Would you let a software agent, powered by AI, search, select, and complete a purchase on your behalf? Will a merchant be able to tell if there’s a human or robot behind the transaction? What about risk and regulation concerns?

Russ Jones and Drew Edmond spent some time at the Glenbrook water cooler this week to consider the rapidly emerging topic of agentic commerce and agentic payments. Tune in as they explore different industry perspectives in order to understand the implications of this new technology in the payments ecosystem.

 

 

 

Drew Edmond: Hey everybody. I’m Drew Edmond, an Associate Partner at Glenbrook and your host for this episode of Payments on Fire. I’m joined today for a Fanning the Flames conversation with none other than Russ Jones, a Partner at Glenbrook. Glad to be here with you, Russ.

Russ Jones: Drew, it’s great to be here at the digital water cooler to talk about agentic commerce of all things.

Drew Edmond: That’s right. This is where we meet up to talk about our favorite topics in payments. Well, today our conversation is going to explore agentic commerce and agentic payments. This has been a pretty rapidly emerging topic in the world of payments this year, and unsurprisingly, it raises a lot of questions in our minds, I think, and I’m excited to get into what those questions are. Probably more questions than answers at this stage in the topic, but we’ll unpack it a bit.

Before we get to those questions, I want to define the topic to kind of set the foundation for our discussion. Agentic Commerce is referring to essentially a new paradigm where these software agents powered by AI are authorized to search, and select, and most importantly for our conversation, complete purchases on behalf of users. And these agents may reside in chat interfaces, apps, voice assistance, or they may be embedded in commerce environments.

And I think this technology might unlock some patterns of behavior that didn’t necessarily exist previously, right? I think the most common example you hear about right now is purchasing something like travel logistics, flights and hotels. So let’s say I want to go on vacation in the fall. I could spin up my Chat GPT voice assistant and say, Find me two plane tickets for under $1,600, including fees, to a city outside of the US, in October, where the average daily high temperature is above 70 degrees. And find me a four star hotel with a pool, and that hotel’s got to cost, I don’t know, under $500, $600 a day. And then when you find it, you send me an alert with the options.

So the agent’s going to go off, perform its research, and five minutes later or whatever, however long it’s going to take, five minutes later, let’s say, I get a message with three different cities that it’s found for me that meet my criteria. I choose one. And then the agent makes all of the relevant purchases on my behalf using the data that I’ve made available to it and the payment credentials it has access to, or has been provisioned to for this purpose.

As I’ve started reading about those types of use cases, I think it was pretty appealing to me at first, but then I started to think about how many filters I check when I book flights, and how often I like to book maybe two one-ways instead of a round trip. And how I don’t actually like flying certain airlines, and I don’t like going to certain airports.

I like to optimize when I take off. I like to optimize when I land. Sometimes I choose a slightly more expensive flight if it matches certain categories. Sometimes if I’m feeling a little bit fancy, I like to upgrade to a nicer seat, but not always. And sometimes I like to read maybe Reddit or Google reviews about a hotel before I actually choose it, or learn any nuances about the hotel before I book it, or see a picture of the pool first to make sure it’s a nice one, or see where it is on the map and see how long it takes to get there from the airport.

So all of a sudden I’m wondering if I actually want an agent to do all this for me because there’s so many different nuances that if I had to tell an agent that, it would probably be faster if I just did it myself in the first place. Now, you like to remind me, Russ, that my behavior isn’t always indicative of the behavior of all consumers. So maybe we’ll ignore my concerns for now on that particular example, but I think they’re valid. I think there’s this notion of discovery that’s a core component of commerce. And I think at this stage and where AI, where agentic payments is, I’m very curious of the discovery component of it.

But there are multiple perspectives that I think we need to explore with this new technology to really understand the implications in the payments ecosystem. Let’s just jump into it and start unpacking a little bit. I’ve one maybe somewhat extreme example of how agentic commerce may be used by consumers. And if we think about it through the consumer lens, as we talk about in our Glenbrook boot camps, there are several elements that can influence changes in payments behavior. Two of the major elements being, how big of an increase in convenience is it, and is it a big perceived financial gain?

