic. I’m really interested in it. I’m always listening to podcasts on the lead developments in ai. I know pretty much everything that’s going on with Gemini, with grog, with Chad, GBT, with cloud, which is from Anthropic. So I know what these models are capable of and what they’re doing. No matter how sophisticated you are, sophisticated FSP, iso, whatever the case may be, you still have to be very careful because one issue that I don’t particularly foresee going away for a long time is the issue of hallucination when it comes to ai. That’s an actual term that describes how these large language models operate, which is very frequently, and this has been in the news and across all kinds of industries, the AI model will hallucinate and output things that are blatantly untrue. But because it’s written in a very intelligent way sound true, or it might reference a statue or a law or a case that does not exist. It

Christopher Dryden (08:03):

Exists. Yeah. I mean, this happens. We’ve seen this where attorneys are using AI and they’re citing cases that don’t even exist. But I know it happened early on with Michael Cohen where he put that in front of a judge. But all the time now, we still see it happen all the time. All the time. There’s sites that we look at, they don’t even exist. And from our angle, this is attorneys and law firms trying to expedite and provide a product faster, better they think, but they don’t really see that it’s,

Leo Arzumanyan (08:41):

They don’t see and they don’t understand the

Christopher Dryden (08:43):

Technology.

Leo Arzumanyan (08:44):

Because AI is a large language model, really, if you want to boil down to it for everyone to understand, it’s a word predictor. The model just predicts the next word in a sequence of words based off all the information out there. Oftentimes it hallucinates the next words because it wants to please you in a way. So that’s how I would sign off on this podcast, is I would just be issuing this cautionary tale of be careful. Don’t just rely on AI to get your work done quicker. Come to people who are professionals who know this industry, the back of their hand, and people who use AI know AI and know what they’re talking about.

Christopher Dryden (09:22):

Yeah, I mean AI for dating apps, sure. But nothing like too serious. Yeah.