Amazon VP on Rufus and the Future of Search

(upbeat music) - Hi everyone, welcome to another episode of the Tech Disruptors Podcast. My name is Punum Goyle. I am a senior retail equity analyst at Bloomberg Intelligence, part of Bloomberg's research department with 500 analysts and strategists working across all major world markets. We are delighted to have Rajiv Maheta as our guest today. Rajiv as Vice President of Conversational Shopping at Amazon, where he leads the development of Rufus, the company's AI-powered shopping assistant, helping customers discover and buy products in a more natural, intuitive way. Rajiv, welcome to the podcast. Great to have you here. - Thanks, I appreciate you having me today. - Great, so on today's episode, we'll do a deep dive into how technology and especially AI is disrupting retail. Changing the way consumers interact with brands and transforming how companies like Amazon are building the next generation of digital shopping experiences. So let's dive right in. Rajiv, Amazon has always been known for its culture of innovation. Tech advancement has long been part of the company's DNA, but this current moment feels especially transformative. As we think about disruption, particularly in retail today, what feels different about this moment to you, particularly with AI reshaping how people discover and shop for products? - Sure. I think AI is making many new features and services available to customers. You're not only a faster way, but in a more personalized way as well. So what AI is doing is that it's enabling Amazon to bring all of its selection, all of the convenience of fast delivery and product discovery to customers using their natural language. So being able to find the perfect holiday gift for my 10-year-old would used to be a preferably arduous task has become one of many things that Rufus can help you with much more delightfully today. - Great. So let's just go back to the starting point of Rufus. You're a team-launched Rufus in about, I believe it was early 2024. So tell us where the idea came from, what gap in the customer experience were you trying to solve for or disrupt when you launched to Rufus? - Sure. Yes, we launched Rufus back in early 2024. And really what we were trying to understand is that as customers were interacting with Amazon, we saw, we're starting to see a shift in what they're looking for and how they're searching for products. And what we wanted to do was make it easier for customers to just tell us what they wanted using their own vernacular, their own words, their own way of describing something. And actually kind of satisfying that request in the most kind of helpful way possible. So yeah, some customers will search for shoes and they'll find the shoes that they're looking for. But others, as we started to see, will actually say, what are the best shoes for running on an outdoor track in the winter in Seattle? And that's a fairly complex answer there, right? Because we've got to take into account like that you're in Seattle, who you are, maybe the size of shoes that are right for you. We may know a little bit about, like, are you a runner that runs frequently or are you a beginner? So we saw that as kind of an unmet need, right? And that was kind of like the, a lot of the kind of initial hypothesis behind Rufus is that you know, could we build something that made it just easier for customers to find that perfect product on Amazon? 'Cause we know that the selection is available, but sometimes finding it could be difficult. And that was kind of the original kind of hypothesis of Rufus and why we built it. - That's amazing. And now Rufus has been around for almost two years, right? So as you've seen the process kind of unfold and you've seen the searches and how consumers have been met, how have you changed or edited Rufus along the way? What were some of the biggest improvements or evolutions you've made to Rufus itself in almost two years now? - Sure. I think if I was gonna point to a couple of things that we've done through our journey is that, one is we've brought more and more personalization to Rufus. Like we believe, we fundamentally believe that the more tailored the recommendations Rufus makes, the more tailored the information or content that it provides customers. When we use more about what you've told Rufus or what Rufus knows about you, it only makes the experience better. So as an example, Rufus knows because I've searched for it very frequently and I've explicitly actually told Rufus that like my 10-year-old is a sports enthusiast. So when I'm looking for the next T-shirt to buy for him or the next gifts that I wanna get from the holidays or his birthday, it can easily take that into account as opposed to making recommendations that are maybe not as relevant to me. So that's one big area over the course of the sort of Rufus where we put a lot of kind of effort and energy. The other part is making Rufus kind of available to customers throughout the shopping journey. So whether you're on a product detail page, you've just seen a bunch of search results for a search that you were doing. You're on the order status page. We know that depending on where you are in your shopping journey, Rufus can be more and more helpful to you. So as an example, like one of the things I was looking for is like, you know, again, I use my son a lot as an example 'cause he's probably where I spend the blind share of our dollars on. But, you know, I was trying to figure out, you know, what size socks I bought him last 'cause he started to outgrow him and I needed to buy the next size up. So when I was, what I would have done previously, so I would have gone to my past order history, I would have