[00:00] [music]
[00:13] >> Hey everyone.
[00:15] Welcome to our breakout session on
[00:17] agenda commerce.
[00:18] We have a pretty packed agenda here, so
[00:20] I'm just going to get into it. Um so my
[00:22] name is Jesse and I'm thrilled to kick
[00:24] things off with you. Um the real stars
[00:27] of the show are to my left over here. Um
[00:29] so we're going to be joined by Kieran
[00:30] and Mike who represent both the kind of
[00:34] commerce side as well as the payment
[00:35] side of the house. You'll be hearing a
[00:36] lot from them. I'm just kind of the
[00:38] context setter. I'm going to tell you
[00:39] about our toolkits and what we have here
[00:40] at Google Cloud. But um let's get going.
[00:44] So a little bit about me, I'm an AI
[00:46] incubation lead here at Google Cloud.
[00:48] I focus on pre-GA capabilities that are
[00:50] launched in Cloud AI. So I run pilots
[00:52] with customers to kind of explore new
[00:55] products and I'm thrilled to to be here
[00:57] today to talk about a topic topic that
[00:59] I'm quite passionate about.
[01:02] So quick agenda.
[01:04] So first we're going to talk about the
[01:05] immense opportunity that retailers and
[01:07] payment providers
[01:08] um that is being unlocked with agents.
[01:11] And then we're going to delve into the
[01:12] building blocks that power agenda
[01:13] commerce here at Google Cloud.
[01:15] Next up, we're going to hear from
[01:16] Wayfair about how they've successfully
[01:18] onboarded 30 million products across
[01:20] various agenda services.
[01:22] And then we'll hear from Fiserv, a
[01:23] leading payments and fintech platform,
[01:25] on how they're seizing the opportunity
[01:26] on payments.
[01:29] So I'll start with a quick story. So
[01:32] when we had our our kid 2 years ago,
[01:36] I was pretty overwhelmed by just the
[01:37] sheer amount of stuff that we needed. Um
[01:39] so a quick trip to go see the
[01:41] grandparents meant packing pack and
[01:43] play, stroller, various creams, diaper
[01:46] bags.
[01:47] Um and you add in two adults and a 75-lb
[01:50] golden retriever, our car was bursting
[01:52] at the seams.
[01:53] So a friend said, "Why don't you get one
[01:55] of those cargo boxes you can put on top
[01:56] of your car? You can load all your
[01:58] luggage and stuff in there." And I
[01:59] thought, "Great idea." But then I was
[02:02] just overwhelmed by the amount of
[02:03] different products out there.
[02:05] I had to have something that fit my car,
[02:07] fit my sanity, my budget. It had to be
[02:09] like safe enough that it wouldn't fly
[02:10] off the uh roof during a a trip on the
[02:13] freeway. And it had to be like pretty
[02:15] aerodynamic. And there were a million
[02:17] products and I was just drowning in
[02:18] browser tabs, um
[02:20] product reviews, and I couldn't really
[02:22] optimize this equation.
[02:24] So I finally just dumped in Gemini all
[02:26] my constraints and my budget and
[02:27] everything. And within seconds it spat
[02:29] out the perfect recommendation for me.
[02:32] So now I was able to basically travel
[02:35] down to see the grandparents.
[02:36] We're all comfy and my golden retriever
[02:38] doesn't have to wear a diaper bag as a
[02:39] hat anymore.
[02:41] So the lesson, LLMs can help us sift
[02:43] through large amounts of data as well as
[02:45] constraints and preferences, and they
[02:47] can help us organize information from
[02:48] many different places, and ultimately
[02:50] help by driving decision.
[02:54] And it turns out I'm not the only one.
[02:55] So according to some recent research, um
[02:58] there's been a 10x increase in brand
[02:59] discovery um through LLMs in 2025 2025
[03:03] alone. And 30% of shoppers are actually
[03:06] initiating that product discovery
[03:08] through AI and it they expect it to kind
[03:10] of grow in the future.
[03:12] And all in all, consumers are already
[03:14] augmenting traditional search and kind
[03:15] of browsing with AI to surface
[03:18] hyper-relevant recommendations.
[03:21] A couple more stats. Uh another study
[03:22] said that 40% of all AI-driven
[03:25] e-commerce sessions land directly on
[03:27] product detail hit pages. And this
[03:29] indicates that AI is actually helping
[03:30] develop a fully formed purchase intent.
[03:33] And why is that? I I think my cargo box
[03:35] story explains it all. We can collapse
[03:37] things like search uh bars, filters,
[03:39] browser tabs, product specs, product
[03:41] reviews, and many other things into a
[03:43] highly adaptive conversational thread.
[03:46] So it compresses the funnel and can
[03:47] filter out complex buying constraints in
[03:49] a single step.
[03:51] So let's talk about the opportunity.
[03:53] Shoppers are already using AI to
[03:55] discover products today.
[03:57] And that's whether you like it or not.
[04:00] So here at Google we think, "Why not
[04:02] meet them where they are, remove some of
[04:04] that friction, and unlock a new buying
[04:05] channel?"
[04:06] So
[04:08] uh with opportunity though comes some
[04:09] responsibility. And brands must meet the
[04:12] moment and optimize their catalogs so
[04:14] that the AI agents can safely discover
[04:15] and transact on them.
[04:17] And the payment processors who invest in
[04:19] secure tokenized checkout infrastructure
[04:21] are going to win the game in this AI
[04:23] AI agent world.
[04:25] And here at Google we think we're
[04:26] building the right products and
[04:28] protocols to support that.
[04:30] And by the way, much of that is
[04:31] completely open source,
[04:33] um ensuring that the digital
[04:34] infrastructure remains transparent,
[04:36] interoperable, and accessible to
[04:38] developers worldwide.
[04:43] So I'm going to hop into the our toolkit
[04:45] and I'm just going to talk through some
[04:46] of our products today. Um you'll get a
[04:48] lot more context around how they're
[04:50] being applied by my um my guest over
[04:52] here.
