How AI Shopping Moves From Chat To Checkout with Adyen's Karan Katyal

AI is changing how people shop, but the real story is what happens after the recommendation. When buying moves into ChatGPT-style interfaces, merchants suddenly face a new maze of platforms, protocols, and trust questions and nobody wants to rebuild checkout for every new “browser” of the AI era. We sit down with Karan, VP, Head of Agentic Commerce at Adyen, to get specific about what’s real today, what’s still early, and what infrastructure has to exist before agents can safely transact at scale. 

We break agentic commerce into four clear models: human-in-the-loop conversational checkout, consumer autonomous purchasing, machine-to-machine transactions, and B2B agentic procurement. Then we dig into the “universal translator” approach: real-time product feeds for discovery, cart creation that plugs into a merchant’s existing tax and fulfillment stack, and a payments trust layer that keeps transactions secure while preserving the merchant as merchant of record.

Subscribe for more conversations with the builders shaping the future of fintech.

Connect with the Hosts & Guest

Sasha Pilch: https://www.linkedin.com/in/sasha-pilch

Karan Katyal: https://www.linkedin.com/in/karan-katyal-14070438/

About This Month in Fintech

This Month in Fintech is our premiere showcase podcast.  Each month, Sasha Pilch will have a deep 1:1 conversation with the leaders in the fintech arena, exploring questions about the industries next moves, where to focus energies, and how to lead into the next chapters and paradigm shifting technologies change everything aga

Transcript

Welcome And Karan’s Journey

Sasha Pilch 0:03

Hello and welcome to the listeners of the This Month InfinTech podcast. My name is Sasha Pilge, and I'm delighted to have our guest today, Karan, the head of Agentic Commerce at Agin.

Karan Katyal 0:14

Thank you so much for having me on the show.

Sasha Pilch 0:16

Great. So there's been some really exciting product launches, especially like within the Agentique Commerce realm, which is the hottest topic right now in our industry, which is really exciting. But before we jump into that, I'm just gonna get to know you a little bit. So you went to Emery and studied economics and political science. You worked at Deere, Isle Capital, and Nova's Partners, and then spent three years at Bain as a consultant. Lots of people that I know that have come out of Bain have gone on to do brilliant things. Like I used to work at Platt, and Zach, the CEO, was ex-Bain, one of my clients now, turnkey. Their CEO, Bryce, is also ex-Bain. So do you feel that that consulting experience has shaped where you are today?

Karan Katyal 1:08

Yeah, I think it it definitely helped. I think as a consultant, you kind of get trained to think in a very like structured way about ambiguous questions. And that that's basically what agentic commerce is, right? It's just a whole lot of ambiguity. So yeah, it definitely helped along the way.

Four Agentic Commerce Models

Sasha Pilch 1:24

Let's talk about the state of agentic commerce today. So you've now been at ADIN for almost three years, and you're heading up arguably the most cutting-edge initiative. So what is the state of agentic commerce today?

Karan Katyal 1:38

Yeah, for sure. Um maybe yeah, let me let me let me just set the scene a little bit, right? Uh we're we're definitely early days, but we're kind of seeing four flavors of agenti commerce start to emerge. The first flavor is what we call kind of consumer human in the loop flows, right? Think about going onto uh Gemini or to ChatGPT or something like that, you know, searching for a new pair of sneakers, uh, and then really doing the discovery as well as the entire purchase directly within that chatbot type interface, right? Um it's almost like an embedded checkout experience that exists not on a merchant's own website, but on one of these, on one of these AI platforms. That's kind of the main flavor that we're seeing a huge amount of building in today. That's what protocols like ACP and UCP are also quite focused on. The second flavor we see is what we call consumer autonomous flows, right? That's really kind of some of the magic of agente commerce. You go out and you say, hey, uh, you know, book me this flight when it falls below such and such price, or or or buy me groceries for the week and it goes off and kind of does it on your behalf. Uh the consumer kind of sets up the parameters up front, right? The budget, the preferences, the rules, but the agent then goes out and executes on that purchase without kind of needing to come back and get approval every single time, right? Um we're very early, very nascent on that. Uh, there's still a huge amount of kind of uh rules and infrastructure and uh and and regulations that still need to be set up around things like trust and intent, et cetera, right? Um if you think about what Google announced last year with AP2, that's what AP2 is is quite focused on. The third flavor then is machine to machine, right? Think about an agent going out and purchasing things like uh API access uh or digital content or or compute or something like that, right? It goes off, it purchases it in the background, you know, from oftentimes from another machine, without a human really necessarily even knowing exactly what's being purchased on the back end, right? If you think about protocols like X402 or MPP, that's where they they sit. So very nascent, a huge amount of infrastructure is being built on here, but it we're we're very early. Uh a lot of merchants and businesses are still trying to figure out how do they even want to take part in that. And then the fourth flavor is what we call B2B flows, right? Where you know you're going out and uh and autonomously doing things like uh like procurement or or paying out suppliers, stuff, stuff like that. Massive, massive TAM opportunity, but it's very, very fragmented and super early. There's not really yet a protocol that exists that tries to solve that yet. So that's kind of the the state of the union as to where where we are uh uh at the moment. A lot of what we have been building so far uh on Addy Agentic has been focused on that first flavor, but we are definitely thinking about some of the other flavors as well.

