1053. Insights: Agentic payments: Revolution or just better orchestration?

0:14Welcome to Fintech Insider Insights from 11FS. I'm Benjamin Ensor and today we're cutting through the hype to talk about agentic payments. Agenting payments are not a new way to pay, but a shift in who makes the decision to pay. So instead of making a payment yourself, you're telling an AI system what you want, and it figures out the rest. It decides when to transact, who to transact with, and potentially even how much to pay. So that's a fundamental change from executing transactions to delegating intent. Now, we've had delegated payments before, subscriptions, direct debits, But those are very predictable.

0:53What's different here is that the decision itself is being done by an AI agent. And that starts to challenge some, maybe many, of the core assumptions that payments are built on, that a human initiates, authenticates, and is ultimately accountable. So what happens when that's no longer true? What does authorization mean? And where does responsibility sit? So that's what we're here to explore in this podcast. And joining me to unpack what's real and what's hype and what comes next in Agentic Payments is a panel of three fantastic guests. So let's meet them. So first up, we have Rory O 'Neill, CMO at Checkout.com.

1:34Rory, it's fantastic to have you with us. Can you give our listeners a quick introduction to you and to Checkout.com, please? Hi there. Great to be on this panel and talking about Agentic Commerce, which the industry has spent a lot of time talking about over recent months, for sure. And we at Checkout, so Checkout is a digital payments business operating globally. We process around 300 billion or more of e-commerce transactions today. We're super excited and believe that agents buying on consumers' behalf will open a world of opportunities to merchants for yet another channel in digital commerce.

2:08And it will offer consumers just a great experience for what they want to delegate to agents as they continue in their commerce world. But I think what we're all working through together and great to get into the discussion with the panelists today is how we innovate and how we build trust in that agentic system so it can operate like the digital commerce that we know and understand and even trust today. Fantastic. Well, welcome. And next, I'm delighted that we are being joined by Sahar Ataripour, an investor and advisor. Welcome to the show. We'd love if you would also tell us a little bit about your background and the kind of investments and firms that you advise.

2:46Yeah. Hi, everyone. Thank you so much for having me. This is a super interesting topic to discuss and there is so much to do still about it. I started my career in consulting post-MBA for near a decade and I got tired of advising people and not actually getting the job done or seeing it through. So I started my own fintech company in 2016, which was, funnily enough, a fully automated AI-driven investment decision-making machine. And I built it and sold it to Allianz Bernstein in 2022. Joined them as a part of the equity hire. I was part of the Hedgeman Solutions team at Allianz Bernstein for two years.

3:29And since I left in 2024, I have started investing myself. So I have created my own portfolio, which is entirely industry agnostic. I've got two fintech companies in my portfolio, but I also advise other financial institutions, including private banks, fintechs and many other businesses at different sizes. and what is generally speaking my interest and what I spend most of my time doing is to look at most of the things that we do at the meta layer from the top and try to see how we have to reimagine and redesign solutions and processes because as you all know today, nothing looks like what it did even a year ago and it's only getting faster and faster and this change is going to be, it's just the beginning of it.

4:21So it's very interesting to discuss agentic commerce and agentic payments as part of that because the part that the technology is ready to do and the part that we can take advantage of and do something with, it's useful and it's doable. There's a huge gap between the two and it's super exciting to talk about what can be done and what are the opportunities. It certainly is. So welcome. And I'm also delighted to welcome Sam Bobojev, founder of Fintech Wrap-Up. Sam, it's great to have you back, especially given your recent deep dives into agentic commerce on Substack and on FinExtra. For people who haven't come across you before, can you give us a little bit of an introduction, please?

5:09Hi, everyone. Hi, Benjamin. My name is Sam. I'm the founder of Fintech Wrap-Up Analytics and Media Platform, also the founder of a wrap-up podcast where I cover developments around payments, agentic commerce, agentic payments, banking and everything around it. So it's great to be on the show. Wonderful. Thank you all three. Okay, well, let's start with the absolute basics for listeners who haven't come across sort of agentic payments before. Sam, perhaps I can start with you. What are agentic payments? How do they work? Well, Benjamin, I would define agent payments as a payment where I do not decide the transaction.

