The Death of Search: How Shopping Will Work in the Age of AI

Authors: Alex Rampell & Justine Moore, a16z; Translator: Felix, PANews (Original article edited and abridged)

a16z General Partner Alex Rampell and Partner Justine Moore discuss how AI agents are changing business and their impact on Google's business model, affiliate marketing, online shopping, and more.

Why are you thinking about the role of AI in business?

Erik Torenberg: You've been thinking about the role of AI in business for a long time. Can you talk about what inspired this article and how these ideas came about?

Alex Rampell: A long time ago, I founded a company called TrialPay. In fact, I've been selling things on the internet for a very long time, even before the internet existed. I was wondering, what will happen to Google first after the advent of AI? Because many people are wondering this: will search volume increase or decrease? Personally, my search volume is decreasing, but not for commercial searches, but for all non-commercial searches. That's one thing. TrialPay is one of the largest affiliate marketers in the world. Affiliate marketing essentially involves referring customers to others. It's one of the oldest business models on the internet, actually predating AdWords and AdSense. You refer someone somewhere, and you get a share of the revenue, or a commission. But will the affiliate model truly become the driving force behind new commerce? On the other hand, consider these very expensive items. You're using AI to deeply analyze them. But without an affiliate model, how do you conduct commerce and transactions? So, first, the ontology of commerce is very interesting. Second, does the entire affiliate model still make sense? Because that seems to be where ChatGPT and other companies are venturing. And thirdly, and this is just me personally, I probably use ChatGPT three orders of magnitude more than I use Google right now, which is interesting. Justine Moore: There are several very large consumer markets right now, the largest of which is probably online shopping. But I think relatively few startups have tried to use AI to address this market so far, although there are more opportunities now because of these very smart language models and agents that can help you make better decisions than you could, and even buy things for you. You might think that this would create opportunities for more people to package this content into products and offer them to consumers, but we haven't seen many people doing that yet. So I think this section is about digging deeper into why this system is so complex, what different types of purchases can AI play a role in, and what do you want to see when you think about the broader market? To what extent will AI enable dynamic pricing? Erik: Alex, can you imagine a world with extreme dynamic customized pricing? Like, you see the same thing on Amazon, but they charge you more, presumably because you have more money? Alex: Well, I mean, people have tried this many times. From an introductory economics perspective, it's certainly a very clever world. Like, how do you capture consumer surplus? Consumer surplus benefits consumers and hurts producers. Apparently, Delta has tried this, or they've tried to. It's like a poor-man's-footprint scenario, where if you have an iPhone, you should pay more than someone with an Android phone because it's more expensive. This actually means your demand elasticity is different than someone with less money. I think there might be regulatory challenges with this. Of course, it would definitely be met with strong backlash from the customer base, but some people have tried this. But generally speaking, it's very difficult to succeed. Why is e-commerce so small? Erik: Before we get into that, let's look back at previous platform revolutions. E-commerce currently only accounts for 16% of total retail sales. If we'd discussed and predicted the share of e-commerce in commerce 20 years ago, we might have expected it to be much higher. But why isn't that the case? Alex: It turns out that the demand curves for instantaneous and non-instantaneous services are different. I used to serve on the board of a company called Wise, and I observed that their main business was remittances. I found that the market for instant remittances is completely different from the market for remittances that arrive two days later. Because if you send something two days later, the demand is probably not as high. It's surprising that you said it's only 16%; that seems very low. Justine: I think that's a high number. I'm not questioning your data analysis. I think that's because I think there are so many cases where people do research online and then go to a physical store to buy. Alex: The hardest thing, and relevant to this topic, is attribution. It's a pain point for everyone. For example, how do I attribute Justine's MacBook sale? And I think the most pervasive and corrosive business model on the internet is "last-click attribution." You give 100% of the credit to the last click. A lot of people can't distinguish between correlation and causation and fall into this trap, and