Sasha Pilch 4:24

Okay, so like where it's at today is clients that you have like Cezanne, like I love their clothes, um, allow people to be in chat or Gemini or Meta AI, having natural language conversation about what they should buy and get the recommendation and then buy it then and there within that interface.

Karan Katyal 4:47

You got it. That's basically that's exactly right.

A Universal Translator For Merchants

Sasha Pilch 4:49

And I also read within your recent press release that it's also a universal translator. So it works on those platforms that are currently used by everyone right now, but then whatever comes next, it can also work.

Karan Katyal 5:06

Yeah, exactly right. Um yeah, so there's there's a there's a few points I kind of make on that one. I think the first one is yes, you're you're you're you you've summarized it all spot on. Um I think as I mentioned earlier, we're still very much in that first flavor, the consumer human the loop flavor. The direction of travel is things are gonna get more and more autonomous. A lot of things still need to be figured out, but we are very, very early. What we are focused on, right, at IDN Agentic is we want to make sure that merchants are ready whenever that future state arrives, right? Real-time product data, uh, cart creation and cart orchestration, as well as payments in a way that is safe and secure, without having to go out and rebuild from scratch every single time. So, this this phrase that you used, universal translator, that's basically what our vision is, right? The key design principle here is merchants shouldn't have to go out and start to pick winners this early. The actual AI platform landscape is going to look very different 18 months from now, as it did, you know, six, nine months ago as well, right? Our job is to give merchants that common layer so that they can integrate once and participate across the ecosystem, regardless of which AI platform or platforms end up winning. Right? The analogy here is like imagine a merchant had to go out and rebuild their website for every single browser, right? That's almost what Agentic Commerce feels like today. Adding agentic really tries to solve for that. Does that make sense?

Sasha Pilch 6:26

That makes total sense. And because we are in this high growth stage, no one knows what's going to be the main platform that we're gonna be using for shopping. So the fact that this is so adaptable to anything is like so much less risk and less cost for your customers. So it's a great approach. And so there's three components to it: the feed, the card, and the payments. Can

Feed Cart Payments In Practice

Sasha Pilch 6:55

you talk us through that?

Karan Katyal 6:56

Yeah, absolutely. Maybe even before I get into the product, let me maybe I'll start by just setting the stage, right? Anytime you, as a customer or as a shopper and you know, want to go off and buy something on the internet, you kind of go through three steps, right? One is discovery. You go out, you know, today you might be doing it on Google, you might be going on to Reddit and looking up some reviews for whatever you're trying to do. But the first step is kind of product discovery. And already we're seeing a huge amount of disruption to that from AI platforms, where increasingly shoppers are beginning that discovery process uh on a on a Chat GPT or in some sort of LLM. The second piece then is the cart creation, right? Actually adding something to the cart and going out and making the purchase happen. When you go off and do that, a million and one things start to happen in the background, right? An order management system gets triggered, a loyalty system might get triggered, uh, you know, inventory, fulfillment, shipping, billing, et cetera. All those things need to happen. And then finally, there's the payment, right? You actually clicking the buy button and having your card details or whatever payment method you're using, getting pulled in and actually having that occur in the background. Also, what's going on in the background is things like risk and fraud, et cetera, are all also being being done on here. Our product suite here, right, uh, Adin Agentic, is really designed to solve each of those three steps in an agentic world. So the first piece is on the discovery side, right? That's what the agentic feed is focused on, right? It's all about making sure that a merchant's products are visible and usable across agentic environments. It gives the AI platform uh a kind of structured real-time access to things like catalog, pricing, as well as inventory data. So the agent is working off accurate information. That's what the feed is focused on. The second piece is where the cart now becomes, it makes the discovery into an actual merchant ready checkout, right? It connects the AI experience back into the merchant's existing checkout, tax, fulfillment, order management, et cetera, stack on the back end. So the merchant doesn't have to go out and uh you know rebuild their entire commerce stack just for every new AI platform. And then the third piece is the payment, right? That's the trust layer that makes the transaction safe and secure as well, right? It handles things like authentication, tokenization, fraud, risk management, as well as payment routing, so that the merchant can always remain that merchant of record and maintain that direct connection with their with their shopper. That's kind of what what the uh what the what the overall offering is really focused on.