searched for socks, I would have scrolled through and seen which ones I bought them. It was much easier for me to just ask Rufus, like, what are the last socks I bought from my son? And it told me. And then all I had to do was like, pick the next size up because that's what he needed. So again, being able to make Rufus available throughout the shopping journey would be a secondary area that we put a lot of kind of attention and focus in through our journey. - Yeah, that sounds wonderful. And actually in a personalization, you hit it right on. It truly is changing the way that consumers are presented with details to transact upon because otherwise you're just searching through open air and really an endless aisle where it's really hard to figure out what is it that you're looking for and the time sunk just is very inefficient. So creating that efficiency has come a long way. But how do you prioritize? So there's clearly a ton of things that need to get done. I'm sure you're looking at a laundry list of enhancements that you'd love to make. How do you decide which one to go ahead with first? - Yeah, it's a combination of a little bit of intuition and also what our customers are telling us. Now that we've had Rufus available to customers and hundreds of millions of customers have used a product, we've definitely gotten feedback from them on A, what's working well and what's not. But also things that they expect a shopping assistant to be able to do. So those play very heavily into kind of our decisions on what we decide to work on and not. And then there's also a little bit of intuition in terms of sometimes customers don't know what they exactly want. And we've got to use our own kind of intuition. And there's like recently launched Price History, which is a feature that's available on product detail pages that allows the customer to be able to look at what's the historical pricing on that item. So that they can build kind of confidence in is this a price that I feel comfortable purchasing something at. And we didn't have a customer explicitly telling us, like, wow, would be awesome if I could see Price History. But we do know that having confidence and building trust in our pricing is crucial and very important at the end of the day when customers decide to shop with Amazon. So that was an example where we used intuition. But it's a combination of what customers are telling us in a little bit of intuition at the end of the day. I guess when you think about striking the right balance between intuition and just customer desires, some things and some things that you're looking to innovate on whether it's the technology or whether it's what the customer is expecting could be risky, right? And potentially, though, could be game changing. So how do you at Amazon think about green lighting something that could be very risky? - Yeah, a lot of it to us comes down to do we believe that the customer benefit or the convenience that we can bring to customers balances the potential risk for the perceived risk and we think about that deeply. And a lot of this is through experimentation also. So what you see from us very frequently is we will launch features to see all segments that customers understand what works, what doesn't work. If it works, we'll kind of accelerate. If it doesn't, we'll pause. Kind of think about why it's not working and then re-launch it. So I do think a lot of it is iterative, especially in areas that could be deemed a little bit riskier. But we always start with the fundamental promise of like, what customer benefit or customer problem are we trying to solve? And do we think that that's going to potentially outweigh it? And there's also a shortened kind of long-term aspect to that, right, which is we also recognize that some behaviors take longer to change and some are super easy to change. So those are probably the dimensions. - Got it. And then I guess as you implement these tests or these new features into the platforms, clearly you're looking at metrics. You're looking at signals that tell you whether this feature is actually doing what you intended it to do. Is it disrupting the marketplace? Is it positioning Amazon better? Tell us about some of those signals. - Yeah, some of the signals that we look at to give us confidence in a feature or capability and if it's working is just engagement. So to begin with, are customers using this feature or capability? A second one is repeat engagement, right? So it's not only that customers use something once, but are they using it repeatedly? 'Cause that's a very good sign of like, not only did they use it, but they found it kind of helpful or it solved a need for them. And then the third is, is it driving the right kind of downstream actions from a customer, right? So customers that engage with a specific feature or capability, are they more likely to return to Amazon? Are they more likely to make a purchase? So we tend to look at kind of a basket of metrics that kind of give us confidence that what we're building is delivering customer value or not delivering customer value. - Right, and then you talked about repetition and that's clearly important in retail. You want to get the frequency back, but you also want the customer to transact. Can you tell us anything about Rufus's conversion and how it's helped drive Amazon's top line? - Sure, I mean, one of the things that we look at is our customers more likely to make a purchase if they use Rufus versus those who don't use Rufus. And what we have seen is that customers who engage with Rufus in their mission are 60% more likely to make a purchase than customers who don't. So we're excited by that fact. And that's maybe one metric that gives us a way of kind