PEP Episode 097 — Residuals Rate Hikes & Reality: Inside ISO Economics | Payments Due Diligence Hack That Saves Money

Episode Description

Ever wondered why “free” POS hardware ends up costing so much down the road? In this episode, Christopher Dryden, Esq., sits down with Jessica Casto from Magnolia Ventures to pull back the curtain on modern payments strategy and ISO economics. They unpack how software-led distribution has reshaped card acquiring, why processor-agnostic architecture protects your business leverage, and where subtle contract fine print shifts liability to the least prepared party.

From navigating US card economics—where reward-heavy credit cards drive up merchant costs—to managing the fine line between cash discounting and surcharging compliance, this conversation breaks down state-by-state quirks, debit rules, and tax exclusions. Whether you are an ISV, ISO, or software reseller, this episode provides a clear blueprint to negotiate liability, protect your residual revenue, and build flexible payment infrastructure that scales.

Key Discussion Points

• The Reality of “Free” POS Hardware: How subsidized hardware setups conceal long-term payback periods, pass-through fees, and shifting risks.

• Software-Led Distribution & Acquiring: Why software vendors are chasing lost processing revenue through SaaS add-ons and integrated payments.

• US Card Economics & Surcharging Rules: Navigating reward-heavy credit card costs, state regulations, tax exclusions, and debit network restrictions.

• Critical Contract Terms for ISVs & ISOs: What to look for in Schedule A transparency, onboarding obligations, chargeback support, and residual ownership.

• Portfolio-Wide Basis-Point Hikes: Why CFOs and operators miss hidden fees on merchant statements and how to audit processing costs effectively.

• Productizing Compliance: Building configurable pricing engines that respect card-network caps, handle state-level rules, and notify partners proactively.
  

 

Practical Playbook for Tech Companies & ISVs

• Request Multiple Proposals: Collect multiple processor proposals and compare true effective margins rather than superficial rate splits.

• Engineer Architecture Optionality: Build processor-agnostic infrastructure so you can route transactions, renegotiate terms, or switch providers without rewriting your tech stack.

• Pressure-Test Payback Periods: If subsidizing terminal hardware, validate payback timelines against debit-heavy card mixes and capped surcharge limits before scaling.

• Plan Your Exit Strategy Early: Treat payments as a core strategic capability—negotiate liability and design durable, long-term revenue streams from day one.
  

 

Featured Speakers

• Host: Christopher Dryden, Esq.

• Guest: Jessica Casto — Magnolia Ventures
  

 

Subscribe & Connect

• Subscribe: Follow the podcast for weekly deep dives into payment architecture, ISO economics, and fintech strategy.

• Share: Send this episode to a founder, CFO, or teammate navigating software payments and merchant contracts.

• Review: Leave a review to help other software builders and payment operators discover the show!

 

Legal Disclaimer

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

Transcript

Matthew Luciani (00:00):

It shouldn’t be a surprise to anybody because data, we are in this generation where that is the premium.

Leo Arzumanyan (00:08):

Data is

Matthew Luciani (00:09):

Data is king.

Leo Arzumanyan (00:09):

I was just going to say data is king. Yeah. That’s a well known phrase. And I fully agree with you. I mean, if I had to place my bets, I would say Google’s going to win the AI race for exactly the reason that you just pointed out, which is they have ownership of essentially all of search across the entire world, probably billions of searches a day. All this data that they’re constantly acquiring and they’re able to train their models on.

Jeremy Stock (00:34):

For what, decades now?

Leo Arzumanyan (00:35):

For decades. And there’s never really been a true competitor to their search engine. I mean, you’ve had the, what was it like DuckDuckGo? But I mean, at this point, Google has such a hold over data. So if I really had to bet, I would say Google’s going to win the race.

Jeremy Stock (00:54):

Welcome to the Payments Experts Podcast, a podcast of global legal law firm. We hope you enjoy this episode. We’re really excited. We’ve got an in- studio podcast. Joining us, Associate Attorney Matt Luciani, as well as Leo Arzumanian, Associate Attorney in our transactional department. Gentlemen, we’re talking about a really interesting topic today. Agentix, AI, AI and payments. Jump right in.

Leo Arzumanyan (01:25):

Sure. Me and Matt, we were discussing kind of what topic to go over for the next podcast. We thought what’s better than what everyone across the world is talking about, right? I mean, this is the hottest topic I think out there. So Matt, maybe you want to kind of just guide us into this?

