Latest / What's Your Problem? / Solving the Trust Problem in AI Shopping

Transcript

Solving the Trust Problem in AI Shopping · What's Your Problem?

0:15
Speaker 1: Pushkin.
0:20
Speaker 2: Hello, hello, Malcolm Gladwell here. Over the past several months
0:23
Speaker 2: across revisionist history and business history, we've been exploring the
0:27
Speaker 2: idea of trust, where it comes from, the moment it's earned,
0:32
Speaker 2: and where we decide to place it. This episode is
0:35
Speaker 2: sponsored by PayPal, trusted by consumers, built for all business.
0:40
Speaker 2: For over 25 years, they've been at the forefront of commerce,
0:44
Speaker 2: helping merchants of all sizes grow. So wherever commerce moves next,
0:49
Speaker 2: you can rest assured PayPal has your back. And now
0:52
Speaker 2: we want to leave you with one final discussion on
0:55
Speaker 2: this special episode of What's Your Problem? Today, Jacob Goldstein
1:00
Speaker 2: hosts a conversation about how consumers and companies are trying
1:04
Speaker 2: to figure out trust in the era of agentic AI.
1:08
Speaker 1: Let's listen.
1:20
Speaker 3: I'm Jacob Goldstein, and this is What's Your Problem? My
1:22
Speaker 3: guest today is Kartik Hosanagar. Kartik is a professor of
1:26
Speaker 3: operations and marketing at the Wharton School at the University
1:30
Speaker 3: of Pennsylvania. And I wanted to talk with him because
1:34
Speaker 3: we are at this moment when the way people use
1:37
Speaker 3: AI is shifting. We're starting to go beyond just having
1:40
Speaker 3: AI answer questions and toward having AI actually do stuff
1:45
Speaker 3: in the world for us, including having AI agents go
1:49
Speaker 3: out onto the internet and buy stuff. Karthik has been
1:53
Speaker 3: studying this shift. He's been trying to figure out how
1:55
Speaker 3: both customers and merchants can solve the problems that arise
1:59
Speaker 3: in the era of shopping with agentic AI. What will
2:03
Speaker 3: it take for customers to trust AI agents to make
2:05
Speaker 3: purchases on their behalf? And what do merchants need to
2:09
Speaker 3: do to prepare for a world where some of their
2:11
Speaker 3: customers are AI agents? In the first part of the conversation,
2:15
Speaker 3: Karthik and I talked about what this all means from
2:18
Speaker 3: the consumer's point of view. In the second part, we
2:21
Speaker 3: talked about what it means for merchants.
2:23
Speaker 1: You know, I will often begin my shopping research, having
2:28
Speaker 1: LLMs do the research for me. I'll send them on
2:31
Speaker 1: a deep research task and have them do all of that.
2:35
Speaker 1: I'll have them synthesize it for me. In some cases,
2:39
Speaker 1: I might be patient enough to read that long report.
2:41
Speaker 1: Sometimes I'll say, convert this into a table and just
2:45
Speaker 1: give me like, you know, what I need to know.
2:48
Speaker 3: This is pretty hardcore shopping. Like what kind of item
2:51
Speaker 3: are you having deep research reports written for? Tell me
2:54
Speaker 3: it's a t-shirt.
2:58
Speaker 1: Yeah, well, you know, sometimes it's like, you know, I'm
3:01
Speaker 1: looking to buy a car right now and trying to
3:04
Speaker 1: figure out what new electric vehicles are around. But honestly,
3:10
Speaker 1: I don't even know the research about whether electric vehicles
3:15
Speaker 1: are as environmentally friendly as originally touted to be because
3:19
Speaker 1: there's all this criticism that, hey, there's all this rare
3:22
Speaker 1: earth mining and the environmental damage of that is very high.
3:25
Speaker 1: So I'm not even saying which car should I buy.
3:28
Speaker 1: I start with, tell me The pros and cons of
3:31
Speaker 1: electric vehicles. In what sense is it good for the
3:34
Speaker 1: environment and what sense isn't? And in what sense is
3:37
Speaker 1: it convenient to own one and in what sense it
3:40
Speaker 1: is not? And then from there, it would be, hey,
3:43
Speaker 1: I've seen on the streets people driving Rivian. What do
3:45
Speaker 1: you think about Rivian? Is that any good? At times,
3:50
Speaker 1: it may not be deep research, but nonetheless, you know,
3:53
Speaker 1: I have a startup as well that I started a
3:55
Speaker 1: year ago. We're trying to make decisions for the startup like, hey, which.
4:01
Speaker 1: Software should I use for cap table management? Which software
4:05
Speaker 1: should I use for payroll? Which vendors? Who's right for
4:09
Speaker 1: a company our size? We are very small. We're just
4:12
Speaker 1: about 10, 12 people. Who's right for California? Who's right
4:16
Speaker 1: if we grow to 100 people and so on?
4:19
Speaker 3: I do feel like this distinction between using AI, using
4:23
Speaker 3: a language model to do research for shopping and and
4:26
Speaker 3: actually handing off the purchase decision to the agent is
4:31
Speaker 3: really important right now. Like I basically trust the AI
4:35
Speaker 3: to give me information about, you know, product reviews and
4:38
Speaker 3: that sort of thing. And I'm curious, like when you
4:42
Speaker 3: see people who are using AI on the research side, not,
4:45
Speaker 3: you know, we'll get back to agentic in a second,
4:47
Speaker 3: but on the research side, like how are they using
4:49
Speaker 3: it differently than traditional search?
4:53
Speaker 1: Well, I think the biggest difference with traditional search is
4:56
Speaker 1: that the search engines are merely curating the information and
5:02
Speaker 1: then you are taking the trouble to synthesize it. And
5:07
Speaker 1: you get exposed to a lot of information about various brands,
5:12
Speaker 1: even if you decide not to eventually buy from them,
5:15
Speaker 1: or even if it's not in your shortlist, you still
5:18
Speaker 1: get exposed to a lot. Yeah. In contrast, when it
5:23
Speaker 1: comes to AI for shopping, I'm outsourcing a lot of
5:28
Speaker 1: the synthesizing. And then depending on the type of buyer,
5:32
Speaker 1: the effort level varies where some might say, synthesize it
5:37
Speaker 1: for me. And then there might be a lot of
5:40
Speaker 1: back and forth, but what about this? What about that?