And I think you can make the argument that agentic commerce has the potential to meet both of these in some cases, right? It’s maybe more convenient for me to have an agent find me a few different options for hair dryers that I want to buy, rather than doing the research myself. Or maybe the agent’s going to do a better job of finding a better deal out there, rather than me having to scour the internet and find the lowest price for a particular item, so maybe there’s a financial gain there. But we also talk about how control and safety are key elements in affecting payments behavior.

So how are you thinking about agentic commerce from a consumer perspective?

Russ Jones: You’ve hit on one of the key issues in my mind. I sort of think about the idea of sending off robotic agents to buy things for people, there’s a continuum here, and you hit on a bunch of the things that matter to people.

This is our Glenbrook perspective that we provide in all of our payment boot camps is convenience drives changes in behavior. And financial gain, perceived financial gain. You can be tricked into thinking is good for you when it’s actually not. Those are the big drivers.

And arguably, if you don’t have one of those two, your big idea is a non-starter. Why would someone do something that’s harder than the old way? And why would they do something that doesn’t provide an incentive? The financial gain is simply an example of an incentive. But you hit on these secondary factors that are really intriguing here. Privacy is one. Safety is another one. You have differences in demographics. I think the odds are better that you might do this than me just because of age demographics. I can guarantee you my mother will never do this. So you have those age demographic things.

You also have psychographic things at play when it comes to payments and psychographics is how people feel about things. And the classic examples are, when it comes to money, are you carefree and trusting or are you paranoid out of your mind? One of those two extremes.

In a real simple example, it would be like your bank shows you every month your statement. I get my bank statement for the month of May. Do you reconcile that statement? Some people don’t even look at it. Their feeling is sort of, I kind of follow where my balance is, my bank tracks it for me, I don’t really need to know the specifics. And then other people are like, Okay, let’s get some coffee, let’s go through these purchases one by one by one, let’s make sure my bank’s records reconcile with my records.

And how that plays into agentic commerce, I think, is whether or not, if you’re on the carefree side of the psychographic dial, you’re going to be like, Wow. In your example of the trip with the plane and the hotel, if they find a great deal, why wouldn’t you just want them to book it for you? That would be the carefree sort of psychographic. The other extreme is like, I would never end in million years let some robotic agent buy things on my behalf without me validating first what it is they’re buying.

Because there’s all sorts of, I hope this thing gets it right, but if it doesn’t, I don’t want to be locked into something I don’t really want. So I think there’s, that’s the big jump in my mind is the jump from, I’d like to have a robot that helps me find things and gives me alternatives and does it quickly. That’s going to save me a lot of time. That’s going to be convenient. But the point you jump over to, Oh yeah, and here’s my payment details, feel free to buy it for me.

Drew Edmond: Right.

Russ Jones: That’s kind of a leap of faith that I think is going to be interesting to see how that plays out with consumers.

Drew Edmond: You made me think, when you talk about the age differences because I’ll take it all the way down to like super young kids, right? You see a younger kid that’s kind of grown up and every screen in their life has been a touch screen, right?

So anytime they see a screen, they’re trying to interact with it, with touch screens. And now I think that as AI becomes integrated more in with the youth, if you will, there’s going to be a world in the future probably where kids have these AI tutors to teach them and they’re interacting with AI verbally, and just the fundamental way that they interact with the technology becomes much different than how it exists even today and maybe what will exist for the short term.

And so maybe in the future there’s going to be a world where people are just more comfortable saying, I’m used to just talking to this AI and them doing things for me. And it gets to the point where that trust is already there. But I don’t think where we’re at today, we have that same trust of like, I can send an agent out and they’re going to do exactly the right thing. People are still afraid of hallucinations, afraid of these things going wrong.

So, if it’s a question of what’s happening today versus, what might it look like in five years, 10 years, 20 years, it could be a very different conversation, but we’re only here today and we don’t have a crystal ball.

Russ Jones: Oh, sure we do. Come on, Drew.

Drew Edmond: Okay, fine. We do, we have one.

Russ Jones: We’re payments consultants. For Christ sake, we only need one data point to spot a trend.

Drew Edmond: That’s right. We talked about the consumer side. I think we’ve seen some press releases come out about the topic from the network, card networks, themselves, from non-card networks like PayPal, tokenization businesses, payment service providers.