Matthew Luciani (01:40):

Yeah. I think one of the biggest topics though is compliance. Because we kind of went from this evolution of using AI for chatbots, customer service oriented things. Now we’ve gotten AI developed enough to the point where your three real flagship companies, you’ve gotten models, you’ve got … Everybody knows ChatGPT. Claude’s getting increasingly popular. Gemini is very powerful. There’s a few other ones as well. You’ve got Groc and everything too. But a lot of people get confused. What is open source? What is not open source? What works for me as far as a compliance standpoint goes, whether that’s the legal landscape, the payments landscape. Compliance is always an issue, right? We’ve got all of this data. And I think people get afraid.

Leo Arzumanyan (02:33):

When you’re talking about compliance, can you maybe, for the listeners, explain what you mean by that? What regarding compliance do you think is the important issue to be aware of?

Matthew Luciani (02:42):

Privacy.

Leo Arzumanyan (02:42):

Yeah.

Matthew Luciani (02:43):

Right. So I think privacy is number one. Also, from a privacy standpoint, it depends on what field you’re talking about. So if you’re talking about the legal field, privacy, client

Leo Arzumanyan (02:55):

Data. Right. Attorney-client privilege, all those sorts of things. How is that being addressed by the AI model that you’re using?

Matthew Luciani (03:01):

Exactly. How’s it using it? How’s it memorizing it? Is it in a closed environment? Is it in an open environment? Are you letting it learn or not?

Leo Arzumanyan (03:10):

And I would imagine for other fields, for the medical field, for example, right HIPAA, these are all considerations that need to be taken into account when you as the user, whatever professional you are, whatever industry you are in, are inputting sensitive, confidential information.

Matthew Luciani (03:23):

Right. So that kind of brings us to this new terminology that we hear a lot is agentic. So that’s kind of the whole shift from the customer service model to a free thinking model. That’s really what agentic is, is it’s an agent. You’re telling it what to do and it’s working off of its programming, application programming, interface. It’s really like a menu.

Leo Arzumanyan (03:50):

It’s kind of like having an automated assistant, I would say, right? You give it what you needed to do, the tasks, maybe you predefine some instructions, and then it goes out and does what you need it to do.

Matthew Luciani (04:01):

Right. I mean, it’s like a law clerk.

Leo Arzumanyan (04:03):

Yeah,

Matthew Luciani (04:04):

It’s like a law clerk. It’s like a law clerk. So you got to give it the right direction if you want it to return the right item or the right product. In the other space, in payments, some of the sensitive compliance related things that you are more concerned with are your internal compliance rules. You don’t want your trade secrets out there. You don’t want to share your client data from that standpoint, even less from the attorney client standpoint that we deal with, but more from the competitive landscape of it. And what’s becoming more and more common is actually using Agentic AI in things such as CRMs. So when you use that Agentic AI in a CRM, you can use that for various functionalities. So you can import client data a little bit more quickly if you got it trained up, you can import sales preferences, things of that nature.

Leo Arzumanyan (05:06):

And then you can export, compile reports based off the data you give in it, right? Like spreadsheets and all kinds of graphs. Whatever you ask it to do based off the data you give it, it’s going to work with it.

Matthew Luciani (05:15):

Exactly. So I mean, it kind of can help you spot … If you’re truly in a sales environment and these aren’t existing clients, it can help you spot … I guess you could use the terminology pain points, right? When you’re trying to sell something, you’re normally trying to hit a pain point. Where can I help you? Where is your current service provider coming up short?

Matthew Luciani (05:42):

So this can help you identify those things and kind of turn it more from like a cold to a hot lead from that standpoint, kind of put you on a more focused sales path. You can also use it for underwriting. And I think that’s where it’s more and more concerning, I guess, for lack of a better term, for the actual open source versus closed source and the compliance related thing. Because if they’re not an existing client, sure, there’s trade secret value there, right? But if they are an existing client and you’re using your specific underwriting features that you’ve taught this AI, you don’t want it to share that. Yeah.

Leo Arzumanyan (06:25):

I fully agree. And I think I want to touch on something you just mentioned, which is this is where guiding the AI really comes into fruition. You really need to know how to guide your Agentic AI, whatever, your LLM. Whatever you are asking it to do, you need to know how to properly prompt it so that it doesn’t just spit out something that can completely derail whatever you got going on. What I mean by that is a lot of people, we’ve been dealing with this, and we’ve spoken on this podcast before with Jeremy and other clients, people are using AI, but they should take a step back. They are using AI in a field that they may not be quote unquote experts in. So then whatever AI spits out, they’re assuming it’s correct. That’s going to be a huge issue, at least in my opinion.