of measuring the conversion or helpfulness that Rufus is providing to customers. - And then when you think about the customer that Rufus can attract or target, how do you think that customer feels about having AI-driven interaction? Your customer race is clearly very broad. It spans demographics, it spans age cohorts. What do you think they feel when they're interacting with this AI shopping agent effectively? - Sure. So I think one of the things to note about Rufus is that it's a set of different customer experiences. One of them is the ability to type a conversational query and get a response, right? So very kind of classic kind of AI chatbot-like experience where you can have a back and forth conversation with an assistant. Another is like that price history feature that I just mentioned where you're not necessarily having a conversation, but you're kind of tapping on a link that's on a product detail page and you get very valuable information that Rufus has gone out both in terms of providing you with the price history as well as a little bit of content that gives you kind of an overview on the item that you're looking at. And I can speak through other features as well, but you kind of get the sense that not all features may be applicable for all customers because a lot of it depends on where they are in the shopping journey and what they're trying to accomplish. But one of the things that we're trying to solve with Rufus is finding those moments where an AI assisted experience actually makes the shopping for that particular customer on that particular journey to be more convenient, simpler or more delightful than it would have been if they hadn't used a Rufus conversational or Rufus powered feature. You make it sound so simple, but Amazon is a big organization and there's clearly a lot going on under the hood to make this all happen. So I want to get a little more technical now. If you think about what's under the hood of Rufus that makes it possible to deliver the kind of AI driven disruption that you've talked about with Rufus at scale, can you talk about that? >> Sure, there's a handful of things that kind of come together to make Rufus an experience that customers enjoy using. It's a combination of a large language model. It's a combination of kind of the personalization technology that we bring, data sources like information that we know about products. So all of that kind of coming together and knowing when to use what pieces of information to provide a response to a customer or a specific customer experience is a big part of what we believe makes Rufus both kind of unique as well as helpful to customers. >> So how much of the innovation here really comes from the foundational model breakthroughs versus just the integration of the commerce data that you have? >> So we believe that a lot of the value that customers find in Rufus and will continue to find in Rufus is it's a combination of all of these. So I wouldn't necessarily wait them differently or when we believe it's actually knowing how to combine these things really, really well. So I'll give you an example, there are times when personalization becomes, I think, very important. Like as an example, when I'm looking for healthy snacks for my household, we happen to be a vegetarian household. So recommending a bunch of items to me that are not vegetarian would be fairly kind of irrelevant. That wouldn't find that helpful. So having a strong kind of personalization signal there becomes almost super, super important. So it's really about knowing that for a specific kind of sort of request that a customer has, like what is it that's going to be most helpful? So again, I can give you other examples of times when personalization may not play as big of a role because I'm actually asking for more kind of generic items, like as an example, I need to be shopping. I use the same kind of gift example. I need to buy something for my niece. I don't know a lot about her interests, except for the fact that she's seven years old. So I'm almost shopping for a seven-year-old, but like additional personalization probably won't help, but being really, really good at knowing what is trending for a seven-year-old or for, let's say, a five to 10-year-old would be much more important than, let's say, a personalization signal because there isn't a lot of personalization in my profile for my niece. So then let's continue with this and talking about the customer now. Disruption only sticks if it really delights. So when we talk about tech disruption, it's often about what feels new to customers. Which Rufus features do you think seem to capture that wild factor? And how does this change the shop reaction? - Sure. I think it really, I don't know if there's like a feature or even a couple, but I think they're very situational for customers and depending on what they're doing. I'll give you a couple of examples, maybe, to bring that to life. Two weekends ago, we were at a friend's house and my wife saw a pattern on a dining room chair that she thought would look great in our hope. And as much as I don't want to buy like new furniture, I said, you know what, like this is pretty easy. What we could do is use like a new image upload feature on Rufus and I could just take a picture and upload that image to Rufus and have it flying other items that would go well with this. That's so instead of buying new furniture, maybe we'll buy new furnishings for our dining room 'cause that would be cheaper than buying new furniture. But that was super useful because like right in the moment at our friend's place, like it's snappy quick picture. I didn't do it then, we got home the next morning. I actually then asked Rufus like, you