[04:53] But I'm going to start off with
[04:55] something a little bit more managed. So
[04:56] we, as you may have heard this morning,
[04:58] we recently launched Gemini Enterprise
[05:00] for customer experience. If you're in
[05:02] the know, you call it GECX. And it's a
[05:05] unified set of solutions that allows
[05:06] companies to provide signature AI
[05:08] experiences to their customers on their
[05:10] own properties.
[05:12] Um and we can use the power of Gemini
[05:14] and Google to do all sorts of things
[05:16] like product discovery, commerce, and
[05:19] even customer support. All this is
[05:21] rooted in your own brand identity and
[05:23] your customer relationships and your
[05:24] full ecosystem. So I'll focus today on
[05:27] the shopping agent, um which is offers
[05:29] like a really streamlined way to deploy
[05:31] um super sophisticated shopping
[05:33] assistants on your own property.
[05:36] So today the shopping agent, it's in
[05:37] preview. We're piloting it with a few
[05:39] customers.
[05:40] And it streamlines that entire
[05:41] purchasing journey from discovery to
[05:43] checkout. And it can leverage multimodal
[05:45] inputs and complex reasoning and
[05:47] basically translate a lot of open-ended
[05:49] queries and suggest like holistic
[05:51] solutions for that. So something like,
[05:53] "How do I outfit an entire graduation
[05:55] party?" Rather than searching for those
[05:57] items one by one and browsing,
[05:59] and it'll deliver real-time product
[06:01] suggestions and take consented actions
[06:03] to complete purchases, and remove a lot
[06:06] of the friction that you see at checkout
[06:08] um on traditional e-commerce websites.
[06:10] And at its core, the shopping agent
[06:12] provides convenience. It can curate
[06:14] personalized choices. It can suggest
[06:16] alternatives to out-of-stock items. And
[06:18] it can even move items into carts and
[06:20] checkout um with the customer's consent.
[06:23] So lastly, it can really help you tackle
[06:26] these complex goals. Maybe you have a
[06:28] really kind of abstract question like
[06:29] the perfect outfit for a special
[06:31] occasion. And it can really un uh
[06:33] anticipate unstated needs and guide a
[06:35] user through complex purchasing project.
[06:38] So that was a quick run-through. I'm
[06:40] trying to talk fast cuz I want to get to
[06:42] the uh the real stars of the show here.
[06:44] But um while the the shopping agent
[06:46] provides this powerful Google-managed
[06:47] solution that really curates this
[06:50] delightful experience on your own
[06:51] property,
[06:53] our open protocols, which I'll talk
[06:54] about today, which are UCP and AP2,
[06:57] empower you to facilitate that agenda
[06:59] commerce um using your own existing
[07:01] tools if you're building high-code
[07:03] agents, and also on a variety of
[07:05] different surfaces as well.
[07:06] So we believe that by leveraging these
[07:09] transparent, vendor-agnostic standards,
[07:11] you can actually build a seamless
[07:12] shopping ecosystem and with complete
[07:14] flexibility and absolutely no vendor
[07:16] lock-in whatsoever. You can build this
[07:18] on whatever you want.
[07:20] So let's take a look at the ecosystem
[07:22] today.
[07:23] So I think there's two major problems.
[07:26] So before kind of a common language for
[07:29] agents to interact with back ends, you
[07:32] had to build a lot of bespoke
[07:33] integrations. So this could be on the
[07:36] retailer side, it could also be on the
[07:37] payment side. Every time an integration
[07:39] required kind of like a custom set of
[07:41] API integrations that weren't
[07:42] necessarily extensible to the next.
[07:44] And um
[07:46] this is something we want to solve. Uh
[07:48] how do we standardize a schema so that
[07:50] an an LLM or even an API can actually
[07:52] understand that language and have a
[07:54] common schema that moves us towards more
[07:55] of like a streamlined communication
[07:57] paradigm.
[07:59] On the other side, if we wanted to build
[08:00] something, uh it has to be flexible. It
[08:02] can't be uh one-size-fits-all kind of
[08:05] monolith.
[08:06] And um you know, we want to be able to
[08:09] support a variety of different protocols
[08:11] that are already being used in the
[08:12] industry today. So we don't want to
[08:14] replace things like MCP or A2A. We want
[08:16] to complement them and integrate them
[08:18] into the framework.
[08:20] So this is where we built UCP.
[08:22] Um maybe you'd like to embed your
[08:23] shopping experience across a broad range
[08:25] of AI services. And UCP, or the
[08:28] Universal Commerce Protocol, is this
[08:30] kind of a single standard for AI
[08:32] commerce. And it covers everything from
[08:34] discovery to post-purchase. And Google
[08:36] built this with um leaders. Some of who
[08:39] Some of our industry leaders and some of
[08:40] those people are in the room today. Um
[08:42] so it's something that we didn't build
[08:43] in isolation with, you know, our own
[08:45] ideas.
[08:47] And it what it does is it creates a
[08:49] common language for AI agents to solve
[08:51] fragmented online shopping. It supports
[08:53] the entire user journey from real-time
[08:56] product discovery to cart management and
[08:57] order tracking. And it ensures that the
[09:00] uh retailers actually remain the seller
[09:01] of record, capturing sales exactly when
[09:04] customers are ready to um buy across a
[09:06] variety of AI services. That could be
[09:08] Gemini. Could be AI mode search.
[09:11] And basically anywhere you want to embed
[09:13] these agenda experiences.
[09:15] So it achieves this through core
[09:16] building blocks that require minimal
[09:19] custom code to implement. And it has
[09:21] basic integrations like native or
[09:22] embedded checkout. It has security
[09:25] secure identity linking, full order
[09:27] management, and interoperability with
[09:29] existing digital wallets and payments.