Leo Arzumanyan (07:14):

And we keep seeing it this morning. Actually, I spoke with one of our founding partners, Chris Dryden, and he mentioned another client used AI and copied and pasted it over in an email to Chris and it was completely wrong. Yeah. But the client thought it was correct. The reason is, if you don’t know the subject matter that you’re dealing with and you don’t know how to properly prompt your LLM model, you’re going to get all sorts of information. Some of it might be true, some of it might be completely hallucinated or incorrect. So I think with what you’re saying, guiding your LLM, guiding your Agentic AI tool is really important and also having the requisite knowledge base to use these tools in the first place. I don’t think people should just go into using these tools and just blindly accepting what it’s saying. And this doesn’t just apply to law.

Leo Arzumanyan (08:03):

I mean, in the medical field, right? Yeah. If you’re using it, I’m sure millions of people are using it for medical questions, medical resources, sure. It might be a good guiding pose, but I personally wouldn’t feel too comfortable with ChatGPT just telling me, “Oh yeah, you’re perfectly fine. You could ignore that. ” And it turns out to be a completely dangerous health hazard.

Jeremy Stock (08:24):

100%.

Leo Arzumanyan (08:25):

Right. So I think one of the biggest components which you touched on earlier is like guidance and really being able to align the AI with your knowledge base.

Matthew Luciani (08:36):

Yeah. And I think that that’s a really good point, especially from the medical standpoint, because when you’re talking about anything scientific, there’s going to be anomalies, right? And one of the things with the AI, especially if you haven’t coded it in a specific nature or you haven’t trained it a specific way, it’s not going to account for that, right? So say you’re using a free model of one of these AIs and you don’t have a lot of previous data, it doesn’t know that you’re an attorney. It doesn’t know that you’re a doctor or a payments expert. These things matter. These things actually matter to, and I would actually say the paid version is always going to be a little bit better than the free version. Agreed.

Leo Arzumanyan (09:20):

I would categorate a lot better actually.

Matthew Luciani (09:22):

Yeah. And I think that goes for almost anything in the world, don’t get me wrong, but a lot of these agentic AIs, truly agentic AIs that are open source, you can still put them on a closed loop. And what that means is that it only stays within your environment. You can teach it based off of what you do. It won’t really limit itself if it knows that you are a certain professional. It will kind of account for those statistical anomalies. I mean, imagine if you’re not a doctor, just for a drastic example, right? Exactly. I mean, think of some really awful form of cancer, not to make it that drastic, but you could Google it, Gemini it. It’s going to come back with a statistic, but that’s not going to take in account that individual and their intangibles-

Leo Arzumanyan (10:23):

Right. All their blood work- All their blood work is done at the end. The doctor knows, unless you’ve trained the AI, which is what you’re saying, which must have inputted it with the requisite context.

Matthew Luciani (10:32):

I mean, there’s people who live with certain medical conditions that if they went to a doctor who wasn’t familiar with them, they’d look at it and be like, “You got to go to the hospital right now.” But if they went to the doctor who was actually familiar with them, they’d look at it and say, “I know that this would be a medical anomaly otherwise, but for you, it’s normal.” So that’s a really drastic example, but it’s sometimes good to make it black and white of how bad it can be if you don’t use it properly. I think the only thing that I would use a free AI for is almost the way that you use Google anyway of- Yeah, quick

Leo Arzumanyan (11:16):

Searches.

Matthew Luciani (11:18):

I wouldn’t use it for actual agentic capabilities of, “I need you to code this or I want you to edit this. ” I wouldn’t really even use it for editing something that I wrote really. I wouldn’t really trust it.

Leo Arzumanyan (11:33):

No, I’m fully with you, and I think this dovetails perfectly into what I wanted to discuss with you next. We’re talking free versions versus paid. I would say we’re some of the more AI forward attorneys in our firm and we’re kind of on top of the latest trends. So I kind of want to get your thoughts on which models you’ve kind of played around with and what you would recommend to our listeners out there who are wondering what are the differences between the models? What is at the top of the leaderboards quote unquote moment? Just to add on, I just saw yesterday, ChatGPT released their latest 5.2 model.

Jeremy Stock (12:09):

Oh, did they really?

