← Behind the Numbers: an EMARKETER Podcast
Behind the Numbers: an EMARKETER Podcast
AI Can Recommend. But Can It Sell? | Behind the Numbers
00:22:17
Loading summary
Transcript56 lines
- [00:00]
A
When a consumer reaches checkout, they're no longer browsing, they're buying. It's a moment of peak intent, attention and engagement. That's where ROKT comes in. ROKT helps brands reach customers at the moment that matters most, delivering relevant offers and content that feel like a natural part of the transaction experience, not an interruption. Learn more@rokt.com. Hey Gang. It's Friday, June 12th. Rachel Grayson, listeners, welcome. Welcome to behind the Numbers, any marketer podcast made possible by rokt. I'm Marcus and join me for today's conversation. We have two analysts. One of them lives on the east of America. That's Rachel Wolf.
- [00:43]
B
That is me. I do live on the east. Thanks for having me.
- [00:46]
A
No one's ever said it like that before. I wish I hadn't. Living in the West.
- [00:51]
C
Grace Harmon hey, Marcus, thanks for having me.
- [00:53]
A
Of course. Today's fact the Voyovkina Cave in Abkhazia in Georgia is the world's deepest known cave. How many Statue of Liberties deep do you think it is? Or Eiffel Towers? I'll take any structure. Very is the answer. It's an impossible guess. 7200ft for our friends across the pond, that's about 2200 meters and narrowly edges out second place, which is in the same area, which is just by 42ft. So it's very, very close, first and second. So for context, the tallest building in the world is the Burj Khalifa in Dubai. That's 2,700ft. El Capitan in Yosemite national park, that thing that, that the free Solo movie is about is 3,000ft. So this cave is around two and a half times as deep as each of those. It would be 7 Eiffel Towers or 23 Statue of Liberties.
- [01:57]
B
That's very impressive. Yeah.
- [02:00]
A
Why would you go in there? Three days it takes you to get in to the bottom and three days to return to the surface. It's fascinating because we talk a lot. Melissa Garside, a visual capitalist, was explaining that we scale mountains and that's something that you see movies about and there's stories about them. But going into the deepest caves in the world is kind of one of planet Earth's kind of last frontiers. It's fascinating. Seven of the 10 deepest caves reach more than a mile underground. Four of the deepest are clustered in the same karst mountain region in Abkhazia, Georgia, where the soluble karst rock creates ideal conditions for deep cave systems.
- [02:40]
C
I'm good at sea level.
- [02:44]
A
Same.
- [02:44]
B
Yeah. Every news story I see about a cave is people getting trapped. So I think I will stay out of the very deep ones.
- [02:52]
A
Yes, well, you don't want to go in Mexico's Cheva cave, that exploration started in February and they are attempting to be. They think that might be the deepest one of all. They haven't found that out yet. How do you. I don't know.
- [03:06]
B
Yeah, I was going to ask, how do you measure. Bring a very long ruler with you.
- [03:12]
A
Hold the other end. Anyway, today's real topic, what do you need to know about the current state of AI shopping assistance? All right, so Rachel, you've recently written a piece about this and we wanted to talk about it. But we'll start with some of the news about AI shopping assistants. A piece from Forbes contributor Jason Goldberg back in March titled why OpenAI's checkout retreat spells trouble for its commerce strategy. So OpenAI launched instant checkout through ChatGPT at the end of September of last year, but it's closed down early March. Rachel, why didn't ChatGPT's instant checkout take off?
- [04:00]
B
I think there are a few reasons. One is that fundamentally OpenAI underestimated the amount of work that it would take to build a functioning E commerce marketplace. I mean, they lacked a lot of kind of basic functionality, such as how to calculate and collect sales tax, which if you think about shopping online, is such an integral part of, of the experience. And same thing with, you know, if you wanted to apply a coupon code, if you wanted to connect your loyalty program. All of these things weren't really built into the interface as it was released. And for that reason people just weren't interested in using it. So Walmart said that in chat checkouts within ChatGPT converted three times worse than its own website. And so when you see that if you're Walmart, you kind of think, well, why am I spending all this money and time and resources to develop this, this thing that doesn't work as well as my own website when I could be putting it into more useful AI initiatives or, you know, the rest of my business?
