Agentic Commerce: What’s Real and What’s Still a Demo - Sam Boboev, Fintech Wrap Up

Speaker A: Hey there. Hello everyone and welcome to another episode of the Leaders in Payments and Fintech the podcast brought to you by, um, Edgar Dunn and company. I'm Beatriz, I'm a senior consultant for edc. And in this series we're talking to the people that are driving the real innovation in payments and financial services. Um, from embedded finance to AI agents to real time payments, the regulatory shifts, uh, our guests are turning big ideas into real world impact. If you're enjoying the series, be sure to follow us on Spotify, Apple or wherever you get your podcasts. Let's dive in. So today I'm joined by Sam Boboev who is a true fintech entertainment enthusiast. Sam, could you tell us a bit about yourself and what you do?

Speaker B: Hi everyone and hi Beatrice. Thank you for inviting me. I'm Sam. I'm the founder of a Fintech Wrap Up. It's an analytics and media platform, uh, where I just basically write about fintech payments and everything that's happening around it. I'm also, um, the host of Wrap up podcast where I speak to, I mean founders and leaders in payments, fintech and technology. And in the past, uh, my background was in product management, uh, I had chance to build payment, uh, wallet and condition products across the globe.

Speaker A: It's really nice to meet you Sam, and we're very curious to hear your perspective today. So today we're dealing with a topic that, at least for me personally is very exciting. We've been hearing a lot of buzz around it and that's agentic commerce. Um, so because we've heard a lot of announcements from the Shemes, from the AI platforms, from the PSPs on this, we certainly heard, I think, more um, of a narrative that I think is not really fully aligned with reality of what's actually practically there, what's real and what is happening. So we wanted to just take a step back and uh, deal with what is actually currently not working and deal with whether the current infrastructure can actually support agentic commerce and agentic payments, uh, currently. So if we just look at the current environment, Sam, um, how would you describe agentic, uh, commerce currently and what is, you'd say real today versus what is more of a demo, right?

Speaker B: Um, that's actually a great question I think, because if I strip it down, what actually is working today is agents are helping us to get to the point of purchase faster so they can help us to do the search, to compare the pricing, show you the images and fill out the forms. The best part, you know, you don't have to do those things manually anymore. But I think where it comes uh, a bit show or when people say like the agent will go off and complete basically the whole transaction journey for you? I think that's a problem. We are not definitely there yet, uh, because at that stage things start breaking very quickly. The problem isn't that the agent can't decide. It's like ah, everything around it messy. We have inconsistent information, everything is unclear. And if something goes wrong we still don't know who is liable and what we should do and how we should behave or how agents should behave. That's why the part when um, agents go and do autonomous transactions is still like under the question. That's why I would say um, getting to the point of uh, a transaction is real, but the rest is still a bit of a demo and everyone is trying to get it right. So I think there is a work to be done there.

Speaker A: Would you say that that's the biggest unsolved issue as of now, the getting to the transaction part, the fully autonomous payments?

Speaker B: I think for me it comes to one thing, making sure that actually the agent will do um. What I meant not what just I said, you know, because these two things are very different because I might say like uh, get me a good flight. But what does it mean? A good flight could be a good air, uh, service provider, or is it the ticket or the length of the flight? What does it mean actually? So I have to explain everything to my agent and unless I translate that into something that very precise that agents can make a decision on, um, agents will uh, like it uh, will not be able to execute or will execute incorrectly. So that one feels very wrong to me. And so the real challenge is like turning the intent, the human intent was always a bit unclear I would say because sometimes I write to my chatgpt, say like do this and that's it. I don't give the instructions. So it starts hallucinating. So I mean getting the intent correctly and executing on that is the biggest problem at the moment.

Speaker A: Yeah, and I think it's also a matter of um, is it tailored to you, your search history, your buying patterns, or is it a matter of what the particular airlines may have displayed in their product catalog as uh, flights.

Speaker B: Exactly. That's absolutely true. And also I mean to the point which you mentioned, like your um, browser history, what if you use a different phone? Right now you don't have any history so it becomes another problem. Right. So, so.

Speaker A: Or what if your account gets accessed by multiple people?

Speaker B: Exactly. Then what do you do? So these are still open questions actually.

Speaker A: Yeah. So you've mentioned uh, flights as one of the verticals or travel in general. Um, I think a key question we're looking at is what are some of the key verticals where we will see the biggest impact and adoption in the near term.

