ALEXANDER PADDINGTON: All right. Good afternoon, everyone. My name is Alexander Paddington. I’m a managing director and partner at Boston Consulting Group. It’s great. I hear there’s a super long line, so I hope the content lives up to the full room here. I’m very excited today to share some of our research of what consumers are saying and what they think and feel about agentic commerce. What are the real barriers to adoption, and what it takes to unlock that? We’re going to dive in deep and really think about: what do customers care about. So this is not about products. We’re not going to be talking about the newest lineup. We’re going to talk about what actually customers care about and what it takes to win them over. Now, if you look at the quote here, Bret Taylor, he’s the chair of OpenAI. He said, “For every company, AI agents will be their most important digital interface, just as important as a website or app.”

I would argue yes, but first we need to win consumer trust before we could get there. And how we win consumer trust is the focus for today. So let’s think about where we are today. The rate of change is accelerating faster than ever. We’ve looked at all the stats. I thought this morning’s keynotes—you saw how quickly things are growing over the last few months. And we see LLMs is really driving huge volume in terms of traditional search. If you think of what comes from ChatGPT, Claude, it’s 1.5 billion monthly visitors today. That’s over half of Google’s search volume. So you can just look at the line. You can see that quickly, this would be the dominant way to search going forward. And it’s not just about searching and asking random questions. It’s actually driving real commerce. 77% of the users are using this as search engine, ChatGPT.

And 64% of US shoppers, they bought something after they use an answer engine. So they’re using this to get insights, information, and it actually does influence their shopping behavior. And that’s a big change. We didn’t have this two years ago. This is all obviously adopted very quickly.

Now let’s go into some of the initial kind of consumer highlights. We’ve done research over the last 12 months in the US about how are consumers using AI today and how they feel about it and what’s going to actually get them to use it more. But this is actually a snapshot. In product discovery, do you use AI for which product categories? And the top one here is consumer electronics: 60% of respondents said they use AI as part of their research and discovery process. Travel, 50%. Grocery and apparel, over 40%. So we see that there are certain product categories that are being part of AI’s discovery, and that’s going to be critical as we think about what happens next. The other stat I wanted to highlight is on the right side. 64% of US consumers, they use AI—not just in one category, but three or more product categories.

So it’s not just, “Hey, I only use ChatGPT to do research for my vacation.” No, I’m actually using it for multiple product categories. And that shows that this adoption is going to start to grow across not just one, but across multiple categories in the future.

Now, we talked about travel, right? Travel’s a great use case. I want to go on vacation. I want to go to Portugal. I want to know how to get there, what flight, what hotel, should I have a travel, what kind of trip. But that’s actually, all that research you’re doing with ChatGPT or Claude is changing how traffic actually happens today. So we looked at airlines, and this is actually JetBlue’s. Their organic search traffic through traditional Google, traditional search engines, is declining. It’s declined 10% in the last year. But in the meanwhile, you think of all the AI engines—think of ChatGPT, Perplexity, and others—they’re growing 200 or 300%. So we say, over time, consumers are going to start to really discover products, services, and really do that through LLMs in terms of how they think about the future.

Now we’re going to make this a bit interactive, right? We have a lot to cover today, but if you could pull out your phone, we have a QR code. It’s also in the app. There’s two questions we’re going to poll the audience. And I will say, Stripe Sessions, you guys are very AI forward. It’s a very AI-forward conference, but I’m very excited to see what you think. The two questions are: which purchase category would you trust an AI agent to complete entirely on your behalf? So you’ll see a few different options. Pick which purchase category. Hey, that agent can do it fully autonomously. And the second question is, what are the barriers that you see preventing you from letting that agent transact? I mean, maybe many of you are like, “Okay, I’m good with agents.” But are there some things that actually do worry you? And what actually concerns you about letting an agent purchase services and products on your behalf? So use the QR code. We’re going to have some of the results towards the end of the session, and obviously very excited to see what this group is thinking today.

Now let’s go into a little bit more of our research. So we think about agentic commerce. There’s different evolutions of agentic commerce, and the way we think of it is really the past, but there’s like four evolutions of how we see this happening in the future. The past is, okay, I do my research, traditional merchants, go to websites, use Google. We think in the future, about 19% of consumers will still rely on that. They’re not going to go use AI agents or LLMs. They’re still going to do the kind of traditional way of doing research and purchasing products and services.

But where it gets exciting is we see a few evolutions. Evolution one is probably the most simple. I use ChatGPT as part of my discovery and kind of product comparison. We think 80% of consumers will start to use that in the future and they’re ready for that. Evolution two is, okay, not just product discovery, but we see that the agents actually able to do not just help me with discovery, help me to purchase. So actually they can transact on my behalf. We think up to 59% of US consumers can actually get to that stage. Now it starts to get exciting as we get to evolution three and four. Evolution three for us is, hey, the AI agents are actually working almost autonomously on their own to actually drive the transaction end to end. So from discovery in terms of selecting, in terms of purchasing, and obviously that becomes a true, almost very little human in the loop to get to that.