know, what are some interesting kind of wall furnishings as well as accents that we could kind of add to our dining room. And it came up with like a bunch of great items that we had available on Amazon that matched the pattern that my wife really liked. Another one that we tend to use, another feature that we tend to use very frequently these days is the ability to like upload a list. Like for some reason, my household still likes to make paper lists for groceries. And as opposed to, you know, like a digital list or like using some sort of app or, you know, building our shopping list on Alexa. So it's now very easy again to use that same feature, but in a different way, which is to take like, you know, a handwritten list, take a picture of it and just upload it to Rufus and be like, add all these items to my cart. And then Rufus knows that, hey, when I write milk, like that means like for my household, it means, you know, horizon, organic, skim milk, because that's what we drink at my house. So again, I think Rufus says, as we've kind of expanded the capabilities and keep and innovating and adding new capabilities, we're trying to meet customers in these moments where there's probably an alternative way of accomplishing the thing that they're trying to accomplish, but it involves either multiple apps, multiple devices, multiple searches, multiple clicks. And without all the personalization and ease of kind of one click or one tapping somewhere and asking a question or uploading an image and getting really, really helpful recommendations. - So then do you think going forward and where we said today, the natural way to search is going to be through natural language, conversations, pictures, is this the new norm? Is this how everyone should be searching today and is searching today? - I think over time, customers will continue to ask more, more natural language questions and they'll do more natural language searches. I don't think it'll, you know, in the short to medium, tomorrow they gets going to replace the way customers search for many of the items that they purchase, especially for items that they repeat purchase, but I do feel like as we continue to do a better job of understanding what customers are asking for, they will continue to ask more and more natural language questions because the responses they're getting are becoming more and more helpful. So I do think customers are starting to habituate more towards doing that because they're getting responses that are more helpful. And it's really upon us to make those responses more and more helpful for them so that they don't have to think as hard or research as hard for what they're looking for, they can just tell us and we can get them something that's helpful. Things like images will continue to play a bigger role for customers because sometimes it's just easier to rather than describing something in words to just send a picture because it's like, I could not describe in my previous example the pattern that my wife saw on the dining room chair at our friend's house. It would just, I would, it would take me way too long and I would get it wrong too frequently to describe it, but a picture is easy because there's no kind of ambiguity or loss in translation. - You're absolutely right. And I think, you know, people are starting to use more natural language and conversations or pictures to get more of what they're looking for. But then as we think about the next step from here, right? It brings us to what I think is the next wave of disruption in retail especially. And that's from moving, taking roofless from being an assistant to an agent. Agentic shopping has been, I guess, the most recent trend word or buzz word in the ecosystem. And I want to ask you about if you feel that that is the next frontier of disruption. When AI doesn't just answer the questions but it really acts on the intent, how are you within roofless approaching that shift? - We do believe that agentic shopping and agents taking more actions on a customer's behalf is not only evolving but something that customers are gonna start to expect more and more of. And we're starting, you know, we've launched a couple of things in the space. One of the things that I'll talk about is price alerts, right? So price alerts is the ability for a customer on any product detail page to say like, okay, I'm looking at a, you know, a hand mixer. And this hand mixer right now is $79 on Amazon. First, I want to know when this maybe goes on sale and if it's on, you know, some sort of deal, notify me when it's down to $60. But the next step we've also taken is the ability to auto buy it when it hits $60. So the ability for the agent to just go out and the minute that that product hits $60 to buy it on your behalf and just ship it to your, you know, your default shipping address that you can set up and your kind of default payment method which you have in your Amazon account. So again, I think those sorts of kind of agent-driven features where you're taking action on behalf of a customer or where the customer's kind of given the agent, the authority and or the desire that they've gotten and got this request out there to go ahead and make a purchase on their behalf as one example of an agent kind of taking action on behalf of a customer and somewhere where I think it'll be just super, super easy and delightful for customers to kind of place that alert and place that buy request out there and we take care of it from there. - Yeah, and it goes kind of hand in hand with what you talked about earlier with the price history, right? So if I can see the price history and I know at some point this mixer was $60, I can just say, well, I want it when it hits this price and it