5:43
Speaker 1: And in other instances, you might outsource the decision itself
5:47
Speaker 1: in terms of, bring it down to the short list.
5:51
Speaker 1: And so you're starting off with a short list of
5:54
Speaker 1: three brands and a comparison table, and that's where the
5:56
Speaker 1: human work begins. A lot of it is based for
6:00
Speaker 1: me in terms of the risk of making a wrong decision.
6:03
Speaker 1: And if the risk is high and I'm spending a
6:05
Speaker 1: lot of money, then I'll actually go back and forth
6:08
Speaker 1: with the LLMs. But if the risk is low, then
6:10
Speaker 1: I'm pretty much, for the most part, outsourcing everything except
6:15
Speaker 1: the final approval decision to the LLM.
6:18
Speaker 3: And that seems pretty common, right? It seems like, by
6:21
Speaker 3: and large, a lot of people are comfortable using AI
6:23
Speaker 3: for doing the research, but they're a lot more hesitant
6:26
Speaker 3: to let the agent take that last step. They're wary
6:30
Speaker 3: of letting AI actually buy the thing. And, you know,
6:35
Speaker 3: it reminds me a little bit of the early days
6:38
Speaker 3: of the web, basically the 90s, which I am old
6:41
Speaker 3: enough to remember. And, you know, you would go and
6:44
Speaker 3: companies had websites and you'd look at stuff. But the
6:46
Speaker 3: idea of putting your credit card into the web was like, no,
6:50
Speaker 3: there is no way I'm going to put my credit
6:52
Speaker 3: card into some random website. And of course, you know,
6:56
Speaker 3: PayPal is in fact an example of one of the
6:58
Speaker 3: early pioneers here. They were navigating these shifts in real time.
7:03
Speaker 3: But now, whatever, 25, 30 years later, we have solved that.
7:07
Speaker 3: Everybody buys stuff online all the time. And, you know,
7:11
Speaker 3: I would say at this moment, I am not ready
7:13
Speaker 3: to just have agentic AI go off and buy stuff
7:16
Speaker 3: for me without my input. I could imagine getting there.
7:21
Speaker 3: I just don't trust it yet. And so I'm curious,
7:24
Speaker 3: you know, we've solved the trust problem before. How do
7:27
Speaker 3: we solve it this time? Like what has to happen
7:30
Speaker 3: to get to the point where consumers trust agents to
7:33
Speaker 3: buy stuff on their behalf, on our behalf?
7:36
Speaker 1: Yeah, I think there's a few things that, need to happen,
7:40
Speaker 1: some of which is probably the tech infrastructure needed to
7:45
Speaker 1: support something like this. The second is a behavior change.
7:50
Speaker 1: People doubted in 1998, 1999, whether it would ever happen
7:56
Speaker 1: that we would put our credit cards and buy. And
7:59
Speaker 1: there were others who said, no, no, no, it'll happen
8:01
Speaker 1: next year. And there were folks who said, no, it'll
8:03
Speaker 1: happen in five years. And so that's the speculative piece,
8:07
Speaker 1: when it happens and for what it'll happen and so on.
8:11
Speaker 1: I remember back in the day, I had my own
8:14
Speaker 1: students who, you know, I teach at Wharton. So my
8:17
Speaker 1: students at Wharton who started the company Warby Parker, where
8:20
Speaker 1: you buy eyeglasses from, when they first came and actually
8:24
Speaker 1: their term paper in my course was a business plan
8:28
Speaker 1: for Warby Parker. And at the end of the semester,
8:31
Speaker 1: I chatted with them. And to me, it felt like
8:34
Speaker 1: at that point when I chatted with them, it was like,
8:36
Speaker 1: I can see some people buying their eyeglasses online. But
8:41
Speaker 1: I think when you're buying eyeglasses, you're wearing one, I'm
8:44
Speaker 1: wearing one. I feel like I want to try it on.
8:46
Speaker 1: I want to see how it feels. And that is
8:49
Speaker 1: extremely important. So I kind of my reaction was, it
8:52
Speaker 1: makes sense, but I still don't see how it's mainstream.
8:56
Speaker 1: But they changed it.
8:57
Speaker 3: Well, they sent you the glasses. The thing of sending
9:00
Speaker 3: you the tester pair, was that because of your feedback?
9:03
Speaker 3: Did you get equity?
9:05
Speaker 1: No, I didn't. I'm sure others brought it up to them.
9:07
Speaker 1: I won't take credit for that, but I will give
9:09
Speaker 1: them credit for kind of saying, we hear you on
9:13
Speaker 1: the problem, but we disagree that there is no solution
9:18
Speaker 1: to the problem. And so that's the part that I
9:22
Speaker 1: was saying is speculative in terms of, when it will
9:26
Speaker 1: happen and what needs to happen for that full behavior
9:29
Speaker 1: change to happen. I think someone will crack it. And
9:34
Speaker 1: I think it'll happen in phases just like e-commerce did.
9:37
Speaker 1: We first felt comfortable buying books and we said, hey,
9:40
Speaker 1: a book is a book. I don't need to worry
9:42
Speaker 1: about how it feels in my hand and so on.
9:44
Speaker 1: And then it became electronics. Hey, I bought books for $ 20.
9:48
Speaker 1: Maybe I can buy a $ 200 or $ 500 electronics online
9:52
Speaker 1: and it's going to be okay. And then it became clothing.
9:55
Speaker 1: And then it just kept expanding. And so I think
9:57
Speaker 1: it's going to be similar with agentic where there are
10:01
Speaker 1: some categories which will be easier than others. And there'll
10:04
Speaker 1: be some customer segments where it'll be easier than the others.
