Latest / Retail Media Breakfast Club / AI Shopping Is Here: How LLMs Are Rewriting Retail Discovery [Agentic Commerce Expert Series Part 1 of 4]
Transcript
- Ep 1 - Discovery has moved upstream, and what that means for retailers === [00:00:00] Kiri: LLMs aren't just answering questions, they're becoming the first step of shopping. And that doesn't mean retailers disappear. It means intent and decision making. Relocate upstream expert series. [00:00:15] That we're about to get into is sponsored by Miracle, and I'm joined by Amelia Van Camp, head of Agen Commerce at Miracle. [00:00:23] To break down what's really happening with discovery and what retailers should do [00:00:30] first before they chase shiny AI assistant launches. Welcome Amelia. [00:00:35] Amelia Van Camp: Thank you. I'm very excited to be here. [00:00:38] Kiri: So let's first talk a little about your role and what it tells us about Miracle's relationship with ag [00:00:45] agentic commerce. I don't, I don't hear head of ag agentic commerce a, a whole lot, uh, yet, and a lot of people know Miracle for its marketplace platform. But tell us a little bit about what the scope of, um, your role is with Head of a agentic [00:01:00] commerce at Miracle. [00:01:01] Amelia Van Camp: It's a very good question. Um, and you're spot on. You don't hear the job title too much right now, but I suspect that will change, uh, in the coming months and years here. So Miracle has historically been a marketplace company [00:01:15] and when Ag Agentic Commerce started. Really becoming a topic of conversation. I would say in the past 18 months, we had an opportunity to think about how we wanted to play in the space. [00:01:26] And we backed up and said, you know, beyond marketplaces, what have we been really [00:01:30] strong at? And we said, we've been very strong at multi merchant order orchestration. And within that world we've built several of our products to be AI first, which I'm very lucky in my role that that is the scenario that I've stepped into.[00:01:45] [00:01:45] So taking that. That first order of business and these products that we've built, we said, what do we wanna add on top that can enable agent commerce? Specifically from a product standpoint. Uh, and then my role stepping into that is [00:02:00] how do you actually bring that to life? How do you commercialize that? [00:02:03] What partners do you choose to enact certain functionality that maybe you're not necessarily going to build yourself, where they have strong domain expertise? Uh, so my role today is primarily US focused, where we [00:02:15] see most of the ENT commerce world evolving. Uh, but I suspect that that will. Edge out into other markets very soon here. [00:02:25] Uh, so my role is a bit of go to market, um, but also figuring [00:02:30] things out on the ground with our customers. [00:02:32] Kiri: Yeah, there's a lot to figure out in real time. Well, let's, let's lay a little bit of the, the groundwork here and talk about what people are actually using LLMs for today when it comes to shopping. I just wanna call out a [00:02:45] really cool stat from a Costa group and they did a, a, a shopper panel where they found that. More than 50% of people used an AI assistant during their Thanksgiving day shopping trip in person last year, and [00:03:00] within that it was like 54%. And within that group, 18% of those shoppers were using. A chat, GPT or a Gemini, um, and a much larger percentage of people were using a retailer, [00:03:15] uh, AI assistant, like a Rufuss or a Sparky or a Target, um, in, in shopping app. [00:03:21] So I just find that so interesting that we're not just talking about e-commerce here, we're talking about all forms of shopping, including in, in the store as well.[00:03:30] [00:03:30] Amelia Van Camp: Yeah, I agree. Um, the dynamic has certainly changed. One of the things that. I frequently do when I'm able to speak in front of a live audience is I ask people by raising their hands, how many of you have used an LLM to start your shopping [00:03:45] journey? Whether that's product discovery, brand discovery, and you usually see about 80% of the group raise their hand. [00:03:52] Then what you ask as the next question is, okay, how many of you have actually purchased products on LLMs? And you'll s you'll [00:04:00] see a smaller number of hands go up. But what's interesting is in a four month time span. The number of hands that are raised continues to increase for both of those questions, and the domain in which the customer [00:04:15] is essentially enacting an agentic commerce experience can be on an LLM. [00:04:20] But as you've rightly pointed out, it can also be on a retailer's shopping agent that they've built. And so I think what, [00:04:30] what this all means is that we as consumers. Are trying to find the best way to find the products and the brands that we need based on the problem solving situations that we're finding ourselves in. [00:04:41] Kiri: So why, you know, if we're talking about, [00:04:45] LLMs as discovery engines, we, we've always had methods of discovering products and brands offsite, like TV ads, radio ads, social media. [00:05:00] What is different about an LLM? Being the place that we're discovering and and deciding what we're gonna buy. [00:05:09] Amelia Van Camp: Well, I think if you take it from the lens of the consumer, I'll take it from my own view and how I [00:05:15] use LLMs. I think what many of us are looking for is ease of use, right? Where can you find the most access to information in the easiest? Forum possible, and LLMs really create this environment. It's a very unique [00:05:30] environment. [00:05:30] how we see the landscape shifting is, as you've rightly pointed out, this topic of discoverability essentially for brands and retailers inside of that LLM, but then even more importantly, the topic of ranking. [00:05:45] How do you ensure that your brand. Your