5:51Rather, I define the intent. For example, instead of me choosing the merchant, the amount and timing, I usually set the goals like book me the cheapest flight or buy my favorite coffee. And an agent makes the payment decision within those boundaries, basically. Well, I would say under the hood, nothing changed in terms of rails. It still runs on cards, bank transfers, or real-time payments. What changes is the decision layer. The agent selects when to pay, who to pay, and how to roll the payment. So basically, agent payments are not a new way to move money. They are this shift from user-initiated transaction to basically system-initiated decision-based or predefined intent, I would say.

6:35Thank you. Rory, it sounds simple enough in principle, but there's some complexity behind the scenes to enable it, right? So if you haven't got a person initiating the transaction, right, what changes from your perspective as a PSP? What has to happen differently, if anything, behind the scenes? Yeah, look, I mean, ultimately it's the trust model, right? Like, so when we make payments online, we need to know who the buyer is. We need to know what they're buying. We need to know that we've got money in the account to pay for the goods and services that they buy online. And all of the trust models that we've built since the birth of e-commerce now have to be updated for a very simple principle change that it's not the human anymore or the physical buyer making the transaction.

7:25It's an agent. So as Sam said, it's delegated authority. So instead of know the customer, it's know the agent. How do we identify agents? How do we know what permissions the agents have? And when purchases are made and things need to be refunded, who's liable? How do we manage reverse logistics? How do we manage refunds? How do we know that the agent has permission to buy for a certain amount or a certain price point? All of these questions that we've answered in the trust model of digital commerce now have to be re-engineered and rethought about so that as we move from using AI for discovery into using AI and agents for purchase, we know exactly the identity of the agent, we know what permissions they have so that we can actually conduct commerce.

8:14And that's the bit that the industry is working through, Benjamin. Indeed. Sahar, is there a difference, do we think, between buyers who are consumers, so retail customers, and buyers who are businesses do we think that we might see agentic sort of payments and agentic commerce more widely being adopted faster by businesses or by consumers? Because when we, you know, when people think about agentic commerce, you sort of tend to think in terms of, you know, someone booking a flight and, you know, the agent goes off and finds you lots of different flights. But is that actually maybe where this is going to start?

8:50Or is it actually going to be businesses, you know, doing repeat orders of, you know, supplies or whatever that might move faster? What do you think? I think there are multiple components to agentic commerce and at the end of it, the agentic payments that some of them are simpler in the case of business purchases. they are in some sense a lot closer to a direct debit or a standing order because like for example it's a resupply of something from a list of vetted suppliers and providers or the restrictions and the parameters of this decision could be a lot more well defined and they are generally speaking a lot better documented as opposed to where you talk about a human interacting with an agent and delegating their agency to an AI agent in order to go through the rest of the process, do the browsing, the selection, the negotiations, the payment, and until the delivery of the task that you have asked them to do.

9:56A lot of things are very nuanced. Humans have a very bizarre way of explaining what they want. And generally speaking, they can't even say what they want very explicitly and in words. And there is a very, I would say, thin line between an agent that if you, because I was thinking about the process of what would be the best UX for the human in order to tell the agent in the most accurate way what to go and buy and what are the characteristics of this thing that they want. But the thing is that we think about things sometimes in a lot more simpler ways and sometimes in a lot more, like there are a lot of feelings involved.

10:42There are things that cannot be really put in words. And if the agent goes and then similar to the chain of thoughts that you see in some of the LLM models, it comes back and says, OK, I did this thinking and these are the parameters and the properties of this decision that I have to make that are missing. And asks you all of these questions. It's incredibly overwhelming. and there is a very thin line between burdening the human to a level that they actually decide not to delegate this decision or just getting incomplete decisions. And at every point in time that you are at an intersection of making a decision as an agent, essentially becomes like a coin flip.