10:08
Speaker 1: And that's where things will start. And then how that
10:11
Speaker 1: percolates or spreads out and becomes more mainstream across all categories,
10:16
Speaker 1: that's the subjective piece. And that's what I was kind
10:19
Speaker 1: of saying. It's hard to say when that will happen.
10:22
Speaker 3: You mentioned books. I mean, famously, Amazon started out selling books.
10:25
Speaker 1: Yeah.
10:27
Speaker 3: Do you have a bet for what's going to be
10:29
Speaker 3: the books of agentic? What's going to be the first
10:31
Speaker 3: category where people are comfortable having an agent go make
10:36
Speaker 3: the purchase decision for them?
10:38
Speaker 1: Yeah, I actually feel like groceries are a really good
10:42
Speaker 1: candidate for me. You know, and this is something that
10:45
Speaker 1: I actually ask my students on a regular basis. I've
10:48
Speaker 1: been doing this for about two, three years where anytime
10:51
Speaker 1: I'm teaching my MBA, undergrad, or executive students, every session
10:57
Speaker 1: I have on agentic shopping, AI shopping, I ask them,
11:01
Speaker 1: would you use an AI agent to shop for groceries?
11:07
Speaker 1: Just your usual grocery list. And perhaps you could argue
11:11
Speaker 1: that if I'm using Instacart or if I'm using even
11:15
Speaker 1: the Whole Foods or the Amazon app to order my
11:18
Speaker 1: groceries where the list is there and I'm saying go
11:20
Speaker 1: buy it, at some level, I'm kind of saying, hey,
11:25
Speaker 1: here's my usual list. You buy it. You make the
11:28
Speaker 1: decisions on substitutions and do it on my behalf. So
11:33
Speaker 1: I've already outsourced that. That's not a big deal. Tomorrow,
11:38
Speaker 1: if someone were to say, here's an AI agent that
11:40
Speaker 1: can do end-to-end, just upload your grocery list once, and
11:43
Speaker 1: then you don't have to ever ask again, that's one
11:45
Speaker 1: I'd be willing to do. The Instacart, the human making
11:49
Speaker 1: that decision, for me, is not that different than an
11:54
Speaker 1: AI agent saying, hey, I'll substitute this particular brand of
11:59
Speaker 1: milk with that particular brand of milk and so on.
12:02
Speaker 1: I know what this customer wants. And I'll know in
12:05
Speaker 1: two orders that this is working for me or not.
12:10
Speaker 1: Am I willing to try two, three times? Yeah.
12:12
Speaker 3: And it's relatively low stakes, right? It's not like buying
12:15
Speaker 3: a car. If you do it and it kind of sucks,
12:17
Speaker 3: you're like, well, I didn't get the right milk this week.
12:20
Speaker 3: Not that big of a deal.
12:22
Speaker 1: Exactly. So I'll happily try that a few times. So
12:25
Speaker 1: I feel that's a good one. I also think the
12:28
Speaker 1: other one is where individuals are doing a lot of
12:32
Speaker 1: research and the research is a big part of it.
12:35
Speaker 1: Insurance is a good example. If I were buying car insurance... Mm-hmm.
12:39
Speaker 1: I would be happy to have Muse or somebody else
12:42
Speaker 1: go to all the different websites, fill in my information,
12:47
Speaker 1: get me a bunch of quotes, and then pick one.
12:49
Speaker 1: Maybe with car insurance, I might still want the final decision,
12:55
Speaker 1: like an approval gate. To me, that's still agentic if
12:59
Speaker 1: it's doing everything.
12:59
Speaker 3: I mean, that's a big purchase, and it's a lot
13:02
Speaker 3: of money. And if it gets it a little bit wrong,
13:05
Speaker 3: you're paying... you know, or buys the wrong policy. Like
13:09
Speaker 3: that one feels way scarier to me than milk.
13:12
Speaker 1: Yeah, but give me the chance to approve it. And
13:13
Speaker 1: that's all I need. You give me all the data.
13:15
Speaker 1: You've done all the work. In effect, from the brand's perspective,
13:20
Speaker 1: AI is the customer. It's doing all of the work.
13:22
Speaker 1: And I'm just kind of saying, okay, everything looks good.
13:25
Speaker 1: Let's proceed. That's the one right I want with insurance.
13:28
Speaker 1: But with groceries, I don't even need to approve it.
13:31
Speaker 1: I'd say just do it. If it breaks two out
13:33
Speaker 1: of three times or three out of three times, I'm
13:35
Speaker 1: not repeating it. But if it works all three times,
13:37
Speaker 1: I'll do it forever.
13:38
Speaker 3: So let's talk about what happens when something goes wrong.
13:43
Speaker 3: If the agent does something I don't want it to do,
13:45
Speaker 3: it makes a mistake. If it goes off the rails,
13:49
Speaker 3: we know that businesses are split over who should be
13:53
Speaker 3: responsible in this kind of case when there's a dispute
13:56
Speaker 3: between the AI platform and the customer. And it's clear
14:01
Speaker 3: that most businesses think that the merchant who fulfilled the
14:04
Speaker 3: order should not be the one who's financially responsible when
14:08
Speaker 3: things go wrong. And so the question is, who should
14:11
Speaker 3: be on the hook if the AI, say, makes an
14:14
Speaker 3: unauthorized purchase?
14:16
Speaker 1: That's a great question. I think anyone who's pushing agents
14:20
Speaker 1: to make them mainstream for shopping, any company that's doing
14:25
Speaker 1: that will have to assure customers, we are on the hook,
14:28
Speaker 1: we'll take care of it. So that's when it goes mainstream.
14:32
Speaker 1: I think that's when you get past the trust hurdle.
14:35
Speaker 1: So if Muse wants to do that for me, or Instinct,
14:39
Speaker 1: another app or startup that wants to do this for me,
14:43
Speaker 1: if they want me to go full agentic, they've got
14:46
Speaker 1: to take on the liability risk and say that if
14:50
Speaker 1: our agent does something that deviates from this intent that
14:54
Speaker 1: you had, then we will reimburse you.
