Latest / Retail Media Breakfast Club / "We're Not Dabbling Here": Amazon's Official Company Line On AI
Transcript
- AI updates from AMZ earnnigs call === [00:00:00] We are not dabbling here, said Amazon, CEO. Andy Jassy on the company's earnings call on Thursday. Well, that is an understatement. Jassy pointed to a thousand plus AI applications that are currently being built at Amazon right now. As he puts it, if you believe your mission is to make customers' lives easier and better every day, and you believe that every customer experience will be reinvented with ai, you're gonna invest very aggressively in ai, and that's what we're doing. Thursday's earnings call was. Packed with AI news and announcements. I'm gonna break the key ones down for you and what they mean for e-commerce and retail media. Let's go. Good morning and welcome to the Retail Media Breakfast Club. I'm your host, Kerry Masters, and every weekday at 6:00 AM Eastern, [00:01:00] I bring you candid and real discussions in the world of retail, media and marketplaces in just 10 minutes. So bring your own cup of coffee and we'll get smarter about retail media together. First thing to know here is that Nova moves to Frontier Tier. Amazon's homegrown Nova model has been upgraded to compete with top tier models like Claude three and LAMA four. Andy Jassy describes it as delivering frontier intelligence with industry lead leading price performance. What is frontier intelligence? Until now. Nova was Amazon's workhorse model that was optimized for cost per token by labeling the latest version Frontier Intelligence. Jassy is signaling that Nova now competes with these top tier models on reasoning, depth and context window, not just price. So the translation for us in [00:02:00] e-commerce land is that Amazon can keep inference inside its own house while still powering higher order tasks, like multi-language product summaries. Building bundles and agentic shopping conversations all without renting open AI or anthropic technology. And this comes back later in the discussion around scalability and price in a big way. Next model update is nova. so Novas, so is Amazon's speech to speech foundation model. So think of Alexa most obvious use case here, speech to speech. Um, and what they've done here is address some key barriers that prevented voice shopping as well as voice actions and a whole variety of diff other contexts. There was some pretty [00:03:00] major barriers with with voice interactions in the past. But they fixed a lot of them. One is lower word error rates, so that's basically understanding people the first time that they ask a question or issue a command. Next is more natural sounding responses. And finally, ability to maintain conversation. Context. So all of these things that were weren't in place previously meant that voice shopping just didn't work very well. There were transcription errors, robotic sounding responses, and a lack of conversation memory. But the updates to Nova, so tackles these issues head on it understands requests the first time. Responds with human-like speech patterns and maintains context across multiple exchanges. So it'll remember things [00:04:00] about you and your family and your preferences and what you like and don't like and what your budget is and things like that. So that kind of fluency is critical for consumers to trust AI with purchasing decisions, especially in screen free environments like in your kitchen or in your car. And the thing that interesting thing here is that when you look at conversational AI and agentic AI landscape. It's all text and image driven, right? Everyone else is Drive is developing text-based chatbots, but Amazon's investment in Nova Sonic so shows a distinctive emphasis on voice commerce, which is tapping into a huge competitive advantage that they have in Alexa. Alexa is in millions of homes, and that isn't an asset that any other competitors have. All right, let's move on to updates around [00:05:00] agents and applications. So of course, we've gotta talk about Alexa plus first. Alexa plus in case. You're not familiar yet, goes beyond just answering questions. It can now take actions for you. It can make a potential shopping concierge, uh, that can compare products, add items to cart, apply coupons and arrange returns. Can also do things like finding contractors for home improvement jobs. And telling you what movies are on at a theater, you and actually buying those tickets and things like that. So this is currently rolling out in the us it's in beta, amazon says that global expansion is planned later this year, and in Jassy's own words, people are really liking Alexa plus this far. So still a little cagey on adoption, which isn't anything new. But I [00:06:00] think you know what this means is these subscription agents could disrupt product discovery, um, potentially, but I think it's more likely a use case for repeat purchases, right? Better to automate routine purchases with instructions like. Keep my household stocked with the cheapest gluten-free cereal and switch brands if unit prices rise by 10%. And so this creates a new battlefield where brands compete to become the agents to fault recommendation, a new kind of invisible shelf. Second major AI agent update is NOVA Act. So Nova Act is still in research stage. It's just a seller in a seller developer kit stage right now. It's not a consumer focused application, but it is a [00:07:00] browser action agent, which can, uh, take multiple steps, um, on, on actions and. What they've done, what the big highlight from the earnings call is that they have improved accuracy on multi-step tasks from today's 30 to 60% accuracy to over 90% accuracy. And this is interesting because if you watch the launch video for Nova Act, they don't talk about commerce at all. It talks about finding an apartment on a internet, um, real estate search engine. So. This is sort of the, the bones of some of these other agents that we're seeing. We're seeing Amazon develop the Buy for me agent that would be built off of Nova Act kind of engine. So. [00:08:00] Yeah. As Nova Act matures, Amazon agents can do this kind of price shopping across multiple websites, then execute the best deal automatically. So this takes what we have seen so far, the very nascent use case of buy for me, and you know, you're looking for a pair of Nikes. Amazon will price shop across the web and offer you products that are not sold on the third party marketplace or through the first party relationship, potentially even picking up a affiliate commission from those sales. So what's interesting here from the earnings call is the fact that. The accuracy has improved so much. So this is definitely one to watch. and the final big piece of news is just around the cost reduction that Amazon has been able [00:09:00] to pull out from ai amazon highlighted significant cost reductions for AI inference thanks to their Tranium 2 two chips. So what does this mean? For us. unlike competitors, Amazon is building across the entire AI stack from custom silicon to user experiences, giving them potential cost and integration advantages over companies focused on just one layer. So again, make this comparison back to AWS. They're really owning all the pipes that they can. And this is driving a big cost reduction for Amazon that they'll obviously use for themselves, deepen their margins, but also pass on to enterprise users of this technology so that developers and companies are using Amazon's AI technology ahead [00:10:00] of. Anthropic and open AI and all of the competitors as well. The takeaway here is they're really investing in these frontier models, the voice first interface and action oriented agents. Those are the three big themes I want you to. Remember here, they're not just enhancing their current business, they're anticipating that the way people discover, evaluate, and purchase products is going to change. And so for retailers and advertisers and brands out there, the message is pretty clear. Prepare for a world where conversational interfaces replace traditional search boxes where shopping agents make routine purchases automatically. And where dynamic personalized content becomes the norm, those who adapt their strategies to this AI driven commerce landscape will thrive. Those who [00:11:00] don't may find themselves increasingly invisible to tomorrow's shoppers. That is it for today. I hope you enjoyed this episode of the Retail Media Breakfast Club. We meet here every day at 6:00 AM Eastern, so be sure to subscribe and get your daily dose of insights, trends, and hot takes in the world of retail, media and marketplaces.