- [04:59]
A
Did they rush it? Is that fair to say? Because this Forbes piece was saying at launch, instant Checkout supported only single item purchases from U.S. etsy sellers. Multi item carts? No, not yet. Promotional codes? No. Shipping promises? No. Is it fair to say that this isn't like a referendum necessarily on instant checkout or AI shopping and buying, but more a model that was rushed out the door when the company had other priorities?
- [05:25]
B
Yeah, I think that's fair to say. And I think you could even draw that out to talk about other AI platforms and their shopping features as well. And, you know, probably talk about this more later. But a lot of these companies have announced these big, splashy features, but for the most part they're not really available to shoppers. And so there is this disconnect between what companies are saying and what they're actually doing. And, and I think in the case of OpenAI and ChatGPT, it does take a lot of resources to build a functioning E commerce marketplace, which is essentially what Instant Checkout would have been. And they just didn't feel like they could do it at this moment in time.
- [06:01]
C
Yeah, I mean, on a broader level, you know, talking about Google, it's announcing features faster than it's launching them. So it's kind of confusing for consumers or businesses to know it's actually available and active at any given time. And then I think that it's really hard for retailers to set up shop. A lot of the holdup for merchants selling through these AI tools is on the firm side, not on the retailer side.
- [06:22]
A
So is it fair to say then, and Jason, Mr. Goldberg of Forbes was, was saying. OpenAI's retreat from instant checkout tells us almost nothing about whether consumers want AI agents to help them buy things. It tells us a great deal about what happens when the most distracted tech company tries to build a commerce platform on. On the size. You can't draw meaningful conclusions about consumer demand for agentic commerce from an experience that was never complete enough to test. The product wasn't beta. It was pre alpha. Is that fair then? After doing the research, Rachel, have you found that we're just early days as opposed to consumers don't have any interest in this at all?
- [07:00]
B
I think that there is definitely interest on the consumer side in figuring out what these tools can do for you right from the shopping experience. I don't know that it extends all the way out to letting it complete a purchase on your behalf, but I think, you know, the fact that Instant Checkout converted so much worse than Walmart's own website tells you that, yes, it was a OpenAI kind of overshot a little bit. But for the time being, at this moment in time, shoppers just don't want this kind of functionality necessarily from these platforms. They're more comfortable with transacting on retailers themselves. They're not willing to hand over their credit card information to these platforms. And so I think there is a little bit of, you know, yes, the technology isn't there yet, but the consumer isn't there. Yet either.
- [07:41]
A
That's interesting to see what route they'll take. This note saying from the piece, OpenAI seems to be betting, at least for now, that its role in commerce is closer to Google's influence the shopping journey monetize through ads and affiliate referrals and let the merchants handle the messy part. Grace, do you think that's a fair assessment of how OpenAI, after taking a step back, sees their role in shopping?
- [08:04]
C
I think there's a lot of challenges of putting together AI with all the intricacies of e commerce business it has lacked to do really diverse experimentation and testing with its ad business. And I do think that extends into what it's been willing to try in e commerce.
- [08:19]
A
Yeah, they're a company that needs money and focusing on advertising and trying to monetize all those weekly users, those 900 million odd, is much more of a priority. And so maybe focusing on that first and then coming back to the commerce piece at some point or putting more effort into it, I think that it's
- [08:37]
C
also figuring out a much broader issue which is whether it needs to be focusing on those consumers or an enterprises right now. So, you know, does it push into having a side quest of figuring out a really strong e commerce business or does it go into focusing on having really strong coding tools? So I think it's not just, you know, one business at a time. It's trying to figure out like five.