Speaker B: So I think um, the first place is the really works where decisions are already quite structured as you mentioned. Like one is flight because usually you more or less know uh, what you want, where you are going, your um, budget, um, destination, things like this. So more or less you know that. So it could be convenient to use agents there. And the second one is E Commerce. Ah it also works but not very precisely. I mean it works if you are buying something consistently. You know if I'm buying regularly things over the months or weeks or days then yes agents can do that. But uh, honestly in my opinion I think the B2B payments are the best positioned uh, for Agent uh to be impacted first. Why? Because you know when you are doing B2B payments usually you are like part of the company and you know that there are rules. You cannot just go and buy anything. Company tells you what policies you have to follow, what your budget is and what you can do and what you cannot do. And those areas where I think uh, agent commercial or agent payments will take off really quickly. Uh, and we already see examples of ramp for example starting working on those types of solutions. It's US based expense management platform or Brex. Ah another one. They're also working in the directions to basically help companies um, to optimize their workflows, their payments and things like this travel as well. Ah using AI agents within their platform actually yeah.

Speaker A: I think it's also maybe a matter of how we define agent E Commerce. Do we mean the discovery part or do we mean the purchase part? If we look at the agenting discovery part I think one of the biggest impacts may be seen in travel just because there's a lot of decision fatigue with researching, browsing prices, hotels, different flight options and if you can just uh, outsource that into the agent, just narrowing it down for you and then you yourself go to the merchant webpage to complete the purchase that already saves quite a bit of time. Then if we're referring to the agentic payments part as you've mentioned, the non emotional repeatable purchases like your weekly grocery shopping, beauty products, um, just simple like toilet roll, whatever you are even subscribing to on Amazon to get delivered every, every month. These are a bit of no brainers. Um and as you've mentioned as well. For B2B use case we can say for example for digital goods that's a very very good use case. Um, and it's also because these are very much impacted by commercials. Not so much about making the decision about choosing a particular brand or um, anything that would involve like the human aspect for decision making.

Speaker B: Yeah, yeah.

Speaker A: Okay. So now that we've looked at the current environment let's dig deeper into maybe the um, architecture, the infrastructure and whether that can support um, agent E commerce. So do you believe that the current payment identity and fraud systems are fundamentally incompatible with agent led transactions right now?

Speaker B: Well I wouldn't say they are incompatible. They, they are rather built for humans. You know, ah, until very recently like we were making decisions, you know, we had our patterns, how we move our um, mouse and what we, how we type and things like this. But uh, new systems are basically now needs to adapt to AI agents because they um, behave differently when you introduce an agent. Like those assumptions previously were built for humans. They do not work now something is like acting on behalf of you and the system doesn't fully understand it. You know even fraud systems get confused because agents like behave really fast. They go through the things very quickly. Sometimes they're very clean work really good. And like fraud systems is this really human or agent and they will start triggering it. But you know there are good agents that are bad agents now agents are most probably good. They're working on behalf of you so you need to allow them to do what they want to do. So the systems like don't need to be replaced. I mean the current systems, they simply need to be adapted for delegation, economy delegation, um, I would say payment, delegated payments where now agents basically behave or act on behalf of you.

Speaker A: Actually I think you've touched on a very interesting topic because we get on a lot of websites that captcha to prove that you're not a bot and now actually the system has to adjust to know your agent and authenticate the agent.

Speaker B: Exactly.

Speaker A: And actually allow the website to be accessed by bots.

Speaker B: That's very interesting use case that you mentioned and I'm really curious how we will um, solve these types of issues because everyone is trying to rush to solve these uh, problems but I don't think we have a clear solution just yet in my opinion.

Speaker A: Yeah, and unlike humans with bots you can have thousands and thousands. Right. So how do you look at every single one without your traffic being impacted, Exploding?

Speaker B: Yeah, yeah, yeah.

Speaker A: Let's also dig deeper into the card rails, um, and the existing checkout logic, um, do you think that that is adaptable enough right now?

Speaker B: I think some parts of it, yes. Some parts of it, no. Uh, what I mean is that what's missing is a layer that sits like, between, I, um, would say what agent is allowed to do and what's not. So we don't have that layer right now. It's, uh, very binary. Either you approve something or you don't. But with the agents, you want to say things like, you can spend this amount in this category under this conversation, under this condition, and that layer does not exist yet in a clean way. And I think that's what needs to be built when it comes to the layers. And I think MasterCard and Visa, they are both working in that direction with their new, um, products and solutions. Basically, in my opinion, that. Exactly the companies, or I would say they want to become the person in the middle. If you look in the current Rails, they had basically the, the person in the middle. You know, they're usually trying to manage everything. They want to become the same layer when it comes to agent commerce. And let's see how that goes. I'm really curious myself.

Speaker A: Yeah. So as you mentioned, we've seen standards, we've seen protocols launched by various players. At some point, it felt like every single week there was a new protocol emerging. So are there any new infrastructure and standards that do you think are required for AI agents to transact reliably across this fragmented ecosystem?