just makes so much more seamless. - Right, 'cause there's in that example, like there are certain purchases that customers, you know, want, you know, our fast delivery on and they want it delivered same day and there's others where you're like, you know, I can wait potentially a few days or even a week or two until it hits this new price point because there isn't kind of a time sensitive thing that I'm trying to hit with that specific purchase. - So clearly Rufus is solving a lot of problems and it's really getting entrenched in both the Amazon ecosystem and within the consumer household. But Amazon being what Amazon is, the story, the disruption story goes well beyond justice one product. So if you could, can you tell us some other ways that the company is applying AI to really redefine the shopping experience maybe across other products too or something else? - Sure, absolutely. You know, I could talk maybe about two other features that are helping customers, you know, essentially find the right product for them and help them kind of in their buying journeys. The one is help me decide. So help me decide what we find many times with customers is that, you know, kind of when they're shopping in a category that they're maybe slightly unfamiliar with or more likely that they just don't shop infrequently, even if they're familiar with the category. And if I stick with the hand mixer category, that would be one definitely for me where I don't really know a lot about hand mixers, but I definitely know that we needed to buy one because like old one broke. So when I was looking for one, what I didn't really know well was, here's a hand mixer and it was like $95 and from a specific brand. What are some of the things that I should kind of consider about that product as well as like, what's kind of an upgrade version? Like if I was going to spend maybe an additional 20 or $40 on one, like what would I get? But if I was going to spend a little bit less, like what would I get to kind of help me decide what's right for me. So that's a new experience that we've recently launched where it helps me kind of zero in on a set of products across like the vast selection that we have in any product category. It allows me to easily kind of trade up or trade down and understand what's right for me. So that's, you know, maybe one example is our help me decide feature. Another one is here in the highlights. So one of the things that we've also heard from customers is that, you know, your product detail pages are a wealth of information but sometimes they're just difficult to, you know, kind of ingest all that information in a written format. So, you know, you're on a product detail page for a television and there's a lot there, right? There's different technologies and screen sizes and compatibility questions that you may have. What here the highlights is is that it's a audio podcast that is completely AI generated. So it's to AI hosts actually describing the product, the pros and cons, why they would, you know, recommend the product for certain use cases, maybe not recommended for other use cases, but it's a fun, engaging way to listen to a, you know, maybe 90 to 120 second clip about all the benefits and things that you should kind of consider about that specific product. But it's, you know, available now on, you know, here the highlights is available on millions of products across Amazon and it's a fun, easy way to kind of understand all the information that we think is kind of relevant and you should kind of consider when making a purchase. So those are maybe two, two good examples of other AI products that we've not only recently launched, but were that are helpful for customers. - There's so much going on at Amazon. You just talked about a few of those products, but I know there's so much more even beyond that. I guess when you kind of take a seat back and think about it, I want to go back to measurement because I know we touched on this earlier, but sometimes as you said before, you don't see the effect of what you've implemented right away, it takes time. So how do you know if what you're implementing is truly reshaping how customers shop and that it's really changed customer behavior, not just the engagement part of it? - Yeah, I think a lot of it is engagement is usually an early signal 'cause you can get engagement information relatively early in a product's lifecycle. But the other thing to look at when it comes to changing customer behavior is that a lot of it is we've got to run the experiment for a long enough period of time. So we're willing to kind of bet with our customers and say, you know, do you, if we start to see the engagement be there, then we start to really look at, you know, does behavior change? And one of the best ways is that you let customers continue to engage with that product over a long enough time. So many times we will, you know, iterate on that product and continue to iterate and iterate until we find the behavior changing and the behavior changing in a way that we think is more helpful for customers. Again, we're not, our objective isn't necessarily the change behavior, our objective is actually for customers to find the things that we build to be helpful. And if they're helpful, we know they'll come back to it more frequently and if they come back to it more frequently, we believe that they will choose to shop on Amazon more and more frequently. - That sounds great. And then I guess before we close out just a few more things that I wanna talk about, you've done a lot of work on, you know, just Rufus and just AI over the course of the past years. What's one or two AI powered features across Amazon that you personally find most exciting or