A lot of these seem to be thinking about this through the consumer lens. They’re trying to enable the payment from the consumer side. Okay, we can allow the chat bot to make this payment, or we’re going to use tokens to do this, or maybe we’ll use stablecoins to do this.

Or, I want to make sure that my payment method is in the mix to make sure that I’m a part of this trend and change. But what about the merchant side? Payments is a two-sided marketplace. How do we get the tens of millions of merchants onboard to make this work? Because if it’s only the buyer’s side that’s out there, it’s going to be a mess on the merchant side.

Russ Jones: Yeah. That’s really true. And if you remember, Drew, the context here, I think, is it’s only been in the last three to six months that agentic commerce has sort of jumped onto the hype curve, if you will, with companies coming out with products and a lot of people thinking about how do I adjust my strategy to take advantage of this idea.

We were chatting about this, Drew, and I, and if you remember, the very first question I ask you is, Is this one-sided or two-sided? Meaning is the vision that’s being articulated at an industry level, is this AI enabled agents going to help people go shopping and buy things for them, the way you described in your example, or is it going to be AI agents negotiating and talking with other AI agents?

Are merchants going to have to be AI enabled? Are they going to have to speak a new type of protocol back and forth to answer questions about inventory levels, discounting, availability, product quality, reviews? And your response was, It’s two sided. It’s obviously two sided.

And I wasn’t all that surprised because I’d read about the agent protocols that are being proposed at an industry level to standardize the way agents talk to agents. And so you translate that in the world of payments. Payments, as you pointed out, is a two-sided world with a buyer and a seller in the world of commerce.

Terminology’s different outside of commerce, but there’s always two counterparties in payments. Two party adoption is probably, I won’t say it’s the hardest, three-party adoption’s harder, but it’s astronomically harder than one sided adoption. Because it’s not enough that you want to do it. Your counterparty has to want to do it as well.

So the way that translates into agentic commerce here is, I want my agent to go shopping for me. I’m really excited. I love it. I want them to buy things. I want them to find the best possible, your criteria was around flight criteria, hotel criteria, and off they go shopping across three different airlines. Think about the other 12 who are not agenetically enabled, right? So you have to have a critical mass on the seller side to make this value proposition work. And on the seller side, they have to see a critical mass on the buyer side for them to invest in this because it’s not going to be cheap, it’s not going to be free, and it’s going to require changes as you’ve been thinking about. Changes in how they do a lot of different things in the back office and how they evaluate risk and risk exposures and things like that.

So the reality here, I think, is that this is going to be a heavier lift. And I’m just talking about the two-sided adoption challenge. We haven’t even spoken about other aspects of this. It’s going to be a heavier lift than the technology itself, I think.

Drew Edmond: I think so too. I think we’re starting to see some, call them attempts, or experiments maybe is a better term. Amazon, for instance, right? You can now go on Amazon and if you’re searching for a particular good and they don’t have it on their marketplace, they can use an AI agent to go off on certain websites and go off and purchase it on your behalf and send that product to you and pass your credentials through and things like that.

Or other folks that are using kind of the merchant of record model to act as an intermediary that probably limits the amount of work that the actual seller of the goods or service will have to do to participate in something like this. But I think those are stopgap measures to a certain extent and to have an actual great customer experience, you’d likely want to see as involved a process as you can to make sure that you’re actually getting what you’re paying for, what you’ve searched for, that you get access to any of the benefits that that merchant may make available to their customers that maybe would be obfuscated in a world where an intermediary sits in between you.

So I think it is early days. I think people are experimenting with what works here, and are probably worried. And part of that is being worried about that scalability issue, trying to reach that final merchant or even just anything beyond the first handful of more sophisticated merchants that have the resources to build this out if they think it makes sense to do so.

Russ Jones: Yeah. And the thing about agentic commerce here, first off, we’re in the world of commerce. We have buyers, we have sellers, we have refunds, returns, we have disputes, we have potentially arbitration. We have criminals. The payments industry, we like to call them fraudsters, but let’s don’t forget they’re criminals.