Speaker B: Um, I think it's going back to the point of, like, the Internet is basically built for humans, not the machines, uh, in the first place. Um, for, um, agents, things needs to be clear and structured, unlike the human, because we, like with our eyes, we decide with our eyes, you know, by looking at it. But for agents, images does, uh, not have any meaning. So what they want is like, product description, the pricing, is it available, what are the conditions and things like this. So you now have to optimize basically your website or your platform for agents. And, uh, the thing is the agent ends up making decisions on like, basically what you are providing. So unlike humans, which can sometimes rely on assumptions like, or intuition, like saying like, okay, I can still try this. Agents won't behave like that. They will, like, rely on whatever you provide. If you provide something good, you are winning. If not, then you are losing basically in this competition.

Speaker A: Yeah, I think, um, at least from our perspective, one aspect that was creating a lot of confusion was for clients with which protocols to integrate with. They were thinking, shall I integrate with multiple ones? Which one is going to stick. Shall I put all my eggs in one basket? Shall I wait until there will be an orchestrator, um, that will integrate with multiple protocols and work with that as a partner? And that has been creating quite a lot of confusion.

Speaker B: So I think Stripe and Google are trying to sort of create that orchestration layer, uh, because again, as you said, not every merchant is able to go and integrate with everyone. It's first of all costly. Yeah. And uh, it requires, uh, knowledge. Maybe you cannot do it, you have to go work with another company. So a lot of things, I think Stripe and Google are trying to somehow come up with solutions to solve these issues. And I'm expecting, expecting more orchestrators to jump into this, uh, let's say wagon to help merchants basically go and just choose between the, um, let's say protocols and start using it with a few clicks.

Speaker A: Yeah. And um, I think it's nice to see all the industry coming together to sort of solve for this use case. But do you think there's new developments that need to be made and would be best positioned to create those developments?

Speaker B: Going back to the, again to, uh, my previous, uh, idea, like, we need trust layer. First of all, uh, we need to understand who, uh, will be the person or who will be responsible if agents make a mistake. In my opinion, it will not be like one company or it will not be like the issuer or the acquirer or the merchant. I think it will be shared between those, all those players all together. Um, once we solve this issue, um, I think there will be more clarity how we want to move forward. And I'm really curious if that will be one company like Visa or MasterCard or if that will be a government or some government body that will set out the rules. I'm really curious. So in my opinion, like, we need to understand, uh, the trust layer to get the trust layer right. And after that we can like, make a significant movement in agency commerce, in my opinion.

Speaker A: It's really great you mentioned the trust layer, because that was my next question. So if I'm Beatrice and I ask my ChatGPT to buy a shirt and the transaction goes wrong, for whatever reason, the, uh, product catalog is misaligned, the pricing is wrong, it's not the color that I requested. Who owns that failure, who is liable?

Speaker B: The question is still open, you know, nobody knows, honestly, because again, in the past, um, for example, I attended the event by Nvidia and aws and the same question was open. We still do not know who owns, ah, like who is liable? So it's again in my opinion it will not be one party that is liable for that for that mistake. It will be like across the system players that's involved in this uh, process of going, searching, making transaction and completing the transaction. So we will have to divide that liability across them, depending on the situation in the case and then decide. But at this moment it's still open. End question. We do not know who is exactly liable if the agent makes a mistake. That's why I think we are using agents in the cases where like ah, for example B2B payments when you already know what you are buying, for example like subscriptions or small ticket uh, purchases that merchants are not scared of or like the issuers are not scared of. Okay, that's small. We can like refund it in case something goes wrong. But again uh, still it's open question in my opinion.

Speaker A: Yeah, I think it's also very bound to the context. First of all, what is it? Is it a charge? But is it friendly fraud? Is it uh, fraudulent storefront? And then also who is the merchant of records? Will the merchant maintain its role as the merchant of record and handle the funds? Or will we see AI agents taking on that role or partnering potentially with the um, merchant of record service provider?

Speaker B: I think agents are far from that at the moment. Uh, maybe sometime in the future they will be like a merchant of record. But not at the moment. I think at the moment, um, let's say agents will still have to partner up with providers in the market. For example, I saw one such example is when Lovable, one of the biggest companies uh, in Europe partnered up with Paddle merchant of Airport company. So they made integration to help the merchants actually that are building on let's say allowable to handle things like uh, tax refunds, disputes and things like this in one place. At the moment it's too much information and it's too complex for agents to handle that. And especially if that happens across the regions, it will be ah, like even more difficult for them to go and uh, take care of this stuff. That's why in my opinion at the moment it's more of a partnership than going and doing these things uh, independently.