The critical thing here, we see 42% of consumers will actually let the AI agent do that in at least one product category. So 40% is a lot, right? There’s still 60% that maybe not, but 40% is a material number. And the last evolution for us is evolution four. Here’s the AI agent, not just does all the discovery, the purchase, selection. They actually will choose how to pay. They’ll choose, okay, let’s use this credit card, use this payment method, maybe they use Stripe Link. How do you get more rewards, more loyalty? They actually choose that on behalf of the customer themselves. That’s about 25%, we think, in the future. So we looked at that from a customer lens, right? We’ve done our surveys with US customers over the last 12 months. But the question is, how does that impact dollars, actual ecommerce spend in the future?

And I think a few things. Based on our consumer research, we estimate that 55% of all US ecommerce could be agent-assisted, and 18% will actually be autonomous agents in the future. And that’s obviously, this space is growing. It’s $2.3 trillion of ecommerce spend today. 55% we think reached that evolution one we talked through. 32% of spend will actually be in evolution two. And the 18% evolution three, that is material. So 18%, but $0.4 trillion, that’s a huge number and that number’s just going to grow, as you know, ecommerce has been growing every year. And this is a massive shift of how we think about ecommerce in terms of happening in the future.

Now let’s jump forward. We talked about it overall. Let’s deaverage it based on what consumers are actually saying. It’s not going to just be one or all my products I’m going to use agents for. It’s going to vary by the product category. And so when we’ve done our research, the most likely ones to have early adoption are some of the simpler ones: household supplies; ordering food from restaurants, maybe on DoorDash; hygiene and beauty products. We see consumers are ready to test autonomy when it is likelihood of failure is low and the cost of mistake is low. And that’s critical. I’m okay if the agent makes a mistake and gets me the wrong soap. It’s not the end of the world. I could have some soap. So this is a critical thing in terms of early signals we see of where growth will come. And let’s think about that.

So based on our research, consumers, how they want to test it? They want to test agentic commerce safely. And we’ve asked consumers a few—what are the most important factors that get them comfort to give full autonomy to an agent? The number one factor, and this will be very critical as you think about your kind of customer experiences and what the customers you serve, 50% said they want high familiarity, the brands or products they’ve purchased before. So if I bought that toothpaste in the past, I’m very happy for the agent to buy it. It’s a very clear defined product. 46% said, “Hey, if I can return it with a full refund, that also gives me full ability to actually trust the agent, and I’m not going to have, obviously, bear the risk.” 40% said low product prices. If the price is low, there’s probably not a big issue if something goes wrong.

And 40% said if there’s low downside risk. So I think a great example: I buy, the agent gets a wrong paper towel? Not a big deal. I could use a different paper towel. But if my agent buys the wrong couch, I think I’m going to be in trouble when I get home. So you really have to figure out what product category is that agent purchasing and that would drive your ability to trust it.

Now we talked about the product categories. We talked about those factors. We see that a lot of these barriers to agentic commerce that need to be overcome, they’re very emotional. They’re tied to trust, they’re tied to control, they’re tied to privacy. And we think that the unlock to get there, payment safeguard is going to be critical. So we ask consumers a few things. If we look on the left-hand side, what do you think are the top reasons that make you hesitate to use an agent? The 50% of consumers say, “We just don’t trust it. We don’t trust the agent.” The next biggest one is, “I don’t want to give up control,” 47%. A lot of consumers, they like to do it themselves. They want to do it end to end. They’re not willing to give it up to an agent. Over a third of consumers also said they worry about data privacy.

Am I actually willing to let the agent do that? I worry about fraud. What’s going to happen if something goes wrong, or the agent goes rogue? So those are real concerns that consumers have today. But on the flip side, we look on the right-hand side, what features will help you get comfort and actually move that needle to get you to use an agent? The number one thing we heard from consumers is, over 45% said, “Loyalty, points, cash back. If you could give me a better experience, better rewards, hey, that actually might get me to start using agent and actually that’s a way to encourage me.” The second one’s fraud protection. If something goes wrong, for whatever reason, will you give me protection to protect me in that issue? And the third biggest one: free delivery, right? I’m okay if they buy the wrong product online, if I could have free delivery and also maybe return for free as well. If we have some of those factors that does get consumers over the hump in terms of the concerns on the left-hand side.