- [08:58]
A
Yes, yes indeed. Rachel, we're talking a lot about OpenAI and the instant checkout. Your research was on what you call third party AI shopping assistants. And ChatGPT falls into that, into that bucket. What do you mean by what are third party AI shopping assistants? What are some other ones? What are some that don't fall into that bucket?
- [09:17]
B
Yeah, so my report was really focused on these shopping tools that are offered by AI platforms. So ChatGPT, Google which has several, include Gemini AI mode, so on and so forth, and also Microsoft Copilot. Those were the three that I focused on for my report. But you could broaden that out a little bit and talk about Perplexity, for example, which also has shopping tools meta AI. As you know, they're sort of dipping into the space as well. So essentially third party AI shopping assistants are these platforms that don't offer products directly to customers, but facilitate the purchase of them versus say Amazon. I guess now it's Alexa or Alexa or Walmart Sparky that are AI tools offered specifically by retailers for shoppers on their sites.
- [10:08]
A
So let's talk through some of those that you mentioned ChatGPT, Google and Microsoft Copilot that you looked at. What's one strength of each? Let's start with the strengths. What's one strength from each platform?
- [10:21]
B
So in my extensive testing of these platforms I found that ChatGPT was pretty good at the research phase, right? If you give it. I was giving all three of them more or less the same prompt which was I'm looking for a bag that's work appropriate, that can fit a laptop in this particular price range and having them sort of generate a list of recommendations and then kind of refining from there. And I found that like ChatGPT in terms of the results, in terms of the interface that it would give you, was really good at walking you through the pros and cons about laying everything out in a way that was pretty intuitive having product images and the results. So all of those things that really helped with the initial research phase of the search for Google. I found that where Google really excelled is they had the most up to date product and pricing data and they also had the most set of features related to, I'm going to call it like finding the best price, right? So tracking price over time, figuring out whether you can, there are any coupons that you can apply, you know, with your loyalty benefits or membership benefits will apply to a particular purchase. And I think, you know, that was really where Google excelled and for Copilot it was kind of in the organization of all of the shopping searches. They were the only one of the three to have a separate shopping tab that would collect all of these sort of commerce related searches which made it really easy to go back to see, you know, what it had recommended in the past, various searches that I had made for different types of products and it was really helpful organizationally.
- [11:50]
A
Grace Rachel writes in a piece though the platform's shopping features look alike, the quality of the user experience is where they diverge. Is there one of these platforms that stands out to you more so than the others when it comes to shopping?
- [12:05]
C
I think that Google stands out pretty strongly. For one thing, there's just that huge amount of trust and it's the default search platform to begin with. OpenAI, you know it's. Rachel had mentioned this before, it's pretty clunky to be buying multiple items and information isn't always accurate, even if it's very good at presenting it for Google. I had mentioned before that it's kind of launching or announcing these features faster than it gets them out into people's hands. But you know, it offers virtual, virtual, try ons, things like that. So I think that they've done a lot more exploration into what they can be offering other than just a listicle.
- [12:39]
A
Rachel, where do these platforms struggle? What's the main drawback from each?
- [12:43]
B
I mean, one thing that I found that's pretty much consistent for all of them is that they're very good at giving you results if you give them very specific parameters, right? If you say I have a certain price range or I'm looking for X, Y and Z features. But I think where they all fall short is in that area of product inspiration, right? Something that social commerce is very good at is inspiring you to make a purchase or, you know, you find things based on your personal style that you then click on and buy. Whereas I feel like all three of these platforms are better at. I'm going to call them like need based, where you have very specific check marks or features that you want to hit and it can tell you whether it does for each one. So I think that's kind of a weakness in a way for all of these assistants. That limits the types of use cases that you might have for these AI shopping assistants overall.
- [13:34]
C
So you can't really do like an unbranded search like you would with Pinterest or something like that, right?