11:27Then at the end, the decision that is made may diverge quite hugely from the initial intent of the human. Therefore, I really think that this process of the translation of the intent to what actually needs to be done for humans, for consumers, is a lot more challenging and nuanced. And it's optimizing for a lot more complex problem as opposed to for businesses, it's potentially simpler. And if I can just give one example, when you buy, you mentioned Benjamin flight tickets purchases. Most humans, if you talk to them and you ask them about their criteria for a flight ticket purchase, they know classes of economy, premium economy, business and first.

12:22while there are at least between 10 to 15 subclasses, which are called the fair classes, within each of these categories of classes that people know, generally speaking, that you might have seen them on your ticket, but different letters of NMK, whatever. And those are very much impacting the price of the ticket, the flexibility of the ticket, the rights that you have as a buyer of this ticket, as the owner of this ticket. And all of these things are seen by the agent. And there are a lot of rules that the agent can see and even kind of predict what the next thing is going to be based on the availability of the multiple options around it.

13:02and if it comes back to you and asks you a thousand questions as in okay in terms of priority would you rather have a heavier suitcase or a more flexible ticket that you can give back until an hour before the flight or change it or like I don't think humans would appreciate that so I really think that we should look at the decisions in terms of the complexity and although they may look very, very simple to us now that we make those decisions. They may not be at the end of the day when they have to be delegated. You're making a number of really, really interesting points there. One of which is that, you know, I think we're very used to always thinking that sort of B2B commerce is more complicated because usually it is.

13:45But actually you're saying potentially in this context, actually B2B might actually be simpler because a business can and probably does define what it needs in a way that we as slightly irrational emotional human beings maybe don't. Sam, I'd love to bring you back in on some of the issues that Rory was talking about, about, you know, how you start to sort of think through what needs to change in the industry if you've got sort of robots, agents, sort of initiating payments rather than humans and some of those issues like refunds and so on. Where are we getting to in terms of sort of thinking through sort of protocols, processes, standards for how agentic payments might be handled?

14:30Well, I see it starting basically gradually, but like ultimately it requires a mindset shift. I think users will adopt agentic payments in a low risk, predictable scenarios first, like subscriptions, audio ordering. But the real shift eventually is moving from approving individual transaction to basically approving a system. Well, that makes a decision on behalf of you. Let's say we saw the boost of the Clawbot development, right? Basically, in the future, what I could do is like I could set up the rules for my bot saying like, hey, so these are the rules. You go on based on these rules, you go and make the payments.

15:13But I think we are a bit far from there. But eventually that changes and we will go there and people will start thinking about controls rather than, I mean, making the payment itself or how to make the payment. Rory, there's a really sort of, there's a really exciting future here where, you know, there are merchants that potentially do really, really well out of agentic commerce and everything's hunky-dory and, you know, sales are growing and it's all good. And, you know, merchants are often quite open-minded about new payment systems because they're like, hey, this could boost our volume and this is all good.

15:49But there's also this kind of nightmare scenario where, you know, agents start buying things that the customer didn't want, to Sahar's point, or the customer didn't know they wanted or realized they didn't want. And then you just get this sort of wall of refunds and so on. And it's just a whole world of pain for merchants. Where are we on that spectrum at the moment? Yeah, gosh, that's a really good question. Look, I think there's, I think for every one bona fide use case we can think of for using agents to do good commerce, I think there's likely a hundred fraudulent ones that sit alongside them.

16:23I'm sure Sahar and Sam would be better than I would about inventing some of those, but but that is where we are in the industry right now. And I think, you know, as we talk about protocols, I think, you know, Sam, we've probably got more agent, agentic commerce protocols at the minute than we have for agentic commerce transactions, if we're honest, right? And I think the, you know, it's very early days, let's be honest, right, in terms of what we're trying to do here. But again, I think we can all see where this will go. It's how we build the infrastructure to get the trust in the way that the system has to work.