Sasha Pilch 9:21

And that last piece about the merchant still maintaining that relationship seems very important because considering like where this could end up going. Like if agents are transacting with other agents, like does the merchant essentially become out of the loop? So it's like good to hear how it's currently working. And

Why Adyen Fits The Role

Sasha Pilch 9:41

why is Agin uniquely positioned to solve this problem? Like, why wasn't it like why couldn't it be Stripe or Shopify or a large bank to have a universal translator?

Karan Katyal 9:52

Yeah, I'd say a few things. Um, I think the first one is just like technical scope, right? Um I think when when all of this started, you know, last year, as we were first digging into it, a lot of us thought that the problem or the long pole of the tent would be payments. What we've learned is it's actually much bigger than payments. Payments is actually pretty solvable, right? It requires like orchestration across product data, across checkout, authentication, tokenization, fraud, and then the payment. And then you get to things like acquiring, issuing, settlement, et cetera, right? Adian already operates across that entire lifecycle through a single global platform, right? All of those foundations that we've been building on for 20 years kind of work really nicely for this new era. Most players often will sit at one level of that stack. We kind of own the entire piece of it. The second piece then is what we call the N times M problem, right? We work with thousands of the world's largest enterprise merchants across every vertical, every geography. That means when we are building Addy Agentic, we're not solving this for one merchant at a time. We're kind of building that infrastructure that scales across the entire ecosystem, right? Every integration that we do or that we build with an AI platform ends up solving for all of those merchants, right? That also, that while that's very helpful for our merchants, it's also very helpful for the AI platforms, right? Because ultimately, right, if I as a shopper want to start to transact with one of these AI platforms, I need to be able to buy from my favorite sneaker store or my favorite uh, you know, favorite shoe store, whatever that looks like, right? That n times n problem is also where we're kind of sitting as that ecosystem player. The third piece is neutrality, right? The universal translator needs to be neutral. The structural reality, right? If you go out and take a look at a lot of the other folks that are building on here, they often have their own surface or their own ecosystem or their own commercial incentive to kind of get merchants and consumers inside their own platform. That's just how their business operates, right? Our business, our primary constituents are merchants and helping make sure merchants can transact wherever their customers are. That really means our incentives are super well aligned with where our merchants are. And the final piece is trust, right? This is kind of foundational. Agentic commerce is ultimately going to start to introduce some pretty hard questions around authentication, authorization, as well as fraud at a level that the industry maybe hasn't seen before. Who's authorized the agent to go off and make the purchase? Is this transaction actually legitimate? All of our existing infrastructure, which is built on top of trillions and trillions of dollars worth of volume, that is the foundational layer that this all gets kind of built on, right? It's not a bolt-on, it's just it's just kind of the the foundation of it.

Intent Fraud And Trusted Agents

Sasha Pilch 12:32

And that trust piece is such an important topic right now. I did a panel with the new head of fraud at plaid, Pedro, and Sbastion from Ledjar and Will Turner from HSBC, and we talked about this agentic trust issue, and it seems to be that intent is the underlying factor as to decide whether or not it was a malicious activity or it was like legitimate. Um, I've also had the chief innovation officer of Experian on the podcast recently, and she was talking about you know, we are in these fast-moving uncertain times where no one knows what the new phase of bad actors and agents on Wilde is going to look like. So, can you talk a little bit about like how Adjun is battle tested and enterprise grade?