14:57
Speaker 3: You mentioned Muse. I mean, that's the meta... agent that
15:01
Speaker 3: is new and buzzy and instinct, which is sort of
15:05
Speaker 3: a startup, right, that wants to do something similar. Invite only,
15:09
Speaker 3: I believe. I haven't been invited yet.
15:12
Speaker 1: I can invite you.
15:13
Speaker 3: Okay.
15:13
Speaker 1: I'll do that.
15:14
Speaker 3: I mean, I'm curious, what do you think it's going
15:16
Speaker 3: to take? You know, there is this leap that a
15:19
Speaker 3: lot of consumers haven't yet made. That leap is letting
15:22
Speaker 3: an AI agent buy something for them. And I'm curious,
15:27
Speaker 3: what is it going to take for that to happen?
15:29
Speaker 3: And And do you think agents like Muse are going
15:32
Speaker 3: to be part of the key to this transition?
15:36
Speaker 1: Yeah, I think agents like Muse are key to the transition.
15:39
Speaker 1: So what Muse and other agents can do is you
15:42
Speaker 1: can articulate your preference, just like you do in a
15:45
Speaker 1: prompt on an LLM. They can go to different websites,
15:48
Speaker 1: gather the information, reason through that, fill forms if need be,
15:54
Speaker 1: make a decision, present it to us. So all of that,
15:57
Speaker 1: those systems do. And like I said, I use it
16:01
Speaker 1: up to that point. They can also go and transact
16:04
Speaker 1: and that's where they need to get past the trust hurdle.
16:06
Speaker 1: And I think for that to happen, first, they need
16:11
Speaker 1: to insure me about things going wrong, as you were
16:14
Speaker 1: just alluding to earlier. So they have to tell me
16:17
Speaker 1: that there is some protection. Just like, by the way,
16:21
Speaker 1: a credit card company, when it's giving us a credit card,
16:23
Speaker 1: they're saying if there's some fraudulent transaction, You're not liable.
16:28
Speaker 1: You can appeal and you can say, this was not me.
16:31
Speaker 1: Something's gone wrong. And they will make sure that no
16:35
Speaker 1: harm is done. And they're providing that guarantee. That's what
16:39
Speaker 1: needs to happen in this world. So someone needs to
16:42
Speaker 1: be the guarantor or the trust builder here. So that's
16:46
Speaker 1: the piece that is needed beyond just the technology needs
16:49
Speaker 1: to work.
16:50
Speaker 3: And it seems like that will happen, right? It seems
16:53
Speaker 3: like the nature of the history of finance is... lowering
16:59
Speaker 3: friction is the way things go. There's less and less
17:03
Speaker 3: friction to buy things. A credit card is lower friction
17:06
Speaker 3: than cash. Shopping online is lower friction than going to
17:08
Speaker 3: a store. And it seems like ultimately, agentic AI will
17:12
Speaker 3: be lower friction still, and someone will figure out how
17:15
Speaker 3: to solve the trust problem, and people will use agents
17:19
Speaker 3: to buy things on their behalf as often as we
17:21
Speaker 3: buy stuff online now.
17:23
Speaker 1: Yeah. I mean, look, I'm not a betting man, but
17:25
Speaker 1: if you were to say, hey, What's your prediction here?
17:29
Speaker 1: Pick a side. For me, it would be, you know,
17:32
Speaker 1: there's no doubt the history of shopping shows that we're
17:37
Speaker 1: always willing to adopt technologies that add convenience while guaranteeing trust. Now,
17:43
Speaker 1: is it going to happen for every kind of shopping?
17:46
Speaker 1: Maybe not. Is it going to happen for every person
17:49
Speaker 1: on earth? Maybe not. But is it going to be mainstreamed?
17:53
Speaker 1: For large pockets of shopping, my bet is that it
17:56
Speaker 1: will happen.
18:01
Speaker 3: We'll be back in just a minute. So we've been
18:14
Speaker 3: talking essentially about the consumer point of view, about what
18:18
Speaker 3: it's going to take for people to let agents buy
18:22
Speaker 3: stuff on their behalf. I want to talk now about
18:25
Speaker 3: the merchant point of view, about the retailer point of view.
18:28
Speaker 3: This is a thing you have written about. You wrote
18:31
Speaker 3: an article about it recently in the Harvard Business Review. And,
18:35
Speaker 3: you know, it does seem like most business owners largely
18:40
Speaker 3: trust AI platforms to represent their brand to customers. But
18:45
Speaker 3: it is clear that there is, you know, to some extent,
18:47
Speaker 3: a trust issue here, that the AI won't accurately represent
18:52
Speaker 3: the brand. What are you seeing here? What are you
18:55
Speaker 3: hearing from the merchants you talk to?
18:57
Speaker 1: Yeah, I think merchants have to worry about this too.
19:01
Speaker 1: And there's many layers of the merchant concerns. One is,
19:05
Speaker 1: are the agents representing the brands correctly? And so I'll
19:10
Speaker 1: give you an example. I was talking to this seller
19:14
Speaker 1: of vintage clothes in London. And so this person owns
19:19
Speaker 1: this brand. They sell in stores. They also sell online.
19:23
Speaker 1: And they sell vintage women's clothing. He actually bought the
19:27
Speaker 1: brand off of the previous owner. And he was kind
19:31
Speaker 1: of saying, it's shocking that we are the number one
19:36
Speaker 1: seller of vintage clothes in the UK. And you go
19:41
Speaker 1: to ChatGPT, you go to Gemini, and you ask this question.