[09:33] So let's zoom out a little bit. Like
[09:34] what is it right here? So UCP sits
[09:36] between those consumer surfaces and your
[09:39] business back ends.
[09:40] And this consumer uh service could be,
[09:43] you know, a merchant property. It could
[09:45] be AI mode in search. It could be Gemini
[09:47] app. Or theoretically any surface where
[09:49] your customers are interacting with you.
[09:51] And it defines the interactions between
[09:53] four distinct actors. So first up, we
[09:55] have the platform, which be the agent or
[09:58] the application. It could The second is
[10:00] the merchant of record. Third is uh the
[10:03] credential provider, which is
[10:04] responsible for uh for things like
[10:06] customer identity, issuing payment
[10:08] tokens, and last is a payment service
[10:10] provider.
[10:11] And if you look at this stack right
[10:12] here,
[10:13] it's basically composed of four
[10:15] different layers.
[10:16] And at the top, we have our services
[10:18] layer.
[10:19] And UCP has a name commerce in its name.
[10:22] Um so, it starts with shopping, but it's
[10:23] extensible and it's designed to be
[10:26] vertically expansible so that you can
[10:27] onboard different types of verticals um
[10:30] using that same kind of concept that UCP
[10:32] is. So, in the future, it could be
[10:34] extended to things like groceries or
[10:37] travel, and basically the service layer
[10:39] defines that.
[10:41] Um we also have thing called a common
[10:42] layer, which you see in that that common
[10:44] box, and that's something that could be
[10:45] used across services um across verticals
[10:48] like um identity linking is a good
[10:50] example of that.
[10:52] Then we have our capabilities. So, we
[10:53] launched um
[10:54] catalog, cart, checkout, order, identity
[10:57] linking, and we're going to be
[10:59] onboarding more in the future. And these
[11:01] are the core business operations for
[11:02] that domain.
[11:03] Um and again, new capabilities can be
[11:06] added to facilitate other vertical-based
[11:08] kind of um interactions.
[11:12] And then last or third, we have our
[11:13] extensions, and these are specific
[11:15] pieces of functionality that decorate
[11:17] those um capabilities. And these are
[11:19] things like discounts and fulfillment.
[11:21] And by having these be decorators as
[11:23] opposed to their own capabilities, they
[11:25] can be kind of injected in the right
[11:27] point um in the transaction flow.
[11:30] And then lastly, as I mentioned before,
[11:31] we have our transport layer. And so,
[11:33] these these three semantic business
[11:35] layers at top are all furnished through
[11:38] um transports.
[11:39] So, these business contracts can be
[11:41] manifested as JSON um that communicate
[11:43] over REST APIs. They can be manifested
[11:45] as MCP or A2A.
[11:47] And then we also have the ability to
[11:49] embed the protocol into an iframe. Um
[11:51] and this is really good for when you
[11:52] want a human in the loop actually
[11:54] clicking a button, maybe a buy button,
[11:56] and you don't want to have that be done
[11:57] by an agent directly.
[11:59] So, before I close out on UCP, a an
[12:02] important distinction here. So,
[12:04] we have UCP, it's fully open source. Go
[12:07] on ucp.dev. You can go on our GitHub.
[12:09] You can create issues and pull requests.
[12:12] And this is open to any business that
[12:13] wants to federate commerce to any
[12:15] surface or any agent anywhere.
[12:17] Um but we've also, with a small pilot
[12:19] group of customers, launched our kind of
[12:21] own opinionated implementation of UCP.
[12:24] And this is a uh allows you to do UCP
[12:26] commerce over things like Gemini and AI
[12:28] mode and search. So, there's a lot of
[12:30] demand for this, and we're looking to
[12:31] onboard more customers. Um I don't have
[12:34] timing, unfortunately, but uh
[12:36] yeah.
[12:38] So, what about payments? Um so, when we
[12:40] think about payments, there's a lot of
[12:42] risk involved, right? There can be bad
[12:44] actors and fraudsters. There could be
[12:46] risk around chargebacks or frankly, like
[12:49] we just might be worried that agents are
[12:51] going to hallucinate a price or a
[12:52] customer intent. And how do we secure
[12:55] against that?
[12:56] So, I'm old enough to remember when you
[12:57] actually were kind of scared to buy
[12:59] things on the internet. It took a lot of
[13:01] uh
[13:02] risk to enter your credit card um into a
[13:05] box on the internet. But we all do that
[13:07] now. So, the payment industry really
[13:09] rallied to build the infrastructure of
[13:11] trust that made e-commerce possible. And
[13:13] what we're doing today with AP2 or
[13:15] agents uh agent payments protocol is to
[13:18] basically build that trust through this
[13:20] agentic interface.
[13:22] And we do that through these
[13:23] tamper-proof mandates. So, I'll talk a
[13:24] little bit more about that.
[13:26] So, agents break that fundamental thing
[13:29] that we're used to, which is where a
[13:30] human is actually clicking the buy
[13:31] button.
[13:32] And AP2 is working to fill that gap. It
[13:35] provides authorization, proving that a
[13:37] user gave an agent the authority to buy
[13:38] something. It provides authenticity,
[13:41] which ensures that merchants know that
[13:42] the agent's requests actually reflects
[13:44] the user's. And it provides
[13:46] accountability so that we know who's at
[13:47] fault when there's a fraud or a dispute
[13:49] over a transaction.
[13:52] And this is all achieved through these
[13:53] cryptographically signed mandates. So,
[13:55] for real-time transactions, a user
[13:57] interacts with an agent. That
[13:58] interaction is captured in a signed
[14:00] mandate that creates this irrevocable or
[14:02] unchangeable record. And the man mandate
[14:05] is a plain English description of the
[14:06] agreement between the agent and the
[14:07] user. And it's completely tamper-proof.
[14:09] It can't be modified after the fact.
[14:12] And very soon, we'll be able to power
[14:15] kind of like human not present tasks.