All those things are at play here. And there’s so many, agentic commerce is lights out, at least the way it’s envisioned. There’s so many visual clues that go on. You’re scouring the web looking for a product, and you find the product, it’s like, Holy cow, I thought this product had been discontinued. I’m so glad I found one last seller who’s going to have this product. And you look at their website, you’re like, you know, this website’s kind of weird.

Drew Edmond: Yeah.

Russ Jones: There’s all sorts of subtle clues that people take as feedback about whether or not the seller is credible.

Drew Edmond: Right.

Russ Jones: And the seller saying they’re credible is not one of them.

Drew Edmond: Not sufficient.

Russ Jones: Providing customer reviews from somebody called Sally. So it’s going to need to be a whole lot of rethinking about how do consumers get comfortable with the seller? How do you get comfortable with product being legitimate and not counterfeit?

And the same thing on the seller side. There’s so many things in the risk management space that are tied to behavioral analytics. Does this seem like a real person, a credible, real person who would be my customer? Having an automated robot buy things at lightning speed kind of takes all those signals off the table from a seller’s point of view.

Drew Edmond: I think that raises up the notion of trust, right, is going to be really, really critical. How do you create trust, especially when it’s two robots talking to each other? We think about this from the payment systems perspective, we saw that Visa unveiled their Intelligent Commerce Network and Mastercard has Mastercard Agent Pay. And I think one thing that stood out to me on the Mastercard announcement was this notion of agent registration. Kind of KYA, know your agent, to a certain extent.

Is this something where we actually trust this bot that’s coming through to make this purchase. And how do we have a framework that’s standardized across the ecosystem? Because if one merchant’s building to this standard and this merchant’s building to another one, the agents have to figure out how to talk to a thousand different protocols on how to interact and verify trust. It’s never going to work. Or at least I don’t think it will. Is agent registration, is that the future of authenticating payments?

Russ Jones: That’s the fascinating thing, I think, and one that’s completely consistent with how card networks think about a lot of things where you have the business side of payments. And we famously in our payments boot camps are always talking about the chain of liability and how it connects all the stakeholders together and assigns liability and the role that the rules play and contracts play, and all that is there.

All that is going to be there in agentic commerce where we’re going to be asking ourselves from a business point of view about who takes liability when an agent makes a mistake. When an agent tries to buy something they weren’t asked to, who’s at fault? Is it you, the consumer, buying something that cannot be returned? Or is the agent, do they have any liability for their actions? That’s on the consumer side, and you can imagine something very similar on the merchant side as well.

You could even go as far as to think, today, the big risk exposure in online commerce is tied to first party fraud or friendly fraud. How do these agents play into friendly fraud? It wasn’t me. It was my agent that did it.

Drew Edmond: Yeah.

Russ Jones: But anyway, my point about certification of agents, that is very consistent with how card networks think about things in that they try to divide the world into business risk and business rules and technology requirements and technology rules, and it’s wholly consistent to require certification and registration of anything that manages card data or touches card data.

If you’re going to be a wallet provider, you’re registered and you’re certified. If you’re going to be building terminals, point of sale devices, you’re registered, you’re certified, at many, many, many different levels. That’s a big part of making sure the overall, at least in the card system, and we’re not talking exclusively about cards here, but at least in the card system, that’s a big part of the integrity in everything is making sure that there’s no rogue components in the system.

Drew Edmond: I think that’s where we’ve seen the card system wrap their arms around this concept a little bit. They’re saying, I’ve got tokens, so we can start with the world of tokens. There could be some biometric provisioning of the agent. If you want to activate this agent, you need to use your face ID or your thumbprint or something like that to tie it to a particular person, maybe. Cards have been the default for a long time, not for all transactions, of course, we’ve alternative payment methods as well.

But do we think the card system is the ideal method of payment? We also see the rise of stablecoins coming up. At the same time we see, we have to consider the PayPals of the world, alternative payment methods in other countries, fast payment systems. How do we think about the payment method aspect?

Russ Jones: That’s exactly right. There’s sort of, I guess two aspects of it. One is, you have what you’re familiar with and that’s a lot of the momentum argument that you would make about consumers have cards, merchants know how to process cards, why don’t we use cards? And certainly that was the thinking in the very, very, very early days of online commerce is the online channel was just a different variation of the call center. A card-based call center. And checks don’t work, cash doesn’t work, ACH is too slow, doesn’t have authorization. Cards kind of fell into be the default for online commerce.