19:45
Speaker 1: I'm in London. Where can I buy vintage clothes with
19:49
Speaker 1: 70s look, 60s look, and things like that? or Vintage
19:52
Speaker 1: Inspired rather is their brand, and his brand would never
19:55
Speaker 1: show up. And he was like, what's going on? And
19:59
Speaker 1: so I actually went through this a little bit. We
20:02
Speaker 1: found what was happening was that there was a post
20:05
Speaker 1: on Reddit where someone had mentioned that this brand is
20:09
Speaker 1: shutting down. And the history there is that the previous
20:14
Speaker 1: owner had put it up for sale And probably the
20:19
Speaker 1: word got around that the previous owner was struggling to
20:21
Speaker 1: make it work. And then, of course, new buyers took over.
20:24
Speaker 1: But there was this post online. So ChatGPT had probably
20:27
Speaker 1: decided that it shut down or something like that and
20:30
Speaker 1: was not showing the brand. And so that's an example
20:34
Speaker 1: of the brand being misrepresented by an LLM.
20:40
Speaker 3: What do you do about that? Like, this is a problem,
20:44
Speaker 3: and presumably it's not in the LLM's interest, right? The
20:46
Speaker 3: maker of the model doesn't want it to be wrong.
20:49
Speaker 3: The merchant doesn't want it to be wrong. Is this
20:52
Speaker 3: a solvable problem?
20:54
Speaker 1: It's a tough one to solve because the maker of
20:56
Speaker 1: the LLM is not going to say, every time you're
21:00
Speaker 1: not happy with an answer, appeal to us, and then
21:02
Speaker 1: we'll have a human look through it and correct it.
21:04
Speaker 1: That doesn't work for them. And so while it is
21:09
Speaker 1: in their interest, there's just too much information on the web,
21:14
Speaker 1: too many vested interests. And then how do you also
21:18
Speaker 1: address people who might complain about subjective things and say that,
21:22
Speaker 1: how come I'm not listed as the top three vintage brands?
21:26
Speaker 1: And in fact, our survey shows that we're among the
21:29
Speaker 1: top three. And so they're not going to provide resources
21:33
Speaker 1: for that. So unfortunately, what it ends up becoming is
21:37
Speaker 1: This is what I tell a lot of brands, that
21:39
Speaker 1: this world is a lot like your customers are sometimes misinformed.
21:45
Speaker 1: Your customers also need to be persuaded. And just like
21:48
Speaker 1: you've spent decades thinking about, how do I ensure I
21:52
Speaker 1: inform my customers about my brand? How do I make
21:55
Speaker 1: sure I persuade my customers about your brand? You have
21:59
Speaker 1: to do the same for AI agents now. And you
22:02
Speaker 1: have to start thinking about the AI agents as another customer,
22:05
Speaker 1: And the same questions you had to have for human customers,
22:08
Speaker 1: now you need to answer those for AI agent customers.
22:12
Speaker 1: And it's a new problem for marketers and brands and
22:16
Speaker 1: so on. So, right.
22:18
Speaker 3: So, I mean, the kind of central theme of the
22:20
Speaker 3: merchant point of view section of your article is companies, merchants,
22:26
Speaker 3: marketers need to think about the agent as a customer.
22:29
Speaker 3: And sort of you dig into that and you lay
22:32
Speaker 3: out these three common misconceptions that you have heard, you know,
22:37
Speaker 3: in your work with retailers and marketers. And so I
22:40
Speaker 3: actually want to walk through those and have you kind
22:42
Speaker 3: of talk us through them. So I'm going to read
22:44
Speaker 3: them to you and then I'm going to ask you
22:45
Speaker 3: what they mean. So the first one, misconception number one,
22:49
Speaker 3: is there will always be a human at the other end.
22:53
Speaker 3: Tell me about that. Like, what are you hearing? Why
22:56
Speaker 3: is that a misconception?
22:58
Speaker 1: Yeah, so I think a common one that a lot
23:02
Speaker 1: of marketers and brands think is, okay, AI is probably
23:06
Speaker 1: another channel like search, where in search engines, we have
23:11
Speaker 1: to think through how we show up in Google's results pages,
23:14
Speaker 1: but there's always a human reading it and making the decision.
23:18
Speaker 1: And so if you think about it that way, then
23:19
Speaker 1: you would say, hey, maybe the human is doing the
23:21
Speaker 1: research through chat GPT, but then the human's making the
23:25
Speaker 1: final decision. But I would argue that the direction in
23:30
Speaker 1: which we are headed is where people will say there
23:33
Speaker 1: are certain decisions where, you know, I don't need the
23:38
Speaker 1: headache of going through all of this stuff. It's not
23:40
Speaker 1: a lot of joy for me to sit and comb
23:43
Speaker 1: through the information and make sense of it. And I'd
23:46
Speaker 1: rather outsource it.
23:47
Speaker 3: Outsource it to an AI agent.
23:50
Speaker 1: To an AI agent. Yeah. Right. So groceries, we already
23:54
Speaker 1: talked about as one example of that. Yeah. There could
23:57
Speaker 1: be many others. I gave the example in the article
23:59
Speaker 1: about like tickets to a music concert. So I know
24:04
Speaker 1: I'm interested in, let's say, Coldplay concert. I want that
24:08
Speaker 1: ticket or in my daughter's case, you know, Taylor Swift,
24:12
Speaker 1: and we want that ticket. So I just tell an
24:14
Speaker 1: AI agent, hey, listen, just keep searching on the web.
24:17
Speaker 1: If it's under a certain price, just go buy it.
24:19
Speaker 1: If it's anywhere in my vicinity, go buy it. Because
24:23
Speaker 1: if you're going to find it and then flag it
24:26
Speaker 1: for me, but I'm busy in a meeting and I
24:28
Speaker 1: don't respond fast enough. And then the tickets are gone.
24:31
Speaker 1: I'd just rather just go search. If you get one,
24:34
Speaker 1: then go just buy it. So those kinds of things,
24:36
Speaker 1: I kind of feel like, you know, they're going to
24:39
Speaker 1: train people more and more to use agents. So this
24:42
Speaker 1: idea that a human being is making the decision is
24:46
Speaker 1: not quite there in those instances. So You know, it's
24:51
Speaker 1: the agent that is getting informed, the agent that is
24:54
Speaker 1: getting persuaded to make choices for the most part, the
24:56
Speaker 1: AI agent.