Karan Katyal 13:24

Yeah, for sure. Um Yeah, I mean, I think so. Essentially, the question is how does like the existing worldview of fraud actually end up translating to this new piece? Yeah. Um, I mean, I think there's a there's there's a lot to it, right? I think one, this point you've raised around intent is exactly right. Especially as you get past the first flow, right? The consumer human and the loop flow, you prove intent the same way you've always proved intent, right? A human ends up clicking the buy button, right? That's how you that's how you do it. When it becomes more autonomous, all of a sudden the intent question becomes a lot more a lot more open. I think there's a few different approaches and a few different players who are trying to come in and own that intent layer. And I think that the goal here is we need to work with everyone. From a fraud perspective, um, we are already seeing, right, or all of our models are already trained on, you know, trillions and trillions of dollars of data, as I mentioned before. It's about now adapting those fraud models to be able to figure out is this a malicious bad bot or is this actually a good agent? Right? Is there some identifier that allows us to figure out who the shopper is that sits behind that agent? Has that agent actually gone out and registered themselves at some sort of registry to make sure that they're actually trustworthy? Those are some of the big concepts that I think we're still grappling with.

Sasha Pilch 14:38

Yeah, that makes total sense. Like in a way, like replicating like what we have now, um, but like for this new agentic world.

Karan Katyal 14:45

Exactly. And of course, it's gonna take a bunch of time, right? The reason risk and fraud models work so well today is there's so much data to train them on. Right now, there's almost, you know, if in effect there's almost zero real dollars of transaction volume across the ecosystem, it will take time for those fraud models to actually catch up.

Sasha Pilch 15:05

Yeah. All of these scams that we now know to steer clear of were once new and people fell for them. And so we're gonna see like a similar pattern. Um

Forecasting Adoption And Where It Hits

Sasha Pilch 15:17

and you know, you mentioned that this is still very much the beginning. What do you think will be the KPI? Like three years from now, what percentage of commerce will happen through agents? Like 1%, 10%? When do you think it'll actually become dominant? I know like it's hard to predict the future, but like just from your expert opinion.

Karan Katyal 15:43

Um I would say without getting super, super, super specific, right? Um, we went out and we did our own kind of market modeling and we interviewed, you know, several hundred merchants as we were doing all of this. Five years or so from now, we expect, you know, call it $2 trillion or so of global e-commerce. We expect to be agent influenced in some way. That's not all fully autonomous stuff. That includes things like discovery and recommendation, assisted checkout. But it's a big enough number that it justifies building the infrastructure now versus waiting for it all to be done down the road, right?

Sasha Pilch 16:20

What percentage is that two trillion numbers?

Karan Katyal 16:23

Uh I think the I think when we were doing the numbers, I think somewhere between call it like eight to twelve percent, somewhere in that sort of a range. The other point to raise here is that the actual penetration is probably going to be quite different depending on the vertical and the geography that you're looking at, right? Categories that have a have a have a ton of decision complexity, right? Things like consumer electronics or travel or fashion or you know, some ports of B2B procurement, those might re those might end up getting um the uh the agentic influence a lot faster than things like luxury goods, where the brand relationship and the tactile experience matters more. Similarly, markets that have a have a ton of AI penetration or or mobile first commerce will probably move faster than others, right? So I don't think it's gonna be one of those things that just happens across the board all at once. It'll probably happen in a phased way. Um, and it'll probably also track back a little bit to that initial framing I said earlier on around the the four flavors that we're seeing.

Sasha Pilch 17:22

Yeah, totally. Because there's enjoyment in like having those experiences of like going into Chanel and like buying handbags or you don't want to lose that.

Karan Katyal 17:31

I mean, that's what it is, right? Anything that you as a shopper feel some sort of transaction utility from, what is the what is the job that the agent is really going off and doing for you, right? So the same reason people love going into stores in person, right? Uh it's not like because you can buy stuff online, you're not going into stores, and I expect that that that pattern will also uh emerge again.

Sasha Pilch 17:53

Exactly. But then like forgetting to buy like bin bags is like not enjoyable. So having agents being on top of all that stuff um makes total sense.

Karan Katyal 18:04

Or or another one, right? Like uh I I could find the perfect flight to take tonight, like in 10 seconds, right? Uh from New York to Amsterdam or Amsterdam to back to New York. And then I need to click another 10 times until I can actually make the purchase happen. And many minutes will pass, and I'm getting zero joy as a shopper during those minutes. Those are the sorts of use cases I think we're we're quite excited about, at least on the consumer side.