[14:17] So, these are things where you say say I
[14:19] want to buy a pair of shoes if and only
[14:21] if the price drops below a certain
[14:22] level. That mandate gets made up front
[14:24] with the price limits and the rules, and
[14:26] then the agent can generate that final
[14:28] mandate once it's uh those conditions
[14:30] are met.
[14:32] So, in the future, I see a lot of cool
[14:34] applications of AP2. Um it can be used
[14:36] to facilitate complex tasks like get in
[14:38] line for me as soon as these Taylor
[14:39] Swift tickets go on sale. Buy me
[14:42] groceries based off of my recent
[14:43] purchases habits and do it every week.
[14:45] And this is where I see a ton of um
[14:47] value in AP2, and it really opens up
[14:49] some kind of workflows that we haven't
[14:51] really thought of before in the past.
[14:54] So, wrapping up, if you're if you're in
[14:57] the room today, you should embrace the
[14:58] challenge, embrace the consumer trends,
[15:01] and start figuring out your strategy for
[15:02] how you're going to be enabling agentic
[15:04] commerce.
[15:06] I think for retail retailers, you'll
[15:07] hear from Hearin here in a second. If
[15:10] you haven't already, start
[15:12] preparing your catalog to be
[15:13] discoverable by agents. I think that's
[15:15] like the first thing you can do. Um
[15:17] integrating to the merchant center.
[15:19] We're going to be adding adding
[15:20] conversational attributes for better AI
[15:22] discovery.
[15:23] And then for payment processors, embrace
[15:25] the open standards. Um this will help
[15:28] you basically integrate with a wide
[15:29] range of retailers, a wide range of
[15:31] surfaces, and help you facilitate
[15:33] ultimately non-center checkout flows
[15:35] that I think will be the future of AI um
[15:38] commerce here.
[15:39] All right. Well, thanks very much. I'm
[15:40] going to hand it off to Hearin from
[15:41] Wayfair.
[15:47] >> [applause]
[15:52] >> Thank you very much. Uh thanks, Jesse,
[15:54] for a great great context setting on all
[15:55] the protocols. I am Hearin Patel, super
[15:58] excited to be here. I'm in the
[16:00] engineering lead at Wayfair. I lead
[16:02] purchase and enterprise platform, and
[16:06] super excited to share you the
[16:07] collaboration journey that we had had
[16:09] with Google and a lot of other partners
[16:11] in building the Universal Commerce
[16:13] Protocol. Uh from a focus standpoint,
[16:16] I'll focus on some of the key decision
[16:18] points that we have went through. Uh and
[16:20] also a lot of the learnings that I think
[16:22] a lot of you folks are here in the room
[16:24] might appreciate.
[16:26] Um I'll start with again a lot of the
[16:29] data points that Jesse shared, uh which
[16:31] is clear that there's a fundamental
[16:34] shift happening in commerce. And the
[16:36] mental model that I I have resonated,
[16:39] and we have learned through this
[16:40] experience, is that for decades, uh we
[16:44] retailers have been trying to perfect
[16:46] this journey of a consumer from search
[16:49] to transaction through clicks and
[16:51] through views. And that model is about
[16:53] to break. Uh today,
[16:56] uh convert customers are already sort of
[16:59] like thinking about uh engaging in
[17:01] becoming conversational. Customers are
[17:03] adopting AI. They're no longer clicking
[17:06] workflows or navigating through browse,
[17:09] category, search. They are engaging, and
[17:12] they are expressing intent. They are
[17:14] asking what they want, and they're
[17:17] they're no longer clicking the buttons.
[17:19] Uh tomorrow, in the world of agents,
[17:22] this interaction goes even further.
[17:25] Where agents are not only going to help
[17:28] customers sort of find products or
[17:29] compare products, but they're going to
[17:31] perform transactions on their behalf.
[17:33] This fundamental shift from
[17:36] clicks to conversation to autonomous
[17:38] action is where I think everybody should
[17:40] be paying attention to, cuz this is
[17:42] where I think things are going to shift.
[17:45] And we are so glad to be working and
[17:47] partnering with Google and many other
[17:49] partners to co-develop this sort of
[17:51] fundamental protocol that's shaping up
[17:54] how commerce going to be in future.
[17:57] Let's talk a little bit more about data.
[18:00] Uh Jesse shared a bit of lot of broad
[18:02] data. I want to share the data that we
[18:04] see on our agentic platform. Over the
[18:06] last two quarters, a lot of this agentic
[18:09] traffic has really grown quarter over
[18:11] quarter, almost doubled it up from our
[18:14] referrals that we are we are seeing from
[18:16] coming from this agentic platform.
[18:18] Traffic that lands on wayfair.com
[18:21] from the agentic channel seems to be
[18:23] higher higher intent. Customers are
[18:26] expressing lot of that purchase intent.
[18:28] And we see actually a positive
[18:31] conversion delta between what we see in
[18:34] our B2C channel versus the agentic
[18:37] channel. Um
[18:39] Also, the most sort of like important
[18:41] part that we are seeing is this is still
[18:44] like very early days. It's single-digit
[18:46] transactions.
[18:47] Um the traffic is still in less than two
[18:50] digits, and hence we think that there is
[18:52] a massive headroom here. And we are
[18:54] super excited to really get these agents
[18:58] learn about our catalog, our depth of
[19:00] understanding of our catalog, and hence
[19:02] I think we think there is a big upside
[19:04] here.
[19:07] With all of this sort of like grounding
[19:09] ourselves in data and the promise of
[19:10] agentic commerce,
[19:12] we in Wayfair feel that our goal is very
[19:16] clear. We want Wayfair to become sort of
[19:19] the really trusted first-party truth of
[19:23] of all things home for all of agentic
[19:25] surfaces, not just Google.