24:58
Speaker 3: Did you see Coldplay?
25:00
Speaker 1: I saw both of those concerts, actually.
25:05
Speaker 3: Okay, back to your article, back to that list of misconceptions.
25:08
Speaker 3: We are at merchant misconception number two, and that misconception
25:12
Speaker 3: is we just need AI crawlers to read our content.
25:17
Speaker 3: And in fact, most businesses say that they have started
25:20
Speaker 3: structuring their catalogs for AI, although for the most part,
25:23
Speaker 3: they're not all the way there yet. They're not fully ready.
25:26
Speaker 3: But why is this a misconception? Why is it a
25:29
Speaker 3: misconception that merchants think we just need AI crawlers to
25:33
Speaker 3: read our content? What's going on here?
25:35
Speaker 1: Well, I would say it's necessary but not sufficient to
25:39
Speaker 1: succeed in a world where AI agents might be shopping
25:43
Speaker 1: on behalf of customers. It's necessary because, yes, you want
25:46
Speaker 1: to make sure AI crawlers can read your content and
25:49
Speaker 1: understand how you show up. But that's not enough because
25:53
Speaker 1: their answers to customer questions or the way they reason
25:57
Speaker 1: through all of the information is not just informed by
26:02
Speaker 1: reading through your page and making sense of it. They're
26:05
Speaker 1: also reading a lot of other stuff. For example, they're
26:07
Speaker 1: reading third-party reviews. Some may be biased. Some may be objective.
26:12
Speaker 1: They're reading through those, making sense of it. they're increasingly
26:15
Speaker 1: starting to process videos. So I increasingly see that Google
26:20
Speaker 1: AI Answers and Gemini Answers are shaped by YouTube videos increasingly.
26:26
Speaker 1: They're shaped by Reddit discussions. So there's a lot of
26:28
Speaker 1: other things that goes into them. So the fact that
26:31
Speaker 1: they read your content does not mean that they are
26:37
Speaker 1: going to make decisions that are favorable to the brand.
26:40
Speaker 1: So just like every marketers had to think through how
26:44
Speaker 1: do you persuade humans? They have to think about how
26:47
Speaker 1: do you persuade AI? But with humans, we understand a
26:49
Speaker 1: few things. We understand $ 9. 99 is not treated the same
26:54
Speaker 1: as $ 10. It doesn't make any rational sense, but it
26:57
Speaker 1: is true that people sort of react to these prices differently.
27:02
Speaker 1: We also know that social proof matters and saying that
27:07
Speaker 1: your friends have bought this matters or you know that
27:09
Speaker 1: scarcity matters for people. Only three items in stock can
27:13
Speaker 1: influence somebody to buy it.
27:14
Speaker 3: Buy now.
27:16
Speaker 1: Buy now, right?
27:16
Speaker 3: Exactly.
27:17
Speaker 1: So we know those things matter for humans. But if
27:19
Speaker 1: I were to ask any marketer, what are these biases
27:22
Speaker 1: that AI agents have? Or how do you persuade an
27:25
Speaker 1: AI agent? The answer would be a blank because none
27:28
Speaker 1: of us truly understand it. So that's why it's important
27:32
Speaker 1: to understand that misconception and really start thinking hard about
27:36
Speaker 1: how we correct that.
27:37
Speaker 3: Does anybody know anything about how to persuade an AI
27:41
Speaker 3: agent or a language model making recommendations?
27:45
Speaker 1: We're starting to understand some basics. And so, for example,
27:50
Speaker 1: I just mentioned that with Google, I have enough data
27:54
Speaker 1: to suggest that YouTube videos matter. And so in your
27:58
Speaker 1: marketing strategy, you should be thinking about influencers and you know,
28:05
Speaker 1: the relevant, who are the relevant influencers? Which influencers are shaping, say,
28:11
Speaker 1: Gemini or AI answers in my business vertical? And who
28:14
Speaker 1: should I go talk to? Or understanding, you know, sometimes
28:20
Speaker 1: even page layouts and how they matter. For example, and
28:24
Speaker 1: this could change, by the way, what I'm about to
28:26
Speaker 1: say could change in six months because new models can
28:28
Speaker 1: come in and they may reason through information differently. But
28:31
Speaker 1: I've already seen, for example, that presenting information in the
28:35
Speaker 1: form of tables somehow has a greater impact than having
28:41
Speaker 1: the information in paragraphs, especially when you're comparing things. So why,
28:46
Speaker 1: who knows? I mean, but for some reason, if that
28:50
Speaker 1: same comparison is presented as a table versus three paragraphs,
28:55
Speaker 1: it somehow seems to persuade AI differently. So there's a
28:59
Speaker 1: lot of small things and big things.
29:01
Speaker 3: So at least for now, language models love tables.
29:05
Speaker 1: At least for now, they love tables. Don't ask me why,
29:08
Speaker 1: but that's what we find.
29:11
Speaker 3: Okay, so we're going back to your list of common
29:13
Speaker 3: misconceptions that marketers, retailers, merchants have about AI. The last one,
29:18
Speaker 3: misconception three, is AI is just another channel. Why is
29:23
Speaker 3: that a misconception?
29:27
Speaker 1: Yeah. I mean, I think the most common way in
29:29
Speaker 1: which brands perceive the impact of AI is to think
29:33
Speaker 1: of it as another channel. Your customers can come to
29:36
Speaker 1: you directly. They can come to you via social. They
29:40
Speaker 1: can come to you via search. And now they can
29:43
Speaker 1: come to you via AI search or via AI agents.