Sasha Pilch 18:30

It's actually stressful. Like I hate that when I'm like, am I is this one seat left on the Delta flight gonna be there? And then yeah, I would like an agent to take that off my plate for sure.

Karan Katyal 18:41

Please, please.

Consent Liability And Regulation Ahead

Sasha Pilch 18:43

Um, so what about the hardest thing from a technical or regulatory challenge that you think needs to be solved?

Karan Katyal 18:53

Um, yeah, that's a great question. So um so I think from a technical perspective, I think the biggest challenge right now is not the payment piece, but the trust piece, right? The authorization and the authentication infrastructure that proves that a consumer actually mandated that specific agent to act on their behalf for a specific transaction, right? And that needs to work on an in an interoperable and quite open way. There's a lot of open questions that still need to be answered. And I mean, if you ask, you know, five, ten different players in the ecosystem who has the right to be the ones who should be the arbiter of that, you'll get five or ten different answers as well. Which is somewhere where I think regulators will have to start to think about quite a bit as well. The other piece on the regulation side is there's a huge amount of there are a lot of key questions still, right, around things like consent, uh, liability, data use, and consumer protection, right? Who's responsible if the agent makes a mistake?

Sasha Pilch 19:52

Right?

Karan Katyal 19:52

Who's actually on the hook for that? What actually constitutes like a valid consumer authorization, right? This stuff is you know non-deterministic. Is that A or B? There's some there's some air, some gray areas there I think that need to be figured out as well. Those questions are not at all answered yet. And how they get answered will end up really shaping how quickly the the direction of travel starts to move.

Sasha Pilch 20:15

Yeah, totally. And yeah, it's like a test and learn thing. We're gonna have to figure it out, but we will.

Karan Katyal 20:21

We will, we will for sure. But there's you know, there's there's launch fast as iterate, and you can and should do that, but there are points also where I think uh anytime money is moving around, you know, those those rules need to be set and folks need to agree to it. Otherwise, things will get very, very messy very fast.

Sasha Pilch 20:39

Can

Uplift AI For Conversion Balance

Sasha Pilch 20:40

we then talk a little bit about add-in uplift, your AI product that increases conversion, reduces fraud, lowers costs? Like how does that work?

Karan Katyal 20:48

Yeah, absolutely. So Uplift is our we we we we released Uplift um you know early early last year. It's the AI-powered optimization layer for the full payments funnel, right? So it kind of sits across all of the found all those foundational elements I was talking about earlier on, right? Tokenization, fraud protection, authentication, uh optimization, personalization, so that you can really optimize the full journey rather than doing it in isolated pieces, right? Oftentimes a lot of businesses need to figure out okay, do we want to try to boost up conversion? Do we want to bring down cost? Uh uh, or do we want to try to keep fraud as low as possible? And it's often you can do, you know, you can try and optimize for one or two of them, but often at the at the cost of the other one. Uplift is trying to say, look, what is the full funnel view of this so that you as a business are as best off as possible? So in practice, what does that mean, right? It means things like showing the right payment method for a shopper, right? It means applying the right authentication flow based on the risk context of how risky this particular person is, right? So you're reducing false positives on fraud that would otherwise block legitimate transactions, right? The advantage here, I've I've mentioned this a couple of times, right? It's all of the payment volume that we've been able to train our models on. Right. We we did something like one and a half trillion dollars of payment volume last year. That is a training data that Uplift is trained on. That data scale is something that is very, very difficult to actually replicate at scale.

Sasha Pilch 22:18

That's excellent because then you're drawing from all of these billions of transactions over the last 20 years in order to have this offering that for your customers decides that friction versus fraud paradigm and where that should sit for these particular transactions, for these specific transactions. That's excellent.

Karan Katyal 22:38

Exactly right. And I mean, honestly, that's the reason why we've we've kind of built the way we've built, right? We've gone out and, you know, whether it's in person or so, I mean I'm in Amsterdam, right? I could spend an entire day today just buying from merchants that work with Adian. In store, online, you know, the the the the shared ride that I that I asked for to come over here. Um yeah, it's that that sort of deep, deep, deep, deep penetration that I think helps a lot. It's also one of the reasons we ended up you know going for our banking licenses years ago. So it all it all kind of fits into this broader piece, which is you know, all these things work in conjug conjunction to make a really nice happy flywheel.