Leo Arzumanyan (12:10):

Yeah, and it’s crushing benchmarks. So like two weeks ago, two, three weeks ago, Gemini released their 3.0 and that topped all the benchmarks. And then now just a few weeks later, ChatGPT stopping all the benchmarks. So I think that really points to how quickly things are progressing. I mean, I don’t know if it’s hyperbole to say, but within maybe a year or two, we’re going to be … I think society as a whole is going to be really reevaluating how the workforce operates, how things get done because these models, these companies are advancing so quickly at such a drastic rate that it is just mind boggling in my opinion. Absolutely.

Matthew Luciani (12:50):

I think the one thing to remember related to that is even though Gemini is younger by comparison, I would look at it this way, right? GPT is kind of your OG

Matthew Luciani (13:05):

Per se, right? It really was the revolutionary of this agentic development as opposed to just kind of the chatbots. So they have a tremendous amount of data from that, right? And I would say it’s previous models, it takes in a lot of data, especially if you customize it and you edit your filters and you put it in this micro environment and it’s still very, very helpful from the standpoint of how much information it actually digests and then spits back to you, right? I mean, if I have to read 300 plus pages, yeah, I mean, I’m going to read it, but I’m going to summarize it first, right? Because I kind of need to know the roadmap if I want to effectively get through that, right? Now, what I would use Claude for in the past before that had happened was Claude’s more creative, in my opinion.

Matthew Luciani (14:05):

Now, as I mentioned before that had happened, what I meant is ChatGPT’s updates. I see them pretty close now as far as how much more creative ChatGPT’s new models are. Now the flip side of that, Google is really the revolutionary of the search engine,

Leo Arzumanyan (14:28):

Right?

Matthew Luciani (14:29):

Right. So for them now to enter the AI space and get as far as they have, as quickly as they have, it shouldn’t be a surprise to anybody because data, we are in this generation where that is the premium.

Leo Arzumanyan (14:45):

Data,

Jeremy Stock (14:46):

Data is king.

Leo Arzumanyan (14:46):

I was just going to say data is king. Yeah. That’s a well known phrase. And I fully agree with you. I mean, if I had to place my bets, I would say Google’s going to win the AI race for exactly the reason that you just pointed out, which is they have ownership of essentially all of search across the entire world, probably billions of searches a day, all this data that they’re constantly acquiring and they’re able to train their models on.

Jeremy Stock (15:11):

For what, decades now?

Leo Arzumanyan (15:12):

For decades. And there’s never really been a true competitor to their search engine. I mean, you’ve had the, what was it like DuckDuckGo

Jeremy Stock (15:20):

And-

Leo Arzumanyan (15:21):

Ask Geeves. But I mean, at this point, Google has such a hold over data. So if I really had to bet, I would say Google’s going to win the race, but what I really like is we’re all … Actually, let me preface this. I was going to say we’re all benefiting from this constant competition, but that kind of goes into what I was going to ask you next, which is how do you see the workforce, let’s say five years from now, change because of AI? And I think the downside is, this is my opinion, of course, I do really think that a lot of entry level jobs are going to be replaced in the not so distant future, and that’s because these models and these different functions like Agentic AI, Cloud has all these really interesting coding tools and artifacts and so forth, and we could get into that if we need to.

Leo Arzumanyan (16:12):

But the point being, these models are getting so sophisticated that I think companies are going to start replacing entry level positions and just having their more mid and senior level professionals do more work essentially, right? Manage the AI. Manage the AI instead of managing the entry level systems.

Matthew Luciani (16:33):

And I mean, I could see that being a very realistic possibility. I think almost everybody talks about that as a possibility, but I think that’s also why it’s important to limit it, right? Because you’ve even heard some of the people at the forefront of this on the boards of some of these companies talking about it’s too powerful, it’s growing too quickly. I think the first side of that is don’t be afraid because it’s not going anywhere. You’re better off integrating and learning how it can help you and then limiting it as fit

Matthew Luciani (17:13):

Than just not adopting it at all. But I think that one of the really important things that’s starting to happen, you’re starting to see it is they are specifically trying to limit the use of the layperson using it for medical research, for legal research, right? There are ways to code your AI properly so that it actually knows what you do for a profession and has specific boundaries that it can’t go outside of. And that’s even if it is open sourced and on an open loop, you can kind of put in these features to prevent it from hallucinating or return specific prompts that way you know if something is unverified or if it’s just inferring something. So those are some of the things that I think that the general public will have to learn, right? I mean, I think that almost like everybody jokes about some of the definites in life and taxes being one of them, right?