16:53So, you know, look, since Sam Altman put out his iconic Age of Agents blog in January 2025, we had Visa and Mastercard jump in with, you know, protocols and services for how their rails would take on agentic payments. we've had the big platforms like open ai you know announce how they would do it inside their the chat gp then they'll announce how they'll do it outside of their chat gpt we've had google obviously announce ap2 and ucp and look i think that in broadly speaking what's happening in a minute is you've got the platforms who are putting protocols around orchestration of the purchase and discoverability of the purchase and you've got the the networks and the schemes quite rightly thinking about identity credentials tokenization.

17:37And they're playing a slightly different different levels in those kind of pieces. And I'm sure we'll talk about how this we see this panning out over time. But I think what's what's really important is, we start to put the consumer at the heart of this and we start to put the merchants at the heart of this. And to your point, like for some businesses, the thought of opening up every single SKU that they offer to agents, without any layer of their brand or any layer of their service is incredibly frightening. And that would be completely against their business model. And I think maybe what we can, you know, whilst it's hard to predict the future, and whilst for sure, these are the worst LLM models that we'll ever use, they'll only ever get better over time.

18:22What we can say is if we go back to how e-commerce started, we know that the most important thing for businesses and the most important thing for merchants was to think about their business model and then think about how they could use e-commerce or digital to accelerate it. It's the reason that IKEA had a fundamentally e-commerce proposition to Chanel. It's the reason that Amazon looked very different to eBay because their business model drove their digital commerce strategy. And I think that what needs to happen here, obviously, is that business models need to be thought about first and how that then wants to correlate to agentic commerce strategies.

18:57and quite frankly, one size will not fit all. We will not be doing identity commerce exclusively inside of any AI platform. It will exist across many places across the world. And for merchants, that means they're going to have to work with players that support all these protocols in all these scenarios to support many, many business models. So again, Benjamin, it's definitely not me to predict the future, but I think that we can draw some correlation from the past for sure. Can I add something here to on top of what Rory just said about the merchants who are not going to like this change? On the contrary, there are a lot of merchants who would really like this change because if you think about it, branding is incredibly expensive and very, very sophisticated.

19:44It's a science of its own and it's very, very difficult to, in this competitive market, to stay on top of the mind of the consumer. So if a company with absolutely haphazard branding and not a very well-known name does their job right in order to be identified by the agents with all the characteristics, like I think people with a blue hat, as they say, like accountant mindset. like if you can translate what your the properties of what you're selling the goods or services into an api friendly manner the importance of branding at that point is near zero unless that human has explicitly mentioned a specific brand which means that actually the point of marketing and branding is shifting um through the tunnel to another point i i think that's i think but i think that's exactly the point.

20:39I think that the point of marketing is going to shift. And you're so right because like, look, ironically, words on a page, pictures on a page will matter less. The fundamentals of e-commerce like pricing, availability, logistics, return are going to be more important. And you're quite right. Like at the end of the day, the consumer is going to set the preferences for the agent. So you're still going to have to know the consumer and you're still going to have to know what preferences they set. But your marketing, your point of marketing will change. I think you're absolutely right. Yeah, in my previous life, I used to work in a consulting firm that they also had SEO capabilities and the writing of metadata in order to be scrollable by Google was just like identifying some keywords that describe something.

21:24But just imagine if you can do that on steroids using the techniques that exist today and describe something that is very, very interesting and easily readable for a machine. you essentially, even if you suck at branding and marketing, you may get the chance. Indeed. It seems like if in the past we had to work on our websites to make it more attractive, but now we don't need to even think about our websites, but rather think about the text itself only to attract the bots and AI. So just to circle back then, are we going to need to build a new sort of liability framework, right? So I completely take your point, Rory.