that's the business model I hate the most, like Honey. It's the kind of thing where you're already on a webpage, ready to buy, and it asks, "Do you want a coupon code?" Of course, why not? 10% off. Click here. You click here. It redirects you to an affiliate page. It places a cookie on your device. It then redirects you back to the page you were just on, effectively stealing attribution for that visit. Why are e-commerce brands struggling? Erik: Let's revisit the industry. It seems like the biggest winners are at the aggregator level, like Shopify or Amazon, and the big independent brands like Allbirds or Casper. They seem to have achieved huge revenues quickly, but haven't been able to become the same sustainable businesses they were before. Their performance hasn't improved as they've scaled. Can you talk about the overall situation in the industry and why this is happening? Alex: It's probably because the products aren't actually manufactured; someone else is usually making them, and that's not a good business model. E-commerce brands are just resellers of goods, and I think that's the problem. Just like what happened in the internet 1.0 era, the long tail of retailers is essentially starting to die out. Justine: I think there are things like Allbirds shoes or makeup that are very dependent on trends, especially in the internet age. Nothing stays popular forever. For example, one year Allbirds is the hottest shoe, and the next year everyone's all about retro Adidas. Now On sneakers are all the rage. This year, I was watching a TikTok video of the University of Alabama's Rush sorority, and every girl was wearing On sneakers, while last year they were all wearing Japanese-style New Balance shoes. Will AI threaten Google's business model? Erik: Let's get specific. What features will be taken away from Google? What features will remain? Alex: I think Google has always had a classic freemium business model. That is, they built a better search engine, and it was much better because of the way they linked. When Google first started, most searches were free because commerce on the internet was still in its infancy, and all searches were free, just information. I remember using Google when it first came out and thinking it was so much better than HotBot and everything else. It was all free, with no monetization involved. They ended up essentially copying Bill Gross's Overture business model, which was a project at IdeaLab that eventually became part of Yahoo. That's why Yahoo ended up owning a portion of Google, and if you look at the history, it was AdWords that made Google the $2 trillion behemoth it is today. What's happening now is that Google is starting to phase out some of the "free" aspects, but the "mium" (paid) aspects are still going strong. Google is eliminating some informational queries, like "Who won the Oscar in 1977?" Those aren't monetizable queries, but people still want to know. People are now asking about ChatGPT, and ChatGPT has 800 million weekly active users, which is a huge number. But the "mium," the paid portion of the freemium model, still exists at Google. How do I know? Look at their financials. Their financials are still growing, but we also know that search volume is actually declining. So, if they're not losing revenue, what are they losing? Alex: I would say the biggest threat to the internet right now is that it's unhealthy, or rather, the World Wide Web is unhealthy. It's unhealthy because a lot of things that once existed on the open internet now exist in so-called "walled gardens." By the way, search has long been fragmented. This wasn't caused by ChatGPT. For example, if you want real-time search, you have to go to Twitter or Platform X. If you want to search for your friends' information, you have to go to Facebook. These various "walled gardens" are one factor in its unhealthiness. Another factor in its unhealthiness is the commercialization of the internet, which isn't inherently a bad thing. But the question is, if you were looking for the best sneakers, who were the people writing about quality sneakers? In 1995, if you had a blog, first of all, you hosted it on your own website. You installed Apache on a server you built yourself, and you did it purely for the love of it. Affiliate links provided a revenue model, but they really polluted the then-open internet. It's like a top 10 running shoe list. You know what that is? To me, that meant the top 10 affiliate marketing revenue. Then I'd find someone to write a bunch of nonsense and optimize it to hell to make money. Think about it before the internet. There's a publication called Consumer Reports, and the cool thing about it is that it's the only publication that refuses to run ads. It's entirely subscription-based, with the idea that you can trust the reviews. We really needed that model, but the whole business model just disappeared. A lot of content is blocked now, and the content that's not blocked is filled with spam. Justine: Honestly, I think the least spammy channel I've ever seen is video. When I want an honest review, I go to unsponsored YouTube videos, which usually get a lot of