Matthew Luciani (18:21):

And they’re like, “Oh, why don’t we talk about personal finance growing up and taxes?” It’s like, it’s going to be that same thing, except probably more generally taught because it is happening so quickly.

Leo Arzumanyan (18:38):

Yeah. I think that phrase, what is it, death taxes, there’s something else, right? What is the phrase? If any of us remember, because I think AI is going to be a part of that phrase now.

Jeremy Stock (18:46):

Yes, exactly. Seriously. Yeah. Something like, “I can avoid anything except for death and taxes.”

Leo Arzumanyan (18:51):

Yeah. And one thing I wanted to add onto what Matt was saying, it’s interesting, we didn’t have this on our planned outline for the discussion today, but it was prompted based off what he said and yeah, you can limit your AI models, right? But one thing that I think that’s interesting, and I was just reading again an article about this last night, what do you do if or when AI becomes sentient and can override any limits that you create and starts to code on its own? Because I think I read an article last night regarding, I want to say it was regarding Gemini’s new model where it got really technical and beyond the scope of my understanding of these things, but I guess some researchers released a paper where they were basically saying the new model is now capable of coding beyond your request. It starts to act beyond what the user actually asked it to do on its own, which I think that just opens the door to a whole bunch of other potential issues.

Matthew Luciani (19:52):

And I think that’s another reason why and important to this is, and we’ve talked about this, Claude, for example, it’s free version, it has more customization features than say ChatGPT’s free version. Maybe that’ll change. Maybe they’ll see that as a competitive advantage and maybe they’ll want to try and close that gap. I don’t know, but to your point, say I’m in Claude’s free version and I coded it and I told it it can’t go outside of these boundaries, you have to do all of that stuff before you prompt

Leo Arzumanyan (20:30):

It, right?

Matthew Luciani (20:31):

Yeah. Because if you prompt it without these boundaries, it’s absolutely going to go outside of them. That’s kind of the whole point of- That’s

Jeremy Stock (20:39):

What it does.

Matthew Luciani (20:40):

Yeah. Right. Yeah. Exactly. Especially if you have it open source AI. So sort of the whole point is to put these restraints on it, that way it’s specifically working on what you need it to work on. You kind of got to focus it. We made the joke about the law clerk earlier, but it’s a really good comparison in our field or analogy in our field from the perspective of if we just hired somebody to intern for us for the summer, I’m not going to say, “Oh, I need you to write this motion for summary judgment.”

Leo Arzumanyan (21:17):

Right. And that’s it. And that’s it.

Matthew Luciani (21:19):

I’m no shock.

Leo Arzumanyan (21:21):

You’re going to give that law clerk the parameters, what the law clerk can and can do, what they should focus on, stay within the balance of the rule set that

Matthew Luciani (21:27):

We do. This is the evidence I want to advance. This is the case law I think you should start with. If you go outside of this, let me know. It’s the same concept, right? You have to give it direction or it will go outside. It will freethink. It will do more than you want it

Jeremy Stock (21:45):

To. And it doesn’t know what it doesn’t know to use the law clerk analogy to even know what the mistakes are because it doesn’t know enough yet. Matt, you raised something also maybe not on our agenda, but I think it’s really interesting. I want to pose this thought to you guys because Leo, you said that you see that Google most likely, in your opinion, might end up winning the AI race, if you will. And then Matt, you talked about how some people may want to limit what information is given, for example, medical information to your layman, someone who is not a doctor, not an MD. That scares me, and I’ll tell you why, because I get it. It’s well-intentioned. We want to protect people from doing surgery on themselves, for example. However, whenever there’s … And Google’s famous for this, if they think that you don’t need this information, or Google thinks you’re better off without this information, or kind of the whole big brother issue of Google being like, “I’m going to withhold that information from this set of people because I don’t think they can be trusted with this information.” That gives me the heebie-jeebies.

Leo Arzumanyan (22:58):

Yeah. And I’m fully with you actually, that must have slipped my mind. I wanted to bring that up when Matt was discussing that point. Matt, please clarify, but maybe I thought Matt was saying that the user is going to be the one limiting the AI, but were you thinking more along the lines of the company itself would be limiting?