name, that your products are actually populating in the result that's going back to the user. [00:05:53] And that can be, I think, a bit of a, a scary point for many of these organizations to be entering [00:06:00] into. Because what we see is that the LLMs, the agents are pulling from multiple different information sources. They're not just pulling from your front end system, they're pulling from Reddit, they're pulling from YouTube, they're pulling from [00:06:15] Wikipedia, and so now all of a sudden, it's frankly an explosion of data sets that a single LLM. [00:06:23] Is using to make a decision. And so we're seeing more retailers and brands kind of ask themselves, how do I [00:06:30] influence that and where do I even start? So it's a big question. Um, we have a pretty strong opinion on it, but. [00:06:36] Kiri: Yes. Well, let's get into that. We're gonna jump around a little bit. So when an LLM is scraping a retailer, maybe that's not the term that you would use, [00:06:45] but you can correct me on that. What is it looking for and what gets missed when the product data is not really set up in an AI friendly way? On the retailer.com. [00:06:55] Amelia Van Camp: Yep. It's a great question. So, um, our chief data and AI [00:07:00] officer explains this, I think beautifully. So she always explains when. Engineers at these companies, at these LLM organizations built these agents. They basically built them to think in some capacity about [00:07:15] trust like children. And the precedent that's been set is you need to look for certain identifiers before you trust a product, before you trust a brand. [00:07:26] And so there's a bit of a gatekeeping factor. So when an [00:07:30] AI agent ends up on your site and it's scraping a product, it gets onto the product display page, but it's looking holistically at the context that the user has put into their search. So are they more sensitive to price? Are they more [00:07:45] sensitive to delivery time? [00:07:46] Are they more sensitive to style? It's taking that into consideration, but then it's also looking at that product and the retailer or the brand that's selling it. And it's saying, can I trust you? Do you have accurate price [00:08:00] information? Do you have strong ratings and reviews? Do you have shipping details? [00:08:04] So all of these different facets that frankly have existed for a long time in e-commerce now become certainly equally important, if [00:08:15] not even more so important in that game of discoverability and ranking to ensure that your products are populated. [00:08:22] Kiri: So these are the sort of intent based attributes coming to the fore here as well. 'cause we're not just talking [00:08:30] about keywords anymore. [00:08:32] Amelia Van Camp: Yeah, so intent-based attributes, um, are honestly another ball game. So you have your baseline attributes, your baseline product data. That's something that we strongly advocate for keeping. [00:08:45] As, as rich, um, and as consistent as you possibly can. That's kind of your first big building block. To this notion of trust with these AI agents, which gets you again to discoverability and ranking. [00:08:59] Then on top [00:09:00] of that, the way that we as consumers today use LLMs is conversation based, right? Which is very different from historic keyword searches, which maybe are two to three words today. It's multiple sentence sentences, multiple paragraphs. So [00:09:15] the user of an LLM is bringing in much more context into their search, and that impacts. [00:09:21] This whole notion of attributes that you add or can't add into your, your catalogs, your product catalogs, you have your baseline set of [00:09:30] information, which you've had for, again, pretty much since the existence of e-commerce. Then what you can add in is intent-based attributes and intent-based attributes fall into this category that's commonly known as GEO or a EO. [00:09:44] So [00:09:45] generative engine optimization, uh, ag agentic engine optimization. Intent-based attributes are basically attributes that are built based on conversation. So they include context, they include q and a, [00:10:00] they include situation setting that is not typically added into a product catalog in a standard setting today. [00:10:08] So when you have a very strong set of core data, strong set of attributes. You've [00:10:15] ensured that you've detected pricing anomalies, missing information, that's your first building block, and when you add these intent based attributes on top, what you're doing is almost gamifying the system a bit more to optimize for your chances of your product to [00:10:30] end up back in that conversation. [00:10:32] As a result, that's populated because you're able to contextualize that product inside of the conversation that the user is having With the LLM. [00:10:42] Kiri: ~So I think what we're sort of covering here is that retail, retail sites don't die. Um, it is just a, a news. Set of decision making, uh, activities that move upstream. Retailers are seeing this referral style traffic from LLMs. Amelia, thank you for your time. We're going to next, sorry, let, just, let me just restart that.~ [00:10:42] ~I'm just, I'm trying to do the segue. Um. ~So I think the overall theme here [00:10:45] is retail sites don't die. The LLMs need the retailer sites just as much as consumers are coming to rely on the LLMs themselves. It's the discovery and the decision making that at least part of that is moving [00:11:00] upstream. So in the next episode of this series, we're gonna talk about what happens when shoppers arrive on a single. PDP that are, and they've already decided what they're going to buy. The PDP becomes the new homepage, which has some [00:11:15] implications for retailers, but also creates new opportunities. Thank you so much, Amelia. We'll catch you next week. [00:11:22] Amelia Van Camp: Thank you. [00:11:23] [00:11:24]