views. Amazon is essentially a giant search engine. And it's polluted, too. A lot of what Amazon does is arbitrage. It's like a polluted ocean of garbage. Costco's Business Model Alex: By far, my favorite business model is Costco. I think Costco is the greatest company in the world because they refuse to sell low-quality products. They refuse to charge high margins. Why they would refuse... it makes no sense. Erik: Is it because they want to pass the profits back to the customer? Alex: No, because it would diminish the value of membership. They make money from membership fees, so they charge you about $100 a year. If you look at their net income, it's basically the number of members... They have over 50 million members, which is a huge number, multiplied by the membership fees, and that's their net income. Everything else is nothing. They started their own chicken farm because the cost of rotisserie chickens was too high. That's how they run their business, refusing to sell anything they're not happy with. Generic brands are just as good. So Costco is the best... Just like the business world we've talked about, like the pre-internet era, the internet era, the AI ​​era, Costco is immune to all of that because they're like Consumer Reports... They treat their customers incredibly well, and that's why this company is worth hundreds of billions of dollars. What shopping categories will AI eat into? Erik: Let's talk about what businesses will AI change? You mentioned a few different types of purchasing behaviors that could be "eaten." Justine: Yeah, so we looked at the spectrum of impulse purchases, and I think those purchases… I mean, they still are, like Coke, a Coke bottle in the aisle. But now a lot of people will, like in the TikTok store, see something while watching a video and think, "That T-shirt looks cool, I need to buy that," or even seriously consider buying something like a house, a wedding venue, or a car, which can cost a significant portion of your income. It's like a one-time or multiple purchase, and you do a lot of research. So I think both of those are very difficult to disrupt with AI. I think impulse purchases are because you don't research it beforehand, and you're not going to any specific place to buy it. Essentially, you see it and you buy it instantly. So, the algorithms are getting better and better at targeting you, so if you see a shirt with your dog's name on it in your TikTok feed, you're more likely to buy that than anything else. But that's a little different from what we call generative AI. Then there's the most complex part of the purchase process. I think it's hard to fully achieve end-to-end AI because while you might start doing online research on ChatGPT, Gemini, or any of the emerging AI-native products, the purchase itself is so meaningful that you'll want to have some kind of hands-on experience—to see, touch, and feel it. So that means within this whole category of products in the middle, we think purchasing could be disrupted by AI in a few different ways. One obvious way is through research. For example, let's say my bag is worn out from all my travels, and I need a bag that can hold my laptop, a large water bottle, and everything, and still fit in the overhead bin on an airplane... If you're busy, you don't have the time to do that research yourself. You could have an AI agent watch all the TikTok videos, read all the Reddit posts, gather real consumer feedback, and then make recommendations. Or you might want to click through the options yourself. But I would say that in this case, if there's a good integrated purchasing solution, you'd most likely be able to do it through an AI agent. There are also things where you already know what you want, like wanting the best price. So I think AI agents can do a lot with price optimization. For example, if you always buy a certain type of laundry detergent, it could search online and ask, "Where can I find the best price for that particular type of laundry detergent?" There's another type of purchase that I think can be mediated by AI to some extent, but it might also have some human influence. For example, with a bicycle, a sofa, or something a little more valuable like a laptop, you want someone to take the time to really understand all of your criteria and help you make the best purchasing decision. This is likely something you'll use for many years, and it's important to you that it works well, is the best choice, and doesn't become obsolete quickly. Today, I think the only way to achieve that is if people dig into Reddit threads like the buyitforlife forum or something like that. Or if they own a truly trusted brand, like Apple, and are willing to pay a premium. I think it's going to be interesting in the future to have an AI agent that really understands you deeply, so you can have a much deeper conversation about these kinds of things. Like, maybe even a phone call where they're dynamically asking you a bunch of questions back and forth, and you're giving them the information they need to go back and research and make a decision. That's kind of how we think about how AI can impact purchasing behavior.