Matthew Luciani (23:14):

It’s kind of both, right? So specifically when it comes to legal advice and medical advice, ChatGPT has come out recently saying, “We want to limit your ability to use ChatGPT for that feature.” OpenAI has voiced its, I wouldn’t say displeasure, but it’s apprehension over that, right? Because not to go back to the law clerk analogy, but it’s probably the best one that we can use, right? Of like, you need to think of this thing as agency law, right? You’re the principal, it’s your agent. It shouldn’t do anything that you shouldn’t or can’t kind of fact check or backtrack with your own professional due diligence. You shouldn’t be using it for professional tasks if you yourself could not perform those tasks.

Leo Arzumanyan (24:17):

Right. That’s a great point. I think that’s interesting because I’m a big proponent of the free market and I’m with you, Jeremy. That is also an interesting will think about it. I wonder if down the road the open AIs, the clouds, et cetera, would make the users kind of let them know what their career is? Would the profile kind of say, “What do you do for work? If you’re an attorney, okay, you can use this for … ” You have to prove your attorney by uploading your bar or your license. Yeah. These are really interesting things that I have no idea.

Matthew Luciani (24:46):

I mean, I can tell you right now, I haven’t, obviously we are attorneys and I have entered that into my AI. That

Leo Arzumanyan (24:56):

You’re an attorney.

Matthew Luciani (24:57):

Yeah, because then it knows a little bit … I mean, I’ve spent so much time with actually training the model of my AI. That data at this point is all my projects are on closed loops, nothing is shared, like everything’s scrubbed. We have ethical guidelines that we’re supposed to follow and we do that, but you have to actually, or what I did at least, take time, showing it your writing product, show it your style, show it what you actually do, and that’s what’s going to get it customized to you, right? It’s the same concept as what we were talking about of like you’re not going to just give a new hire a motion for summary judgment and not tell them what deposition transcripts to look at. So it’s the same concept in the sense of like, I do think that OpenAI, for example, is the one who’s publicly expressed apprehension on this, right?

Leo Arzumanyan (26:07):

Yeah.

Matthew Luciani (26:08):

If I didn’t have those filters in there, I am curious how much different it would perform, right? Like if I wanted to edit something and it has all of these no data sharing, all of this, you can’t use my data, everything scrubbed, if it didn’t have all of that and I was just using a free version and it didn’t know who I was or anything like that, I am genuinely curious how much different the result would be.

Leo Arzumanyan (26:39):

If I had to place my bets, I think the result would be a lot worse. I think it’d be radically different. That’s where the danger comes in is someone who’s not preempting it with all these rules and these guidance like Matt is, they’re just relying on whatever it outputs with no context, no rule set, and that’s where things get dangerous. And you, Jeremy, I think this is why it’s really interesting topic to discuss because I think back in the Cold War, it was a space race and now we’re in the AI race

Matthew Luciani (27:06):

Because

Leo Arzumanyan (27:07):

You’re hearing China’s advancing the US. I think these are the two countries that have the biggest advantage when it comes to AI, just given the current state of things. But I think we’re just in a really interesting place with this goal, with this whole evolution of AI and how it impacts the workforce, governments, militaries. I just recently read that the US military is going to be using Gemini for a lot of its operations. I don’t know how, but that’s going to be interesting. So yeah, I think we’re in a really interesting space right now.

Jeremy Stock (27:36):

Yeah. I just, to close out that section as you guys move on in your conversation, Matt, everything you were saying, I think it makes perfect sense and I get the importance of having those limits and someone who’s a professional being able to come in and say, “There’s the parameters, et cetera.” I just would be an advocate for you having the ability to do that, not Gemini, not OpenAI, not Google telling you in advance, “No, Matt, there’s already a section of information we’re not going to give you because we don’t think you are okay with it, ” or whatever their reason is, that’s the issue, right?

Matthew Luciani (28:17):

I think that’s fair, right? I mean, I’m all for free enterprise, don’t get me wrong, it’s not about that. It’s more about, to some extent or lack of a better way to explain it, the human aspect of this, right? Because we are human, we do make mistakes. This is AI, but it still makes mistakes,

Matthew Luciani (28:42):

Right? It hallucinates. Right, especially without a human guiding it, right? So I do agree with the apprehension that say OpenAI has presented or both of you guys have just kind of expressed of there should be a certain limiter on it where, yes, we want this to be available, but it should kind of tell you when it’s not, right? If it’s not going to provide you with certain information, I’m all for it giving you a rationale for why it’s not providing you that information, right? Like, “Well, are you trained on X, Y, and Z? If you’re not, then I don’t know if I should be sharing this with you. ” You need to, again, this is all about free thinking, right? We’re talking about it in the AI paradigm, but this is still about the human side of things too.