Sasha Pilch 23:14

Yeah,

Banking Licenses And Stablecoin Debates

Sasha Pilch 23:15

owning the whole stack has been so like powerful for Adjun. Now that you have the banking license, and congratulations on that, it's definitely not easy to get. Um does this make accounts, capital, card issuing, real-time payouts simpler, more cost effective? And does it mean that Adjun will be less dependent on stable coin for treasury, considering and like other use fintech use cases?

Karan Katyal 23:45

Um it's a great question. Um okay. So I think first thing I'll say is our banking licenses in the US, uh in the EU, and the UK already kind of give us real-time settlement and payout capabilities that solve a lot of the same problems that stable coins are positioned to address for for other fintechs, right? Things like cross-border speed, uh treasury efficiency, cost reduction, uh, et cetera. So in that sense, I'd say yes, we're a little bit, maybe a little bit less dependent on stablecoin infrastructure to solve those use cases because we already have the pipes built. However, we do watch the space quite quite carefully, especially for certain kind of B2B uh and or cross-border constraints where the economics are actually quite interesting uh and the infrastructure is maturing. So it's not really uh an ideological position. It's more about as we look at it use case by use case, we're ultimately pretty much agnostic as to what the underlying technology is. So if it makes sense, I think it's something we'd be.

Sasha Pilch 24:44

If it's like certain countries, certain exchange rates, stablecoin could make sense, and then agent's gonna serve that as the best option.

Karan Katyal 24:51

Totally, totally.

Sasha Pilch 24:53

And do you see stable coin transactions via agents being an important thing as well?

Karan Katyal 25:01

Yeah. Another excellent question. So one, like I reserve the right to change my mind and all of this stuff. So if I go back to those four buckets we talked about at the beginning of this, right? For consumer human in-the-loop purchases, especially in developed markets, I'm not really, yeah, I don't really see a use case in the near term for stable coins to get on there, right? Uh whether it's the card networks or any of the other ways that shoppers already like to pay all over the world, I think those work quite nicely for those use cases. More or less, I kind of feel the same way even for consumer autonomous flows, right? If I'm sitting in the US, I like getting my points. I like to be able to do chargebacks if I if I need to. Uh I like going into my you know, Chase Sapphire Lounge at JFK before my next flight. Why would I give up on any of those things, right? When you start to get to the third bucket, the machine-to-machine bucket, that's where I think some of these stablecoin pieces do start to get a little bit more interesting, right? When you start to think about things like, you know, microtransactions or or or or very high velocity transactions. All of a sudden, you know, although, you know, perhaps now there is something that a stablecoin can do. I'm not yet, you know, I don't have super high conviction though that card networks can't also try to solve that or or solve that in a way that that might work between things like batching and other things along those lines. On the B2B side, yeah, I think there's a lot of a lot of potential there for stable coins to start to come in and figure it out. So yeah, I think it'll be really, really interesting over the course of the next few years to see how that starts to happen. Um yeah, so it's a little bit of a roundabout answer, but that's kind of like.

Sasha Pilch 26:43

No, I know. I I think it makes total sense.

Karan Katyal 26:45

Like what do you think? What's what's your take?

Sasha Pilch 26:47

Well, I I agree. Like an agent or like a company isn't going to benefit from getting enough Amex points to go to the first class lounge. So what's the incentive there for them? And then for especially transactions that are happening cross-border, if stable coins can make that so much more efficient and you know cheaper and faster, then why wouldn't they use it? Like that's kind of like where my head's at. But that said, it is nascent and we don't have the controls in place that make it easy to do a chargeback or you know, talk to someone and like get something reversed, etc. So there's that risk to it as well.

Karan Katyal 27:36

I think I think probably one of the big drivers will be like how how much like a human is this agent really acting? So, what I mean by that, that first use case, consumer human the loop, it is a full human. The second one, consumer autonomous, it still wants to make decisions the way you make decisions, right? It's just acting as an agent on your behalf, so you know, you might still want it. As you start to get away from that though, right? Machine to machine and B2B, totally agree. I think there's lots of other things that these agents might optimize for uh down the road.

Acquisitions Loyalty And Usage Billing

Sasha Pilch 28:07

Great. Um, so we're nearly at time. Last um last few questions. So, you know, we talked a lot about Adjun's competitive advantage by building everything in-house. And um, I had lunch at your New York office the other day, actually, with Sakrit, who I used to work with at Platinum.