- [13:38]
B
You can't really be inspired because you have to tell it what you're looking for, which is kind of a drawback when you don't know what you're looking for. And a lot of things, like, let's say for apparel, I think a lot of those purchases are based on kind of intangibles, right? Like you look at an image and you decide whether to buy something based on, I don't know, maybe how it drapes on the model, or you like this particular cut over that cut and you can't really describe that necessarily to an AI assistant.
- [14:04]
A
You write that AI platforms are widely used product research tools, but they still struggle with conversion. What's the number one reason why?
- [14:13]
B
I think the issue, as I may have said earlier, is just the lack of trust, right? People aren't ready to give AI platforms the authority to shop on their behalf because these tools are untested. Nobody really knows if they're going to go and make unauthorized purchases on people's behalfs when you just hand over the credit card data. But I think the other issue is that it's not only the payments piece. I think people just don't trust AI platform's recommendations wholesale. As you can see from this chart on the screen, over 60% of AI users search for more information outside the AI after they've been recommended a specific product by an AI assistant or chatbot. And this data comes from a January report that eMarketer did with publishing. So, you know, what this tells me, tells everybody, is that people need to do their research outside of what these chatbots are saying. And maybe that is on Google. Maybe you get something on AI mode and then you promptly switch to a Google search to find out more, or maybe you go to a retailer's website itself. But I think that's just a huge barrier.
- [15:19]
A
I want to talk a bit about what this looks like down the road. We have some forecasts.
- [15:22]
C
I would have something to add on to Rachel.
- [15:24]
A
Please, please, please. Absolutely, sure.
- [15:26]
C
So, first off, I agree with everything that Rachel said. You know, people want to be in control of checkout. They want to get their own. They want to find personalized ideas, they want to find new products maybe without having to give the details of, I want this color, this price, this brand. And they, like you said, they want to handle the spending on their own. That trust element around autonomous purchases isn't there. And then you had mentioned that there are still doubts around the recommendations themselves. You know, are they accurate, are they generic? And that's pretty split, as you can see from this chart on the screen. About 31% of US adults trust AI somewhat or a lot to make recommendations. 41% don't trust it, and 23% are neutral, according to YouGov. So that really just doesn't speak to overwhelming confidence.
- [16:09]
A
You're talking about the trust piece. So it's, are they accurate? But also, is part of this. Is it paid for? Right? Like, is there, is that like a sponsored item that you're getting presented? Like, how do you, how do you solve for that? Is that just the same with Google, you know, search results that you'll say sponsored on the ones that are sponsored and not on the ones that aren't? Like, is that the only way, by being transparent and labeling which ones are and aren't, is that the only way to get around it? Or do you think people eventually will say, I don't really care as long as it gives me the best thing?
- [16:39]
B
Yeah, I mean, I think that's an interesting question that all these platforms have to kind of figure out. You know, I think ChatGPT and Google seem to be of the mind that if you clearly label what's an ad and what's not, then that kind of gets around the trust issue.
- [16:54]
C
That's good enough, right?
- [16:56]
B
Yeah. So on the subject of advertising, 2/3 of consumers say they would trust chatbot recommendations less if they saw ads, while 57% would trust the advertising brand less. And this is according to a survey by partner Centric.
- [17:10]
A
So I want to talk a bit about where this goes next. We'll start with where we are today. Our forecasting team estimates that 88,0 million Americans, it's a quarter of people in the country, are Gen AI shopping users, meaning people who enter a prompt into a gen AI system to carry out any shopping related tasks. Shopping does not mean buying. Most people are shopping looking. And this excludes AI powered search summaries. So think of AI overviews, things like that. We think that number is going to climb from about 23% today to about 30% in three years. So across the 100 million Americans, Mark, that's not everybody by any stretch, but it is a fair amount of people. It's about half of gen users. So if you're a geni user, then there's a much greater chance obviously that you're going to be a geni shopper. Chris, I'll start with you for this one. How will we be using AI shopping assistance a year from now? What's AI shopping gonna look like?