29:48
Speaker 1: And that perspective just says that, okay, we've got a
29:52
Speaker 1: bunch of channels and we need to show up in
29:54
Speaker 1: this other channel. And what I worry about that is
29:58
Speaker 1: that you are mistaking something that I view as a
30:03
Speaker 1: system level shift in how buying is happening And you're
30:09
Speaker 1: mistaking that for another distribution channel that emerged. The analogy
30:13
Speaker 1: I would give is that when Netflix got started, all
30:17
Speaker 1: the movie studios said, okay, here's another channel through which
30:21
Speaker 1: I can distribute my content. I can get my movies
30:24
Speaker 1: out via the movie theater. I can get it out
30:27
Speaker 1: via pay-per-view on TV. I can also give it to
30:31
Speaker 1: various TV channels or I can give it to Netflix
30:34
Speaker 1: through streaming. And this is yet another channel And what
30:38
Speaker 1: they realized later was that, whoa, it's not just another
30:41
Speaker 1: channel because what's happening is it's changing my customer behavior
30:45
Speaker 1: and training them to work in entirely different ways. And
30:48
Speaker 1: it's changing who has the power and so on. And
30:51
Speaker 1: ultimately Netflix took over that industry because Netflix said, hey,
30:55
Speaker 1: maybe I don't need your content anymore. I can just
30:57
Speaker 1: make my own content and deliver it and so on.
31:01
Speaker 1: And so I think there are similar analogies here, which
31:04
Speaker 1: is to say that If you just think about it
31:06
Speaker 1: as another channel, you're missing two things. You're missing one,
31:10
Speaker 1: the fact that your customer behavior is changing in big ways.
31:14
Speaker 1: And as a result, you could lose market share because
31:17
Speaker 1: if all you're thinking is it's another channel and I'm
31:19
Speaker 1: on that channel, your competitor that is thinking smartly about
31:23
Speaker 1: how to inform and persuade this new type of customer
31:27
Speaker 1: AI agents will be ahead of you and will steal
31:29
Speaker 1: market share from you. The second thing is you're also
31:33
Speaker 1: missing how the power dynamics might change. If people are
31:38
Speaker 1: doing more shopping through, I'll pick an LLM like say ChatGPD,
31:44
Speaker 1: then does it shift the power balance and those LLMs
31:48
Speaker 1: become more powerful? By the way, Amazon has acknowledged that
31:51
Speaker 1: this could be a huge impact on them, which is
31:54
Speaker 1: why Amazon is reluctant to provide its full product feed
31:58
Speaker 1: to the ChatGPDs of the world. And instead, Amazon is
32:02
Speaker 1: building its own AI agentic shopping experience, which is called Rufus,
32:08
Speaker 1: and trying to persuade people, hey, have AI do your
32:12
Speaker 1: shopping for you, but do it within our Amazon environment.
32:16
Speaker 1: That's because Amazon recognizes that the power balance will shift
32:20
Speaker 1: if the agent doing the shopping is not their agent,
32:23
Speaker 1: but a third party, because then they're at the mercy
32:25
Speaker 1: of the third party deciding whether it'll finally buy it
32:28
Speaker 1: from Walmart or whoever else. And so they recognize that.
32:32
Speaker 1: And that's an example of saying AI is not another channel.
32:36
Speaker 3: Well, I mean, that last example is compelling, right? Amazon
32:39
Speaker 3: is one of the few retailers with the scale and
32:43
Speaker 3: the money and the technical expertise to be able to say, oh, no,
32:47
Speaker 3: we're not going to let these outside AI agents shop
32:50
Speaker 3: with us. We're going to build our own. Like, clearly,
32:54
Speaker 3: businesses are worried that If they don't adopt agentic commerce,
32:58
Speaker 3: then they risk losing customers or sales to competitors. So
33:03
Speaker 3: if you're a small company, what should you do?
33:05
Speaker 1: Yeah, I think you need a holistic strategy. So by
33:08
Speaker 1: the way, when we talked about e-commerce, it's not like
33:10
Speaker 1: stores don't exist today. They are there. And a good strategy,
33:13
Speaker 1: like we talked about Warby Parker, figures out how do
33:16
Speaker 1: we mix stores for humans who want that experience walking
33:22
Speaker 1: into a store and trying on glasses with an online
33:25
Speaker 1: strategy for humans who want the convenience of going to
33:28
Speaker 1: a website and with four clicks getting their eyeglasses. And
33:32
Speaker 1: so it may be a strategy where you say, how
33:35
Speaker 1: do we have a strategy where we provide an experience
33:39
Speaker 1: for humans who want to shop themselves? And also an
33:43
Speaker 1: experience for agents who have already taken over a big
33:46
Speaker 1: chunk of the shopping work on behalf of humans. And
33:50
Speaker 1: so how do you support both?
33:51
Speaker 3: Although now you're walking back to the It's Another Channel, right?
33:54
Speaker 3: Like that is the It's Another Channel story. But the
33:57
Speaker 3: Netflix analogy suggests that you might end up getting eaten
34:03
Speaker 3: in the end.
34:04
Speaker 1: Good pushback. But by the way, notice that I did
34:07
Speaker 1: not say, how did you, how do you support humans
34:11
Speaker 1: who want to shop on their own with humans who
34:13
Speaker 1: want to shop through AI. What I said was, how
34:15
Speaker 1: do you support humans who want to shop and AI
34:19
Speaker 1: agents who are shopping? And so I'm really saying it's
34:23
Speaker 1: another type of customer. It's another audience and you need
34:26
Speaker 1: a real strategy for that audience. And so for that,
34:29
Speaker 1: it might mean, I'll give you again an example. You
34:32
Speaker 1: might say that Either the way I respond to it,
34:36
Speaker 1: I have to make shopping so enjoyable and so luxurious
34:40
Speaker 1: and so high-end that it's worth it for the human
34:43
Speaker 1: to show up. And a very high-end luxury brand would
34:48
Speaker 1: have to ask, how do I make it so amazing
34:51
Speaker 1: that the person says, I'm not asking any AI agent
34:54
Speaker 1: to do anything. I want to show up at the
34:56
Speaker 1: car dealership and I want to sit in this car
34:59
Speaker 1: and experience it. And that is what shopping is going
35:02
Speaker 1: to be. So you either make that call or you've
35:06
Speaker 1: got to make the other call, which is to say
35:09
Speaker 1: our primary customers will be AI agents. And so what
35:14
Speaker 1: I'm going to do now is have a team that
35:18
Speaker 1: is solely focused on winning with AI agents. And that
35:22
Speaker 1: is to ask, how do we inform these new customers?