Karan Katyal 28:24

Nice. Oh, excellent.

Sasha Pilch 28:25

Yeah. And um, she was saying that Adjun for the first time ever made MA activity. So she told me about Talon.1, and then like later that week I saw the announcement of Orb. So are we gonna see more acquisition activity from Adian?

Karan Katyal 28:43

Uh I'll probably I'll stay in my own lane on that, on that question. I won't I won't I won't speak speak too much too much, but maybe I can talk a little bit about them as well. One, I think uh both both of these acquisitions I think make a ton of sense on their own, right? But when you start to look at it through the lens of agenti commerce and the problems that need to be solved at scale, it all starts to come together a little bit, right? So I mentioned at the very so I'll I'll go in a little ramble here. I mentioned at the beginning, right? We spoke to hundreds of merchants as we were building this, right? Every step of the way, we were working with merchants to figure out what are you excited about, what are you afraid about. And there were kind of three big things that came up every time. Uh the one that we heard on every single conversation was this fear of disintermediation, where hey, I'm a merchant or brand. I might, you know, I might spend tens or hundreds of millions of dollars every single year on customer acquisition and on marketing and on developing a deep and emotional connection between me as a brand and my shopper on the other side. I don't want there to be some AI platform that now entirely owns that relationship. What's the other side of the coin to disintermediation? It's loyalty, right? That's where Talon 1 starts to come in, right? Talon 1 addresses loyalty and promotions, right? You know, I've already talked about merchant control, I've already talked about disintermediation. Loyalty ends up being the mechanism by which merchants are able to start to win repeat relationships, even if that first transaction happens via some sort of AI agent. So Talon 1 ultimately we think will end up giving merchants the tools to build and manage those relationships and not totally abdicate the customer relationship to these AI platforms. On the orb side, orb really starts to come into the world of usage-based billing, right? As commerce becomes more agentic and more dynamic, the billing models that sit underneath businesses will need to start to become a little bit more flexible. Static pricing doesn't always work in a world where an AI agent is starting to make these purchases. Or if you think about some of these new business models around charging for uh you know, uh based on usage for API access or for digital content or anything like that, you now will need help figuring out how much to charge and when to charge and how that's all going to end up working. And I think that's where Orb will become really, really interesting. And I think, yeah, I think what's most important here is both of these products are genuinely valuable today, even before Agentic Commerce reaches scale, right? Talon 1 is the best in class loyalty infrastructure for enterprises today. Orb is the best in class billing orchestration engine as well today. So regardless of how things go with agentic, and you know, as you can tell, like it's something I'm quite bullish on, even leaving that aside, they're also solving real-world problems that our merchants are facing today and will only become more powerful in the future.

Sasha Pilch 31:27

That makes so much sense. Um, to be honest, I when I first saw it, it was a bit like question mark of like, well, why those two companies? But yeah, the loyalty piece, because it's gonna increase lifetime value of customers, and we need to be ready for that in an agentic world. And the orb piece with billing makes total sense. Like when I was a VC, um, we did like a lot of work in the CFO tech stack space and orb and metronome and sequence and like all of those type companies. And so now I can see like why they're becoming so much more valuable. So thanks. That was like a very, very um helpful explanation.

Favorite Books And Final Thoughts

Sasha Pilch 32:10

Um okay, we're actually over time now. So I'll just wrap up with like the one question I ask all guests because I'm very interested in what these amazing people I get to interview read. So, what are your favorite books? One personal can be fiction or non-fiction, and one professional.

Karan Katyal 32:28

Okay, cool. Um personally, I love historical fiction. So I think probably I'd say Lonesome Dove or The First Man in Rome, both excellent reads. Professionally, uh, I read this book a few years ago called Range, and I loved it because it it was like confirming all of my own biases about myself. So it's like it basically says like don't hate on being a generalist. You know, do more, like like cultivate interests that are outside of your lane, and it'll just end up translating. So those are my actually three picks I give you.

Sasha Pilch 32:59

Awesome. Great. I can't wait to read them. And that's all we have time for. So thank you so much, Karen. This was a really awesome conversation. I learned so much, and I'm excited that the listeners get to learn from you as well.

Karan Katyal 33:11

Amazing. Thank you so much for having me in the show.