But if we’re going to build, as an industry, we’re going to build a new layer on top of the internet that is built from agent technology, it doesn’t require the active participation of consumers and merchants. It’s agents buying and selling from agents. I don’t know if that card assumption is as solid as what it might be. Just to your point about stablecoin, stablecoins based systems are certainly on the rise and people are taking them seriously about a way to be smart and flexible about payments.

And, at the same time, you highlight the rise of fast payment systems. If I was sort of responsible for a fast payment system, I’d be all over agentic commerce as a big market opening, asking, particularly from a seller point of view, why don’t you want instant non-revocable payments? Did I mention low cost. I forgot that. If that’s a merchant requirement, I forget.

Drew Edmond: Right. No, they love high-cost payments. I think even just the notion of should this be a pull payment versus a push payment is a big question. I think there’s a reason why push payments haven’t necessarily even taken off, especially in the United States for e-commerce transactions, right, because there it comes with a whole host of challenges.

Russ Jones: I was kind of like lamenting that merchants are going to lose a lot of risk signals to evaluate incoming payments. But that’s really only the world of pull payments, right?

Drew Edmond: Right. Exactly.

Russ Jones: In the world of non-revocable push payments, a lot of those issues just fall off the table. Who cares if this doesn’t behaviorally map against how a buyer behaves? The point is you just got paid. Someone just placed an order, you got paid, and it can’t be reversed.

Drew Edmond: But then you got to go back to all the way to the front and say, is any consumer going to use that if they have no protection against what shows up after their robot goes off?

Russ Jones: Yeah, maybe the guarantees come from the robot. Now there’s a smart robot. So Drew, you’re talking earlier about all the technology companies that are jumping into this, and we were sort of just drawing out that there’s more to this agentic commerce idea than just a technology challenge. But one of the things that has been interesting, at least to me, is we’re seeing companies come forward with what they call MCP servers or Model Context Protocol servers.

They’re sort of based on the idea, particularly on the merchant side, and PayPal has introduced one of these, Adyen just last week introduced their MCP server. And it’s based on the idea that merchants are going to want to talk agentic protocols back and forth to agentic buyers. And it’s going to be hard because they don’t normally do that. That’s like a new world So these things in some ways are like gateways. It converts the world of familiar callable APIs that a merchant would be familiar with into the world of AI dialogue back and forth between agents.

It is always fascinating all the places, the idea of the gateway emergence. There’s payment gateways, token gateways, you would think about Plaid as a bank access gateway. And now you’re seeing these products that in some ways are agentic commerce gateways as well.

Drew Edmond: Well, then if I’m an enterprise merchant, do I, especially if I have multiple PSPs, which one do I, what am I using? Am I using all of them? Am I using a single one? Am I using a third party that sits in front of all of my PSPs to enable their ability to speak? it’s definitely an area I need to learn more about myself.

Russ Jones: What do you think a PSP should be doing right now? Specific to agentic commerce.

Drew Edmond: I mean, well, I think one thing is a focus on fraud, certainly. I can’t answer the MCP question right now, but I think the fraud question, it would be a focus. If I have merchants that want to be accepting these types of payments, how do I help make sure that this doesn’t ruin their business with a bunch of bots coming through that they’re not prepared to handle on their side?

Russ Jones: Do you think merchants should look at this as another channel? The way they would think about an online channel, a call center channel, where they think about an agentic channel?

Drew Edmond: I think you have to. I think that you have to think about your back office as a merchant in terms of the data, right? You have to make sure that these transactions are tagged appropriately so that you can understand what’s happening in segmenting that channel specifically to say, Okay, these are the approval rates and decline rates that are happening here, these are the decline reason codes that I’m getting from these particular transactions, this is the number of chargebacks that I’m getting from these transactions.

And understanding what your consumer, what your customer experience is, your, ultimate buyer experience is that’s coming through that channel. Do they use that channel and you never hear from them again because they had a poor experience on however they bought using agentic payments to get that good? And what can you do to make sure that you are communicating with them? Through email, through whatever channels, communication channels,

Russ Jones: Do you think you use email to talk to robots?