- [18:06]
C
I think the biggest part is just how user behavior changes and how strong integrations are. If these traditional shopping habits stay really sticky, if these experiences stay really fragmented, I think AI is gonna stay a lot more of a research and discovery tool rather than turning into this personal shopper. And I think that the next year we're just gonna see more of a test of whether consumers are willing to hand off more of the shopping journey to AI and also just how strong those services are. I also would go back to the stat you mentioned that you also should keep in mind that there's, you know, intentional and then unintentional AI use. There's a difference between me going to chatgpt.com and asking it to pick something out for me and me putting in a Google search and then interacting with an AI overview that is presented to me.
- [18:52]
A
Yeah, people might not even realize they are AI users. They're just going to the same old place, but it happens to be using AI. Rachel, you wrote that shoppers may be more comfortable interacting with retailers own AI assistants because of trust, personalization, things like that. What does the landscape look like in a year? Is there a clear front runner emerging? How are people interacting with AI shopping?
- [19:13]
B
Yeah, I mean, my prediction, as you just said, is that shoppers will just be more accustomed to interacting with what's already on retailers websites because those are the tools that can really deliver the best experience, right? In terms of personalization, in terms of recommendations and in terms of delivering actionable insights. You're already on the retailer's website, you already have a certain amount of purchase intent and so you can go from there. And I think to Grace's point, it ultimately comes down to whether these AI shopping assistants, third party shopping assistants, actually improve the experience for the consumer, right? If there is genuinely a reason to use these AI shopping assistants versus again just going to Google or just going straight to Amazon and using Rufus, I think that will be the key. And certainly all of these platforms are trying to a certain extent. But whether they'll be successful depends on whether they can deliver on their promises.
- [20:09]
A
Let's end with what to do right now. Brands and retailers, what's your number one recommendation for them?
- [20:15]
B
So I think brands, retailers, you have to test and learn, right? You have to think about how you're showing up on ChatGPT, how you're showing up on Google, how you're showing up on Copilot, but you also have to think about all these other platforms that people are now using for commerce related searches or you know, just searches in general. And so maybe that means looking at Claude and looking at Meta AI and looking at Perplexity and all these other platforms. And the technology moves fast so you have to be prepared for any potential shifts, whether it's somebody launching in chat checkout and then promptly ditching it a couple months later. You have to be flexible to manage those changes.
- [20:56]
A
Grace, anything to add?
- [20:57]
C
I think it's just a pretty daunting task right now to meet shoppers where they are. I mean you have to be ready for all these zero click searches where people are getting information and maybe experiencing some brand awareness in chat, but not clicking through, making sure that product data and reviews are constantly available. For AI system systems to find an index, you know, keeping up with regular SEO and keywords for traditional search, I think it's a really daunting task right now. I think I'd have a hard time giving a single recommendation.
- [21:24]
A
Well, if you do need more recommendations, Rachel's got you covered. The full report she just put out is the state of AI shopping assistance Pro plus subscribers head to eMarketer.com you can find it there or there'll be a link in the show notes. That's what we've got time for for today's episode though. Thank you so much to my guests for hanging out with me. Thank you. First to Rachel.
- [21:41]
B
Thank you so much.
- [21:42]
A
And of course to Grace.
- [21:43]
C
Thank you, Marcus.
- [21:44]
A
Yes, indeed. And to the whole production crew. Danny for helping out on this one. Luigi hanging in the background taking credit for doing nothing. And everyone for listening in to buying the Numbers Marketer podcast made possible by rockd. Thank you guys for listening in. Subscribe and follow to hear about new episodes and leave a rating in review if you can. It's the equivalent of donating to a podcast. So if you guys have a few minutes that they mean the world to us, they really, really help us out. We'll be back on Monday. Until then, happiest of weekends.