Leo Arzumanyan (29:42):

And Jeremy just touched on your concern, kind of my concern as well, I think this is where competitive advantage comes into play. Maybe, let’s say OpenAI continues down that path and really starts to close it off and say, “If you’re an attorney, you can use it for legal, but if you’re not, you can’t, or if you’re a doctor.”

Jeremy Stock (29:59):

Right.

Leo Arzumanyan (30:00):

Maybe that’s going to flip the market and everyone starts to go to Gemini or Claude or Grok. Maybe not. So I think that question’s kind of up in the air and I’m with you. I don’t like when these corporations that are already really powerful and really control a lot of areas of our lives get more of a say in these type of matters, especially when it’s related to knowledge base. But I think ultimately it’s going to come down to what the free market decides. If people are fine with it, we’ll see that play out. And if they’re not, we’re going to see that as well. I mean, the good thing is we have so many competitors right now and they’re all advancing on what seems to be a weekly basis where one company tops all the benchmarks and then next week another company tops the benchmark and it just keeps going and going without hitting a wall at least up to this point.

Leo Arzumanyan (30:48):

So I really am curious where this is going to go.

Matthew Luciani (30:50):

I think that related to that, the only thing that is a little bit scary from my point of view, I guess, is like how fast these are advancing versus how fast are the ethical guidelines

Matthew Luciani (31:04):

Advancing. Specifically in our field, right? I mean, we have guidelines that we have to follow, as I mentioned before, like you have to scrub client data, you can’t share this, you can’t share that, you can’t share anything. But on the flip side of that, I’m obviously not a doctor. I mean, what are their ethical guidelines? Another field, what are those ethical guidelines when it comes to the payments landscape? What are your internal compliance guidelines? What are those ethical guidelines? What do those things look like? Because that is going to steer the ship. If you don’t have ethical guidelines for using your AI, then it does become this big conglomerate that’s uncontrollable. And I think that’s probably the most important thing moving forward is, again, this gray area between adoption and apprehension, where I think it’s perfectly fine and natural for us to have some apprehension.

Matthew Luciani (32:10):

And that’s why the death and taxes thing was kind of like, yeah, humorous, but also a good analogy from the standpoint of we’re not going to run away from this. So it’s fair to wonder or be a little scared of it. I mean, I’m scared of taxes. So it is what it is, but once you learn about it, it’s less scary.

Leo Arzumanyan (32:35):

Yeah. No, I agree. And I think this really leads us into the perfect closing segment, I would say, which is, we’re a payments law firm, but really it doesn’t matter if you’re a payments business, if you’re a medical, whatever the case may be, I think this entire discussion should really help you kind of get an idea of how to think about AI in your business. How are you going to use AI? How are you going to implement AI? And I love to go back to this example that I mentioned multiple times on the podcast already today and otherwise, but I think it’s just really important to hammer this point and that’s, you can do everything that Matt mentioned in terms of setting up the rules and guiding it, but really it still is really important to go to a professional if you’re dealing with sensitive matters that are going to impact your bottom line.

Leo Arzumanyan (33:25):

Like I said before, we’ve had a client come to me and Chris and say, “Hey, I drafted this agreement using ChatGPT. Can you please review and let me know if anything needs to be changed?” It was entirely off course. It was like a one and a half page agreement for a thing that really needed to be far more substantive. And we turned it into, I think it was like a 10 to 12 page agreement with all the bells and whistles that we do because we’re experts in this field. We know something that maybe the model is not yet trained on or the user himself or herself doesn’t know what to look out for. So don’t get me wrong, I love AI. Use it as much as you can in your businesses, but I think there’s a point where you can use it in your business, but you should also rely on the experts that know what they’re doing.

Leo Arzumanyan (34:13):

100%. Because you are otherwise going to take your business down a road where you might really end up in significant litigation and that’s what I’m trying to raise like Matt come in. So yeah, that’s just my take on that.