35:25
Speaker 1: How do we persuade these new customers? And unleash this
35:28
Speaker 1: new team to go chase after that business And make
35:32
Speaker 1: that work. I think you've got to first make that
35:34
Speaker 1: strategic choice. And once you've made that, your playbook is
35:38
Speaker 1: going to be extremely different for these two strategies. But
35:42
Speaker 1: that second strategy is, again, I'm still saying it's not
35:45
Speaker 1: another channel. It's here's your new customers. How do you
35:48
Speaker 1: win those customers? That now means you hire new researchers
35:53
Speaker 1: who are figuring out how are these AI agents reasoning?
35:57
Speaker 1: I've actually just created a new startup called Bodium Labs,
36:00
Speaker 1: and what we are doing is doing that on behalf
36:02
Speaker 1: of brands. We are basically building systems that brings in
36:06
Speaker 1: interpretability or explanations for how AI systems reason. So it's
36:11
Speaker 1: no longer this channel. It's almost just like you have
36:14
Speaker 1: a data science team that's thinking about how do you
36:16
Speaker 1: inform and persuade humans. You have an AI research team
36:19
Speaker 1: that's thinking about how do you inform and persuade AI.
36:26
Speaker 3: We're going to take an ad break now and we
36:28
Speaker 3: will be back in just a minute. Your HBR article
36:40
Speaker 3: lands with this idea that AI is a new kind
36:44
Speaker 3: of customer, right? That the, at this moment, we're in
36:47
Speaker 3: this hybrid moment where people are using AI, but people
36:49
Speaker 3: are still clearly the customer. Um, we will probably soon
36:55
Speaker 3: be in a place where the AI is actually making decisions,
36:57
Speaker 3: actually buying things. And so like, ultimately, what does that mean?
37:01
Speaker 3: What does that world look like? How do you think
37:03
Speaker 3: about it?
37:04
Speaker 1: I'll share one thought I have about how it plays out.
37:09
Speaker 1: By the way, in software settings and in B2B settings,
37:13
Speaker 1: we already see it. You asked me for early examples
37:15
Speaker 1: where this is already happening. Many people are building their
37:20
Speaker 1: online storefronts through tools like Cloud Code. And they're not
37:28
Speaker 1: hiring developers to go build it. They'll just say, hey,
37:30
Speaker 1: build a storefront for me. I want to sell XYZ
37:33
Speaker 1: and Cloud Code or Lovable or these other no-code tools
37:39
Speaker 1: that will have AI write the code, no humans writing
37:41
Speaker 1: the code, will build the storefront. When they're doing that,
37:45
Speaker 1: they're making decisions. They're making, for example, how are you
37:47
Speaker 1: going to accept payments? They, the agents. Yeah.
37:52
Speaker 3: So the agents are making decisions about things like payments,
37:55
Speaker 3: these fundamental business trust things.
37:58
Speaker 1: Yeah. Yeah. And if you now take that to the
38:01
Speaker 1: consumer world, I think increasingly we will have a shopping assistant.
38:06
Speaker 1: I don't know who wins that. Is that meta with
38:08
Speaker 1: Muse or is that going to be a next generation
38:11
Speaker 1: thing that Apple launches with Siri and Maybe in the future,
38:17
Speaker 1: or Google does something, but it might be on the
38:19
Speaker 1: phone itself. It might come built into the phone. But
38:22
Speaker 1: a system where you might say, hey, I want to
38:26
Speaker 1: buy this kitchen timer for when I'm cooking in the kitchen,
38:33
Speaker 1: or I want to buy a new phone, or I
38:36
Speaker 1: want to buy an alarm clock because I don't want
38:39
Speaker 1: to set it on my phone, whatever the reason. But
38:42
Speaker 1: you might just speak that into the AI. And you
38:45
Speaker 1: might either say, do all the work, make a decision,
38:50
Speaker 1: confirm and get my approval before buying, or you might say,
38:53
Speaker 1: proceed to buy it and you outsource that. And it's
38:57
Speaker 1: going to be that seamless for a customer. But in
39:01
Speaker 1: that world, we then have to think through what does
39:04
Speaker 1: that mean for businesses? You know, how do you stay
39:07
Speaker 1: relevant as a business when you're branding? What does brand
39:12
Speaker 1: even mean now in a world where AI agents are
39:14
Speaker 1: deciding that? What does persuasion mean when AI agents are
39:18
Speaker 1: making the choices? What does advertising look like in that
39:22
Speaker 1: kind of environment? How do you advertise to AI is unclear,
39:26
Speaker 1: but all of those will have to evolve. And I
39:28
Speaker 1: think there will be a whole AI economy coming up.
39:32
Speaker 1: When AI is buying and selling, you might have a
39:34
Speaker 1: whole new layer of the economy open up with services
39:38
Speaker 1: targeting as either helping brands sell to AI or helping
39:43
Speaker 1: AI make better decisions when buying on behalf of some people.
39:47
Speaker 1: And so I think that's an interesting new era that's
39:50
Speaker 1: potentially coming up in the next several years.
39:53
Speaker 3: It was great to talk with you. Thank you so
39:54
Speaker 3: much for your time.
39:56
Speaker 1: Thank you for having me. It was my pleasure.
40:05
Speaker 3: Karthik Hosanagar is a professor of operations and marketing at
40:09
Speaker 3: the Wharton School at the University of Pennsylvania. Today's show
40:13
Speaker 3: was produced by Gabriel Hunter Chang, engineered by Sarah Bruguier,
40:17
Speaker 3: and our showrunner and editor is Jake Harper. I'm Jacob Goldstein.