[19:28] What this means is we really need to
[19:30] focus on three things, and we have a
[19:32] three-pronged strategy here. We want to
[19:34] become the trusted home source. And the
[19:36] way we think of doing that is really
[19:38] focusing on our catalog data, making our
[19:41] catalog data really reliable, structured
[19:44] that the agents can understand, and
[19:46] hence when they can understand, they can
[19:48] navigate it better, and they can really
[19:51] help the customers make purchase
[19:53] decisions.
[19:54] Secondly, we want our catalog to be
[19:56] instantly shoppable. What that means? It
[19:59] means that our commerce stack has to
[20:01] have these APIs that are well organized,
[20:04] well governed, that can easily integrate
[20:07] into protocols like UCP. And lastly, we
[20:10] will deliver the promise. We want to be
[20:13] the source of truth on maintaining that
[20:15] end-to-end relationship with the
[20:17] customer, which is really critical in
[20:19] maintaining trust in this agentic world.
[20:24] All of this,
[20:26] you cannot do it on your own. You need
[20:27] to be in an ecosystem that is really
[20:30] powerful. And we are so glad that we are
[20:33] co-building this sort of fundamental
[20:35] protocol with Google. Our collaboration
[20:38] with Google and many other partners to
[20:40] build UCP or Universal Commerce Protocol
[20:43] is very foundational.
[20:45] What's been really, like, important that
[20:47] we have seen is the depth of
[20:49] collaboration. We're not just partnering
[20:52] with Google so that our products show up
[20:54] on Google surfaces. We are codifying lot
[20:57] of the patterns and the guardrails that
[20:59] are super critical for the retail
[21:01] industry as a whole.
[21:03] We are ensuring that product accuracy,
[21:06] transaction integrity, and trust are
[21:08] maintained.
[21:09] And what's really important that
[21:12] throughout this journey, it has allowed
[21:14] Wayfair to focus on enabling, designing,
[21:18] and launching an infrastructure layer
[21:20] and an API layer that integrates not
[21:24] only with Google but many other agentic
[21:26] platforms.
[21:30] We have a demo upcoming, but I wanted to
[21:32] show you how the whole journey is really
[21:35] changing with agentic commerce. It's a
[21:39] much more simplified view than the
[21:41] traditional sort of seven-to-eight steps
[21:43] journey that you see on most commerce
[21:46] platforms right now.
[21:47] It is becoming much more simpler.
[21:50] Customers are no longer going through
[21:52] multiple loops and bouncing between
[21:55] browse, category, search, product views.
[21:58] They're really simplifying into
[21:59] expressing intent, finding the product
[22:02] that they want, and going ahead with
[22:04] that purchase.
[22:05] And since UCP is a shared language, it
[22:08] ensures that when an AI agent is finding
[22:11] that lamp for you, it is actually
[22:13] looking into the product catalog that
[22:15] you already provided through Merchant
[22:17] Center or Merchant APIs to Google.
[22:20] UCP experience also makes sure that not
[22:23] only this experience is frictional-less,
[22:26] uh it is not faceless. Wayfair continues
[22:30] to be the seller of record. It ensures
[22:32] that when the customer clicks on buy
[22:35] from Google screen, it is a Wayfair
[22:37] transaction, and we keep the
[22:39] relationship with the customer with our
[22:42] teams from our warehouses to the service
[22:45] desk.
[22:47] Okay, I I promised a bunch of key
[22:49] decisions, and catalog readiness or
[22:51] catalog was a key decision for us.
[22:54] On this one, we did not try to overthink
[22:56] on what type of products the customers
[22:58] will be comfortable handing it off to
[23:01] the agents to buy. We actually focused
[23:04] really on filtering a very narrow set of
[23:06] products with that we thought that
[23:08] agents will have trouble with. Things
[23:10] like, uh items that are only available
[23:13] in certain states, or items that are
[23:16] having special restrictions. We focused
[23:18] on just filtering those items, and we
[23:20] enabled our entire 30 million
[23:22] product catalog for agentic
[23:24] integrations. We wanted to expose this
[23:27] entire catalog primarily for two
[23:28] reasons. It's a new channel. It's a new
[23:31] interface. We wanted to learn how
[23:33] customers will actually shop through our
[23:35] breadth of the catalog. What are the
[23:37] learnings it will provide to us as they
[23:39] engage with that? And if you are already
[23:42] part of Google's enterprise, you're
[23:44] integrated with Merchant API, onboarding
[23:46] was very easy.
[23:49] Okay. I'm going to show a demo that
[23:52] should
[23:53] talk.
[23:59] Oop.
[24:00] There you go.
[24:02] As you can see, the experience is really
[24:05] intuitive. It's fast.
[24:08] It's engaging. It allows the customers
[24:11] to really get narrower with that
[24:13] purchase intent.
[24:14] And it's all integrated. It's a click of
[24:18] a button uh where they get the consent
[24:20] and place the order.
[24:36] Great. Um another decision point that we
[24:40] ran into, and lot of you who are
[24:41] thinking about an agentic commerce, you
[24:44] will run into, is what should be your
[24:45] agentic architecture.
[24:47] And
[24:49] the core principle we tried to follow is
[24:52] agents don't want to get redirected to
[24:54] your websites. They want to perform that
[24:56] transaction directly where they are.
[24:58] And the obvious trap that we could have
[25:00] got into is try to build sort of bespoke
[25:03] checkout layers that work for different
[25:06] agentic platforms. And while that could
[25:08] have worked, we quickly realized that it
[25:11] could actually get really messy very
[25:14] fast, hard to scale, hard to manage, and
[25:17] over time it could drift away from what
[25:19] our Wayfair internal B2C architecture
[25:22] looks like. So, we made a different
[25:23] decision. Instead of creating something
[25:25] new, we had an existing partner API
[25:29] ecosystem, with a partner API
[25:31] architecture that already was enabled
[25:34] for marketplaces, for other partners.
[25:37] So, we thought about treating agent as
[25:39] yet another partner.
[25:42] This helped a lot. At a high level, this
[25:44] architecture is very simple.