22:05This is a way off. There's a lot of thinking to be done. There's a lot of just people getting used to it. And we'll see a lot of maybe agentic discovery where there's still a human in the loop for payments before we get to sort of agentic payments. But let's say at some point consumers, businesses do start getting comfortable with agents that they've given good guardrails to and so on to buy maybe routine replacement objects and things like that. at what point then do we then need maybe this is to you sam do we need a sort of new liability framework do we need a new way of thinking what happens if there's a mistake right because the agent orders the wrong thing the customer didn't want the thing the agent ordered you know maybe it's the customer's fault right um but right now the merchant tends to get the blame right tends to get the charge back.

22:57Do we need a new framework for that? Well, I think we need it, yes. The current framework, again, assumes, as you said, explicit intent. We believe that there is a single responsible actor. I mean, for example, in my case, if I'm making a purchase, it's me. And also, we also expect that transaction level authorization. But agent payments break this all three. You know, you have continuous authorization in this case and also decision-making change. And there is multiple actors involved because, again, some of those actors is like the developer of the agent. So in my opinion, the system needs to move towards identity and permission boundaries and audit trails as the basis for assigning responsibility.

23:46So if we talk about liability. And I also believe that in the long term, the developers who developed the bot and agent will be more responsible when it comes to liability. And as Rory mentioned in the beginning, we also see networks trying to own the, I would say, identity layer when they say like we will go and basically ensure that the bots or AI agents that are making the payments are trustable or even the developers or the companies that develop these bots could be trusted. And let's see how that plays out. But in my opinion, generally, we will need a new liability framework. moving forward.

24:35We need to wrap up this first section, but to wrap it up, Rory, I love the point you made earlier. There's potentially more sort of protocols and standards being developed than there have been agentic payments at this point. Where do you think we might see, and I take your point about predicting the future being very hard, where do you think we might see agentic payments starting to be embraced first? Can you see any sectors, any countries, any transaction types where you think, yeah, this is perhaps where it's best suited so far. Yeah, definitely. And look, we've actually spent a bit of time, um, you know, running survey with consumers.

25:12We've surveyed around 12 ,000 consumers across the globe and around 800 merchants on exactly that question, Benjamin, like, you know, where do they see agentic commerce actually getting traction? Um, what are the barriers and, you know, effectively what types of transactions? And some of it is this is like, you know, consumers will admit like 50%, more than 50 % actually will commit using AI to discover products today. But 48 % of them are absolutely categoric that they are very nervous or feel unsafe about using full agents to conduct commerce or using agentic payments. And the top three barriers, of course, are how do they know it's me?

25:50How do we manage returns? And how do I set the limits? So that gives us a really good indication for what's on the mind of a consumer and what the industry has to work together to address in its liability models and all those kind of pieces. So, you know, that being said, outside of the research, you know, we've seen that, you know, there's a whole host of like transactions that people will want to do. More often than not, people will say that they'll use agentic commerce so that they don't spend too much, actually, so they control their budgets. But we see it really being used for purchases that humans really don't want to make?

26:31And the answer to that question is different for all four of us that are on this podcast today. Some of us will enjoy making some purchases. Others will hate making the same purchases. So that's what's going to happen if we were to predict the future is that consumers are always going to use this type of technology to make payments for the stuff they don't want to make. And it's fascinating to see in the research that like more than a third of us would use agentic commerce to buy for a loved one because we trust the agents to get it right, but we won't use them to buy for colleagues because we're worried that they'll get it wrong.

27:02So maybe that gives you an indication of the type of psyche with inside the consumers and their adoption of agentic commerce. Wonderful. That's a fantastic point. All right, we're going to take a very quick break here and then we're going to come back and explore where the value in agentic payments sits.

27:22This episode is brought to you by Indeed. Stop waiting around for the perfect candidate. Instead, use Indeed Sponsored Jobs to find the right people with the right skills fast. It's a simple way to make sure your listing is the first candidate C. According to Indeed data, Sponsored Jobs have four times more applicants than non-sponsored jobs. So go build your dream team today with Indeed. Get a$75 Sponsored Job Credit at Indeed.com slash podcast. Terms and conditions apply.

28:18And participate in McDonald's while supplies last.