Speaker A: Yeah, we also came across um, Padel and Clever Bridge and uh, it's certainly one of the use cases where we're looking at maybe more on the long term because we still think that uh, merchants themselves would preserve the role of merchant of record. And the appetite is not as strong for AI agents to move into that very tricky space. But at least from my perspective it does make sense for certain use cases. So first of all if it's a closed um, AI ecosystem, like if it's a super ah, app, if it's uh, Amazon, then it makes sense also for let's say low ticket purchase values, digital goods, um, or low liability products.

Speaker B: Right, right. No, that could be true. And again maybe step by step we'll start um, delegating this stuff to AI agents as you said. Like for example with Amazon closed loop, maybe they will not like give everything at once but step, uh, by step we'll move in that direction. In the long term that's possible. I cannot predict of course, but that's possible that AI will start handling all of that.

Speaker A: Yeah. So if we look in the long term, let's say agent E Commerce suddenly accelerates, who do you think is the most exposed to that change? Is it the issuers, the merchants, the networks, the payment service providers?

Speaker B: Well I think I would divide this basically into two categories. It's very hard because again there are too many players. I would rather divide into um, smaller players, startups that are coming up in the market and I would say that's a bigger players that already exist in the market. For example, I will take the example of Visa and MasterCard, um, I think the both will be affected. Visa and MasterCard for example, they're in a very good position to become a trust layer. Ah, they already have strong network, strong access to data and resource of course to build. However the problem with that could be that they move slow, you know because again they are way too big and it's very risky for them to start implementing everything at once. However um, startup companies, be it a short acquired, doesn't matter, they are uh, they tend to move faster. They will start building as an AI native companies, everything on AI, uh platforms. So they tend to move faster and to grab the market where they see potential and opportunities. Of course sport, again partnering up with bigger players because you cannot go without three players in the market. So I think there is an opportunity for both small players and big players when it comes to the future.

Speaker A: Yeah, certainly the small players are more agile and they lack the five different um, approval processes.

Speaker B: Yeah.

Speaker A: To get decisions done.

Speaker B: True.

Speaker A: So we've looked at agent E Commerce, the hurdles behind it. Now if we consider success, how do you think that should be measured? Um, would it be conversion, the speed, uh, would it be the cost? Maybe if we lower the cost for the consumer doing retention for the consumer? Or do you think it's something else entirely?

Speaker B: I think in my opinion it is whether the agent or the AI got your intentions right and executed on that. It's of course a bit hard to, I mean, measure in my opinion. It's not like if transaction went through or not, because there are multiple things in this whole process. I mean, if, to understand if the agent got your, um, let's say requirements right or not, uh, if agent, uh, picks something and I'm happy with that, did they avoid mistakes? Do I trust it enough? Like things like this play a role, like to understand if the agent got your requirements right or not. I think because again, uh, there is a risk. If agent gets it wrong, then you lose money. You could lose huge amounts of money. So in my opinion, the metric is to understand if the agents got your intent or requirements right and executed on that. This is the metric for me, at least for now.

Speaker A: And I think it's also a matter of who you're defined as success for. Is it for me, the consumer? Is it for the merchant? Is it for the AI agent itself?

Speaker B: True, true, absolutely true. Look, but if you look into the thing that I said as a success metric, like if the agent got it right or not, because again, the thing is, it's across the board, everyone wants to understand if the agent got it right or not. So I think that could be a very good success metric. But let's see again.

Speaker A: Yeah, and maybe just to close off with a final question, if we look at the customer journey currently, which part of it do you think will look fundamentally different with Agent E Commerce?

Speaker B: Um, I think the biggest change is that people stop interacting with checkout the way we do at the moment. Like we go to the website and then we see different options and we do checkout. So it's going to change. Instead of going step by step, in my opinion, you just define what you want and it happens, basically. So like, instead of saying I'm buying this, you're saying like, this is what I'm trying to achieve and the agent handles that the rest of the. So this is the shift that, um, I'm assuming will happen very soon in the market. And so basically instead of doing things manually, you like give the agent instructions and it executes on behalf of you. These are the changes that I'm seeing in the near future happening in the market.

Speaker A: Amazing. Sam, thanks a lot for sharing this with us today. Um, and thank you all for the listeners who joined the Leaders in Payments and FinTech by EDC. We're always eager to hear from you. Um, so head over to our website, uh, linked in our show notes and click Connect With Us. Uh, to share your ideas or just say hi to us. And if you've enjoyed today's episodes, please, um, hit, follow and subscribe whenever, wherever you watch, um, or listen. There's plenty more great conversations that we will have in the future. Until next time, take care.