[25:47] On the left, you see agents, you see
[25:49] partners. In the middle, you have the
[25:51] partner API architecture layer, and on
[25:54] the right, you have all the core APIs
[25:57] that Wayfair already has for its B2B and
[25:59] B2C channels. The middle layer
[26:01] essentially has two key concepts. One is
[26:04] a core access layer that allows you to
[26:06] give authentication, identification to
[26:09] plug into your core layer. And the
[26:11] second layer is a protocol translation
[26:13] layer that allows you to take
[26:14] requirements from UCP and translate it
[26:17] for your own uh internal uh systems and
[26:20] internal integrations. This gave us a
[26:23] lot of speed. We were able to launch
[26:25] this in less than a quarter.
[26:27] We were able to apply lot of this to
[26:29] scale to other partners. We are
[26:31] integrated with Perplexity and the next
[26:34] AI platform uh to come.
[26:37] And third is it allows us to maintain
[26:39] safety and security. We control still
[26:42] fraud, payment, fulfillment, and all of
[26:45] those integrations so that the trust is
[26:47] maintained with the customer.
[26:52] Uh fast forward couple of months, we
[26:53] have launched in in Q1. We have some
[26:56] great learnings I want to share. Uh two
[26:59] key learnings from a customer behavior
[27:00] standpoint, I think really, as I said,
[27:04] uh the purchase intent is really strong.
[27:06] Our customers are coming up with a very
[27:08] strong intent. Mhm. So, when they land
[27:11] on the product view, they're really
[27:13] clear on what they want. And hence, not
[27:15] only conversion is is increasing, but
[27:18] also they are ready to walk away.
[27:21] So, the bounce rates are higher. So, you
[27:23] see two sides of those uh two dimensions
[27:26] of uh
[27:27] uh any sort of commerce API. Your bounce
[27:29] rates are higher, your conversions is
[27:30] higher. What that tells you is you
[27:33] really need to focus on your catalog.
[27:35] You need to understand how the agents
[27:37] are able to read your catalog both from
[27:39] a breadth and a depth standpoint.
[27:41] Pricing, imagery, availability,
[27:43] fulfillment, lot of the core core things
[27:46] really need to be in a shape where your
[27:48] agents can parse them so that it can
[27:50] match the intent of a customer back to
[27:53] your catalog.
[27:54] Secondly, this agentic experiences and
[27:57] LLM-powered experiences are really
[27:59] great. They're very powerful. They're
[28:01] able to understand the breadth and depth
[28:02] of your catalog, and they're able to
[28:04] narrow the intent of the customer very
[28:07] clearly, very fast. And hence, what we
[28:10] are able to sort of like offer back from
[28:12] our experience standpoint is also great.
[28:15] Bottom line uh is the customers'
[28:17] behaviors are changing. Um you as
[28:20] retailers really need to think about how
[28:23] fast your strategy is going to change to
[28:24] match the speed of what the customers
[28:26] are learning about discovery, about
[28:29] integration, and also about engagement.
[28:32] And hence, I think um just keeping pace
[28:35] to it, not thinking too much, uh really
[28:38] trying to adopt, it is the message that
[28:40] I want to give.
[28:41] Lastly, as we say in Wayfair terms, we
[28:44] are never done. Uh on the left, you are
[28:48] see a bunch of things that we launched
[28:50] nearly Q1. Uh our 30 million Wayfair
[28:54] products are discoverable and
[28:55] purchasable on Gemini and AI mode. Great
[28:58] learnings on purchase intent, traffic
[28:59] growth, and customer order growth. Uh in
[29:02] upcoming quarters, we are actively
[29:05] working with Google on three main focus
[29:07] areas. First is sort of like really
[29:10] evolving our UCP integrations so that we
[29:12] can focus on basket building. Right now,
[29:15] we have simple purchases. We want to
[29:16] enable multi-item cards and so forth.
[29:19] Promotions are key to commerce, and
[29:21] hence, I think really enabling
[29:22] promotions will go a longer way. And
[29:24] then, we want to keep on working on
[29:26] catalog understanding so that we can
[29:28] keep on enabling newer and newer
[29:30] workflows to increase the catalog depth.
[29:32] With [snorts] that, I want to say lots
[29:33] of thank you to Google team who worked
[29:36] on UCP. Uh and a great shout-out to
[29:38] Wayfair on the customer technology team.
[29:40] Some of the folks are here in the room
[29:42] uh that has helped us enable this.
[29:45] Uh and our supply chain team who who
[29:46] extended the partner API uh integration
[29:49] layer. With that, I'm going to hand it
[29:51] off to Michael.
[29:53] >> [applause]
[29:59] >> Okay.
[30:01] Good afternoon, everyone.
[30:03] Let me just check.
[30:05] Okay, so Mike Edwards.
[30:07] I have
[30:08] product strategy responsibilities in the
[30:11] merchant solutions business
[30:13] at Fiserv. Let's just check we have
[30:16] slides moving. We do. Just 30 seconds on
[30:17] Fiserv for those of you that may not
[30:19] know what we do. We're a global
[30:22] financial technology
[30:24] company. We sit at the intersection
[30:26] of banking and of the merchant business
[30:29] of commerce.
[30:30] We support thousands of the world's
[30:34] largest financial institutions
[30:36] and millions of merchants
[30:38] all around the world. All the way from
[30:41] small business owners
[30:42] through our Clover point-of-sale system
[30:45] and commerce platform
[30:47] all the way up to the largest
[30:50] brands in the world including
[30:52] some of the largest retailers,
[30:55] quick-service restaurants, digital
[30:57] brands, grocers, and and and and more.
[31:01] So, okay, let's
[31:02] lean a little bit into agentic. McKinsey
[31:06] frames
[31:08] agentic commerce as a seismic shift as
[31:12] agents start to autonomously navigate
[31:16] the shopping experience.
[31:18] And they think this could be a
[31:19] trillion-dollar
[31:20] industry by 2030
[31:23] as we think about US retail.