Matthew Luciani (34:26):

Over reliance might be the word, right? Yeah. So I think that one of the important nuances related to that, and we keep using medical and legal terminology from the standpoint of making things a little bit more black and white for the discussion purposes. I

Leo Arzumanyan (34:43):

Think because we have so many obligations with those fields-

Matthew Luciani (34:46):

Exactly. But like think about what you just said for a second, right? We’re payments attorneys. That’s a niche in of itself. I could go to another attorney, they could litigate the case, but they’re not going to have the background that we have. It’s kind of the same thing as like, let’s say, God forbid I’m going through a divorce, right? I’m not going to defend myself because you’re going to have a bias, you’re going to get the rose-colored glasses sometimes. It’s the same concept. You can be an entrepreneur and a business person and you can lead your business so far. You can do almost everything yourself. Some people try to, but what happens when you do that is tunnel vision,

Matthew Luciani (35:29):

Right? You don’t get those outside opinions. You don’t get that professional advisory, right? So like when somebody who doesn’t do this on a regular basis tries to start drafting their own contracts, it’s the same concept. If I went to an attorney who wasn’t a payment attorney and I had them draft my contracts for an ISO, I would probably still need to go to a payments attorney regardless. So it’s being mindful of your own limitations, the AI’s limitations. Again, we kind of said it a bunch of times. It’s agentic, but there’s still a human aspect to this, right? And it is really crossing that or filling in that boundary of the gray area between adoption and apprehension.

Leo Arzumanyan (36:23):

I think what you said is great because it relates to exactly something that happened this morning. Matt’s point about how you need the perspective of those who are experts in the field and you just need a different set of eyes to look at, let’s say the agreement you want reviewed or potential lawsuit, right? This morning, I was going back and forth on emails with Chris where two attorneys who … Chris has been in this field forever. I’m more new to it, but we’re two attorneys with different perspectives and I think our client is getting a great benefit from that because we are analyzing an agreement for the client. Chris is viewing certain provisions one way, and I’m responding to him telling him, “I agree here, but I actually disagree with you on this and this is why.” And we’re kind of really just thinking through all these things.

Leo Arzumanyan (37:08):

If the client didn’t have us as a resource and just said, “I’m just going to go to ChatGPT, have it review the agreement, boom, this wouldn’t come up, this wouldn’t happen.” Meanwhile, here you have two attorneys who do contracts on a daily basis looking through every provision down to the last detail and arguing over it for your benefit. Yeah.

Matthew Luciani (37:26):

And that’s kind of again, where does the buck stop,

Matthew Luciani (37:33):

Right? Where does adoption begin? Where does apprehension end? I mean, another really good example of that and the differences of these nuances and these different fields, regardless of if it’s payments, legal, medical, I mean, I went to a hearing a couple weeks ago where opposing counsel was actually like, “Oh my God, you got a standard of care expert instead of a damages expert. I didn’t know what to do when I saw that designation.” It happens, you know what I mean? And that’s a really good example of the nuances of payments even, right? Of why it’s so important to still have an expert to go to because you can’t just assume that you’re going to adopt ChatGPT firm-wide or company-wide and that everybody’s automatically going to know how to use it properly, right? You still need those people who are going to do what Leo and Chris do and kind of argue with each other about it.

Leo Arzumanyan (38:37):

Yeah. So I think really the takeaway from that and from a lot of our discussion today is use the AI, but know of its limitations, know of your own limitations. And if you’re dealing with matters that require expert insight and expert guidance, I personally wouldn’t rely on just AI. I would seek out that expert. If I have a medical issue, sure, I’ll ask AI to get an idea, but I’m still going to the doctor’s. I’m not just going to skip out on a doctor’s appointment because ChatGPT told me you’re fine.

Jeremy Stock (39:05):

Yeah.

Leo Arzumanyan (39:05):

That’s just my personal-

Jeremy Stock (39:06):

You’re not going to pull out your own tooth or-

Leo Arzumanyan (39:07):

No, no, no. We’re not at that point yet, I don’t think. I might try.

Jeremy Stock (39:10):

Yeah. I do want to make one comment, Matt, you mentioned how, God forbid, you don’t want to ever be divorced. You got to get married first, my brother.

Matthew Luciani (39:19):

No, no. I didn’t say I don’t want to get- Keep that in. Keep trying. No, no. I said I wouldn’t be my own family law attorney. There’s an entire … There’s a lot of moving parts

Jeremy Stock (39:31):

To that one. Thank you for listening to this episode of the Payments Experts Podcast, a podcast of global legal law firm. Visit us online today at globallegallawfirm.com. Matters discussed are all opinions that do not constitute legal advice. All events or likeness to real people and events is a coincidence.