Drew Edmond: No, this is for the person that actually bought it, though. They still need to know information about what they bought, right. So there’s still that communication there. But actually maybe, the robots might be crawling your emails to construct the itinerary for all the things they just bought you for your vacation. So there probably is information in there that you need to provide. So I think there’s so many layers to this that are untapped, I think, right now.

Russ Jones: We’ll have to change what a Captcha does. It’s not to screen out the robots, it’s to make sure you properly identify the robots when you’re selling high margin goods to them.

Drew Edmond: Yeah, it should just be a welcome sign, I think at that point. Welcome robots.

Russ Jones: Maybe a final thing I’m interested in, Drew, you spent a lot of time working in the world, working in the subscription economy and the world of recurring payments. How do you think this idea plays out into recurring payments and, and subscriptions in general?

Drew Edmond: Sure. It comes into I think the world of financial management a little bit. Take agentic payments out of it. We’ve been moving in this, towards this world of more control over your subscriptions. We’ve all got a million subscriptions now. Everything that can be recurring, has tried to move in that direction.

And so, now we see things like the ability to pause or cancel your subscriptions through your bank, your bank app, or doing it through a third party that manages your finances for you.

And so you could presume that there’s a world where you have agents that are monitoring your subscriptions for you. Maybe I didn’t catch that email from my video streaming service that they just raised prices again, another few dollars a month. And maybe I’ve configured my financial management through my bank or through some third party to say, I’ve got agents scouring my transactions and if they see that any prices go beyond what I’m expecting, that that service gets canceled. Areas like that where, am I putting more of my financial management in the hands of a service that is following some framework and can go and take actions on my behalf.

That’s a potential area that I could see it maybe coming into play.

Russ Jones: Another area that we really need to touch on here, I think, when we think about the stakeholders and the payment industry, there’s a lot of them, right? And we’ve spent our time in this discussion looking at consumer behavior and trust and whether or not you’re willing to go lights out, letting a robot buy things for you.

We looked at the seller side and merchants and how they might think about this. We talked about the technology and the business side of payments. But what about the world of regulatory requirements and legal considerations? This is a new world, a lot of the things we think about in commerce really are drawn out of regulations, commercial commerce regulations.

I’m wondering about the global trend for strong customer authentication to make sure that the human buyer is deliberately involved in approving every purchase. That’s a legal requirement in Europe. We’re moving that direction in India and Japan. There could be some stumbling blocks there.

Drew Edmond: Yeah, I think it has to be fundamentally reimagined in this world, right? I mean, strong customer authentication is built assuming that the humans, they’re making a choice that they can go, if necessary, and authorize it using different ways if it gets to that point.

If they want to maintain that level of authentication in a world where human is not in the loop for that transaction, it seems to me that that framework breaks down pretty quick.

Russ Jones: So PSD3 in Europe, you know we’re currently on PSD2 that has the strong customer authentication requirement. The regulators are just putting the final touches on PSD3 in the coming year or so. You willing to say PSD4?

Drew Edmond: Four. I’m ready.

Russ Jones: Agentic commerce, PSD4.

Drew Edmond: Yeah. If it grows, then they will have to do something about it because it’s not going to fit into, my guess is it’s not going to fit into the framework that exists today, and any changes that they’re going to make for PSD3. Unless they’ve gotten to it in the past few weeks without comment from everybody that they would want comment from. I think it’s too early, early for them to build that. Right. We need to see more of this in action. We need to see how consumers use it. We need to see the the impact on fraud. That is probably not going to be great given how things typically go with new technology and payments and fraud.

Russ Jones: Well, I think we’ve raised more questions than we’ve answered.

Drew Edmond: I think so. I think we may have to come back and revisit this topic when it matures further.

Russ Jones: Yeah, I don’t think this is the last time we’re going to be talking about agentic commerce, for sure.

Drew Edmond: Well, thank you Russ, for joining me on this one. Looking forward to the next time we get to talk about it.

Russ Jones: Okay. Thanks for joining me, Drew.

Drew Edmond: All right, to all of you listening, thanks for joining us and until next time, keep up the good work and goodbye for now.