[31:27] There's some other estimations on the
[31:28] screen as well that don't hit a
[31:30] trillion, but the potential is clearly
[31:32] huge. And I'm going to spend a bit of
[31:34] time today looking at what are some of
[31:37] the enablers to get to that scale
[31:40] from the lens of a payments payments
[31:42] company, payments and trust.
[31:44] Um
[31:46] And as we go through, we'll look at some
[31:47] of the issues that
[31:49] we face in reaching that scale as well.
[31:54] Another way to look at this, I think
[31:55] actually it's probably quite easy for us
[31:57] to enable an an agent to go make a
[31:59] payment. I think where there are
[32:01] concerns
[32:02] from merchants across the ecosystem is
[32:05] can we do that? Can we enable that
[32:07] payment to take place in a way that
[32:09] preserves the
[32:11] authorization integrity that doesn't
[32:14] introduce new fraud that's uncapped. Um
[32:18] and perhaps equally as importantly
[32:21] doesn't change our margins, the way we
[32:23] operate,
[32:24] doesn't change significantly any of the
[32:26] payment economics.
[32:31] Okay, so
[32:32] I'm going to talk a little bit about the
[32:35] agentic landscape. I actually think
[32:37] about this landscape and you could group
[32:39] these boxes a little bit by these three
[32:41] layers. The first layer is an
[32:43] intelligence layer. So, think about the
[32:45] product catalog,
[32:47] think about intent discovery.
[32:52] Agents and protocols.
[32:55] Then I would say we also have the
[32:57] experience layer.
[32:59] So, this is
[33:01] the place at which consumers are going
[33:03] to enter the ecosystem, the agentic
[33:06] surfaces.
[33:07] And then finally, as we move to the
[33:09] right side of this the screen
[33:12] we'd look at the transaction layer. And
[33:15] this is payment processing, but it's
[33:17] also post-purchase
[33:19] order management.
[33:22] I think Jesse and Hitendra spoke a
[33:24] little bit about what I would consider
[33:25] to be the intelligence layer and the
[33:27] experience layer, so I'm going to focus
[33:29] on
[33:30] the transaction layer
[33:33] where success I think requires three
[33:35] things. It requires trust, it requires
[33:37] security, and it requires transparency.
[33:42] And I think we can
[33:43] dig dig into these items with a number
[33:45] of key
[33:47] principles.
[33:48] The first of which is
[33:50] we have to have the identity and
[33:52] permission has got to be rock solid.
[33:55] If we're going to allow an agent to act
[33:58] we have to understand who that agent is
[34:00] acting for
[34:01] and what rights and limits have been
[34:04] given
[34:05] to that agent. And so, tokenization
[34:09] delegated permission, strong
[34:12] authentication become really important.
[34:15] Clearly, we can't give agents the free
[34:17] reign.
[34:18] They have to act with a bounded
[34:19] authority and that authority should be
[34:22] managed by the consumer, by the
[34:24] customer.
[34:27] Every agentic action should have an
[34:30] audit trail, should be auditable.
[34:33] If an agent makes a purchase
[34:35] merchants, customers, the networks need
[34:39] to understand why that purchase was
[34:41] made.
[34:46] And
[34:48] need to understand why that purchase was
[34:49] made. And so,
[34:51] things like having clear logs
[34:54] having intent signals
[34:57] and understanding the transaction
[34:59] context become really important. The
[35:01] transaction context helps when you get
[35:03] into the post-order phase when you look
[35:06] at chargebacks and disputes even as you
[35:08] feed into your loyalty program that's
[35:11] important so that those items are not
[35:13] just they don't become
[35:14] guesswork. And transparency across the
[35:17] ecosystem as far as that audit trail is
[35:19] concerned is important.
[35:23] Fraud controls need to evolve from
[35:27] protecting data
[35:29] to protecting decisions.
[35:35] In the agentic landscape
[35:38] it's not just okay, it's not enough
[35:39] anymore just to protect an unauthorized
[35:42] card
[35:43] use. We now have to protect unauthorized
[35:46] autonomous decisions from being made.
[35:49] And so, agents that are verified
[35:52] agents that are trusted
[35:54] where the customer has defined what
[35:56] those agents can and can't do
[35:58] becomes critical. And equally, merchants
[36:00] need instantly to understand if that
[36:03] agent is verified or if in fact they're
[36:05] under some form of automated attack.
[36:09] And then finally, I would say standards
[36:11] matter
[36:13] in our ecosystem. I think collaboration
[36:15] matters even more.
[36:17] You have the networks that are building
[36:19] trust frameworks. You have the platforms
[36:21] that are building protocols.
[36:24] You have merchants that are working
[36:25] across the ecosystem.
[36:28] And our role at Fiserv certainly we see
[36:31] our role in helping to bring that
[36:33] ecosystem together
[36:35] so that trust can be applied
[36:38] consistently across the entire ecosystem
[36:40] everywhere and not just in pockets.
[36:45] And so
[36:47] a bit more about
[36:48] Fiserv
[36:50] we kind of view agentic to be successful
[36:53] where the an agent can act transact
[36:56] securely and a merchant can understand
[37:00] who acted, why they acted, and with what
[37:02] authorization they they acted. And there
[37:06] are a number of challenges as we see
[37:07] them.
[37:08] Jesse touched on this in his
[37:10] presentation.
[37:11] The first challenge is actually the
[37:13] ecosystem itself. It's fragmented. We
[37:16] have multiple protocols. We have
[37:17] multiple standards. Those standards are
[37:19] evolving quickly. I think it is a valid
[37:22] concern for merchants to understand if
[37:25] they are expected to integrate to each
[37:28] and every platform, to each and every
[37:29] standard and if they're expected to
[37:31] manage and maintain
[37:33] those connections and those standards as
[37:36] they evolve.