most disruptive for customers, the ones that you think will change the way they shop in the next few years. It could be Rufus, but it could also be something else. - Yeah, I obviously have a little bit of personal bias towards Rufus and we've talked a lot about that. I think a couple of the other ones that I enjoy a lot is our interest feature. So I have what interests is the ability for you to tell Amazon something that you're kind of literally interested in and what we do. And that could be as, you know, as short as, you know, I enjoy sci-fi books or as, you know, kind of narrow, like somebody like myself who enjoys collecting like sports memorabilia and collectibles. And for Amazon, then to be going out and continuously looking for new products that are coming out that are based on your interest. And what that does is it essentially creates a feed of products for me that are hyper personalized and tailored for something that I've set up as opposed to having to kind of proactively go in and do that search on a daily or weekly basis to see what's out there. It's like Amazon's always searching on my behalf even when I'm sleeping. And then for me to just open up my feed of products and say, oh, is there something new and interesting here? So for me, it's like an awesome shortcut to all the new stuff that's available in our store. And I tend to use pretty frequently. So that's one that I think is super exciting and something that I use very frequently. Another one, another feature that I use also very often and that I think is very, very helpful for customers is like kind of, it's called keep shopping for. But it's like, you know, when I essentially start looking for something on Amazon, but then I either get interrupted or I wanna come back to it later. The ability to kind of continue that mission at a later point in time is awesome. And Amazon remembering everything that I did. So I'm not starting all over again. So to have an easy place to come to, that's got the products that I've looked at previously and to be able to know, you know, why I like the specific product or not and to be able to kind of tap in and out of them easily. And then with our new compare feature to be able to compare those products across each other. It's another feature that I use very frequently 'cause I can't tell you how many times I'm shopping for something and then I get interrupted and then life goes on and I need to come back to that at a later point in time and having an easy place to kind of pick up where I left off. Also is something that I'm super excited about something that we've had for a little bit of time and that we continue to kind of innovate on and improve. - You know, it continues to amaze me how many pain points Amazon continues to solve even after all these years and yet there's still so many more. So we've talked a lot about what Amazon is doing today with an AI shopping and how it's just making shopping more frictionless, more efficient and really more pleasing for the customer. But when you think about the future and the future of shopping particularly, what do you see the greatest opportunities for AI-driven disruption in shopping in the next few years? If I talk to you again in five years Rajeev in 2030, what will you have already said that we'll have come true? (laughs) - Wow, yeah, so I think as we look into the future, I think personalization will continue to play a very critical role for customers. The more that customers tell us about what they're looking for, what is important for them when they're shopping, whether it's by category or time of year, or kind of the attributes that are most important to them, I think you're gonna play a very critical part because I think customers are gonna expect from us that we just know that about them, so that the recommendations that we make take that into consideration. So I do think that we were talking a few years from now, I think what you would find is that customers are gonna expect that everything that I see in the store has been tailored just for me because we know enough about them and the customers have told us enough about themselves where they just expect that to work. Just like you would expect a great human to know about you, you would expect a great shopping assistant inside a Amazon to know that about yourself. It's, if I was gonna use an anecdote, again, like a personal anecdote, like my wife and I have been married for a long time, right, so over 20 years, we know each other's kind of preferences, but that's come out of like many, many years of being together. It's not that hard to believe that a shopping assistant should be as good as my wife and picking out like the perfect item for me because we should also get to know our customers that well and know when to use that information to make a better recommendation. - So is that where Rufus has had it over the next two years? Should we expect that out of Rufus? - We should expect Rufus to get to know our customers better and better, and to be used personalization to make better and better recommendations every day. - And then if I could ask you one last question, if you had to describe the future of shopping in one word, what would it be? - Exciting. - Exciting, okay? Well, Reggie, we wanna thank you for joining Tech Disruptors today. For our listeners, we wanna thank you for tuning in. If you liked the episode, please subscribe and leave a review. And check back to hear conversations with the leading disruptors in the Tech Landscape. If you wanna learn more about our research, including deep dives on topics like AI, check our work on the Bloomberg Terminal at BI Go. We also wanna thank Aditya Somani for his help in editing the podcast. Thank you.