[37:38] Finding a partner to help simplify and
[37:41] remove that complexity
[37:44] is extremely important.
[37:47] The second challenge is actually
[37:48] omnichannel itself. How do we create
[37:50] consistent experiences for our customers
[37:53] as they move across channels? We used to
[37:56] thinking about this in the payment world
[37:58] from an in-store to mobile to e-com.
[38:02] And as we introduce agentic as
[38:04] effectively a new channel, we should
[38:06] take the same approach and apply the
[38:08] same approach.
[38:10] If we can keep our existing tech stack
[38:13] and use it across all those channels
[38:16] we can deliver the same consistent
[38:18] payment types.
[38:20] We can remember the client through
[38:21] loyalty programs and we can retain that
[38:23] connection
[38:25] to the clients and we can do it at
[38:26] minimal cost.
[38:29] I think keeping the consistent tech
[38:31] stack, keeping customers happy with
[38:34] consistent experiences
[38:36] you also keep your CTO and your CFO
[38:39] happy as well and you you retain the
[38:42] investments that you have made. You
[38:44] leverage the investments that you've
[38:46] made in your existing tech stack. Things
[38:48] like optimization
[38:51] programs and such forth.
[38:54] And then finally,
[38:56] trust and risk.
[38:57] So, as I said before, you're not just
[38:59] protecting payment data, you're
[39:00] protecting decisions.
[39:03] Being focused on identity and trust is
[39:07] is super important.
[39:11] A little more granular detail.
[39:15] This is what a an enablement layer can
[39:18] look like
[39:20] that brings together all of the
[39:21] ecosystem
[39:23] partners into one place. And so, for the
[39:26] merchant, the outcome is that they get
[39:27] to integrate once.
[39:30] They can stay ahead of the evolving
[39:32] standards that can be managed at the
[39:34] center.
[39:35] And they get to
[39:36] preserve payment integrity
[39:39] end-to-end.
[39:41] It is actually worth saying as we look
[39:43] at this, Google has
[39:45] um
[39:46] uh, some real strengths to bring into
[39:48] this uh, ecosystem.
[39:50] Uh, if you think about the Merchant
[39:52] Center
[39:53] um, as an on-ramp, many of your
[39:56] organizations will already be using
[39:58] Merchant Center today. You'll already
[40:00] have product catalogs loaded. They can
[40:03] be leveraged and extended into the
[40:05] Agentech ecosystem without having to
[40:08] recreate them.
[40:10] Uh, uh, UCP uh, Jesse talked to him and
[40:13] talked to
[40:14] open standards. It's going to allow you
[40:16] to bring more of the things that you
[40:18] know about your customer,
[40:20] loyalty and tender types, uh, into into
[40:23] the ecosystem as well.
[40:26] At Fiserv, we think about
[40:29] um, preparation is about being
[40:32] transaction ready.
[40:34] And so, just a few areas to to finish
[40:37] with, areas to focus on as you're
[40:38] thinking about embarking on
[40:42] your Agentech journey.
[40:43] The first is data and and both my
[40:46] colleagues talked about
[40:48] um, uh,
[40:49] cleaning data, clean data from a product
[40:51] catalog point of view. Absolutely the
[40:54] same applies from a payment lens as
[40:56] well. Having clean feeds, clear pricing,
[40:59] mapping your customers uh, to their
[41:02] preferred loyalty and and uh, and tender
[41:04] types,
[41:05] uh, will allow agents to transact
[41:07] reliably and consistently. And as I
[41:10] think about
[41:12] tender types and I think about loyalty,
[41:14] um,
[41:16] make sure you bring these. Number two,
[41:18] make sure you bring these into
[41:20] into the Agentech ecosystem.
[41:23] Agents are going to change the
[41:24] relationships that you have with your
[41:26] customers.
[41:27] You're going to have shopping journeys
[41:29] that start on third-party uh, sites
[41:32] rather than your website, which means
[41:33] you're going to lose some visibility
[41:35] into that customer.
[41:37] It's really important that you bring
[41:40] into the ecosystem the things that make
[41:42] your customer your customer,
[41:45] understanding who they are from a
[41:46] loyalty standpoint, understanding who
[41:48] they are and how they prefer to pay are
[41:51] ways that you can continue to keep the
[41:53] connection with your customer. And so,
[41:55] you can prepare now by
[41:57] starting to map, have an understanding
[41:59] of my customers' preferred payment
[42:01] types,
[42:02] uh, stored cards, all of those things,
[42:05] uh, to be ready to bring that into into
[42:06] the ecosystem.
[42:08] Uh, and then, connectivity.
[42:12] Make sure you find a partner that will
[42:13] help you come into this ecosystem and
[42:15] access multiple Agentech layers,
[42:19] Agentech surfaces,
[42:20] uh, and remove some of that complexity
[42:23] into your business and hopefully allow
[42:25] you to
[42:26] retain and make use of your existing
[42:28] technology and the investments that
[42:30] you've made in your tech stack.
[42:33] Uh, and then finally, I would say
[42:34] testing.
[42:35] Merchants that can test now, that can
[42:37] experiment now, uh,
[42:39] in in its nascent form,
[42:41] uh, will learn what works for them, what
[42:44] breaks for them, and as adoption
[42:47] continues to drive and grow, we'll be in
[42:49] a better place to understand how to
[42:51] drive revenue uh, for for their
[42:53] organization. So, pick a test group, uh,
[42:57] target a test group as you can. It could
[42:59] be a sub-segment of your loyalty base
[43:01] and begin uh, and begin some testing and
[43:04] some experiments and learn.
[43:07] Uh, and I'm going to wrap there and just
[43:08] say that at Fiserv, we're focused on
[43:11] continuing to enable merchant
[43:13] discoverability
[43:15] in the Agentech uh, world
[43:17] uh, and making sure that you are
[43:19] transaction ready whenever an agent
[43:22] turns up.
[43:23] Uh, so thank you.
[43:27] >> [music]