Now let’s talk about some of these examples a little bit deeper, right? This one’s evolution three. This one I think is starting to emerge, but it could go a lot further. Let’s talk about groceries. In groceries, obviously everyone buys, you go in person, you buy it online, you buy Instacart. What if your agent is now able to shop across all different grocers? And maybe you’re shopping at Costco, you’re shopping at Kroger. What if you ask your agent, “Hey, this is what I bought last week.” It knows exactly what you purchased, how much, and what quantity. It asks you, “Do you want the same items again this week?” And as you talk to the agent, it says, “Hey, no, maybe I move the pasta, move the onions. I don’t need that this week. I already have enough.” It does all of that for you and gives you the final cart details and say, “Are you ready to purchase?”

I would argue if you have the right experience and the right comfort, you will say yes. The agent could go, do all the work for me, and I’ll give trust to the agent, and that’s true value add. And groceries is a great example, right? High familiarity, tend to be lower ticket items, and those products aren’t going to be like trying to choose between a couch. So this is a great example, but let’s go a little bit more advanced.

Okay. Groceries are simple, right? Buy milk, buy pasta. What about buying a car? Will I use agents to help me buy a car? I would argue yes, but it depends on the journey and there’s different value adds you can see. There’s two journeys here I’ll kind of talk through. The top one, journey one, I have an agent, my personal agent. I want to buy a new car. I want to get some details about whether it’s Volkswagens, BMW, Mercedes. Tell me about the models. Does it fit my lifestyle? Tell me all the details and the reviews. I want to have it all nicely put together and recommended based on what I need. The agent can do that. Obviously it could scrape websites, but what if the agent actually goes directly to the manufacturers, and they have agents as well that can give them all the details they need to actually make you make the right decision of which kind of car do I want to actually try.

So that’s journey one, right? Agent to agent there. Second journey. Okay. Now these are my top three cars. I actually want to go take it for a ride. Let’s see how it is on the road. And in journey two, we have you talk to your personal AI agent and say, “Hey, reach out to these car dealerships. Tell me what’s available for test drive, look at my calendar, and actually book that appointment and tell me when I can go to a test drive against those three cars.” Again, my personal agent, speaking to the dealer’s agents, doing all that detail—and if I think of the olden days, it’d be me calling up, not answering, leaving voicemails is a huge hassle. What if the agent could just do that for you, and suddenly it helps me pick the right car, helps me actually test out the car, and maybe it actually makes a big difference in terms of huge value purchase.

So this is a great one. I think if you see agent-to-agent interactions, it doesn’t have to be just micropayments. You can have agent-to-agent interactions across all use cases, and this is a great one I can say across all categories.

But that said, there’s a lot of focus on these micropayments and microtransactions, and that is unlocking new commerce models and agent to agent. So we think of the biggest use cases, high frequency, low value, the lesson is sent often some of these transactions. Think of paying per website, think of paying for API calls, think of compute time. We see agents starting to transact between each other, and it’s phenomenal how that’s starting to grow in terms of where the volume’s going. But is this going to continue? I would argue yes. If you look on the right-hand side, there’s some key trends that are going to drive the growth. Let’s think of globally connected IoT devices. There are 21 billion IoT devices in the world today. That’s going to double by 2030. So suddenly these devices actually need to be online and they can have agents tied to them in terms of how they transact.

Second data point: think of global traffic on the internet. 51% of that is automated. It’s no longer humans only driving traffic. Over 50% is being driven by, ultimately could be, by agents. And you can actually see that as that volume starts to grow, that opens up new futures for agents in terms of transactions and for commerce. And then one more page, and then we’ll get into some of our polling results. We talk a lot about like what’s the experiences, what consumers will need to get comfortable. I would argue, as retailers and merchants, they build their personalized experiences, they’re embedding agents, and they’re going to add lots of value that’s really unique to their different products, customers, and services. Think of Walmart. Walmart now in the Walmart app, you can use Sparky. It’s helping you with product discovery and argue that Walmart is preparing for an AI-first future, where now I can shop with an agent to help me in that full Walmart ecosystem.

Let’s look at Lowe’s. Lowe’s has a Mylow assistant. Mylow assistant helps me not just for purchasing and discovery—it actually gives me step-by-step product guides to help me in terms of the purchases that I do. And so it’s really tied to the merchant, it’s tied to the experiences, and this is how we think that merchants will win in the future.

Now we’re going to switch over. I’m looking forward to seeing some of the results from the crowd today. So I’m going to introduce my colleague Gaurav. He’s going to pull up, I think, the results, and then we’ll save some time for some Q&A and some closing thoughts. All right. So our first question, “Which purchase category would you trust an AI agent to complete entirely on your behalf?” Okay, so grocery is 41%, travel is only 12. Okay. So grocery, food or beverage. So if you’re shopping online, if you’re using Instacart, if you’re shopping in Costco, looks like groceries are going to be the one, that use case we talked through. But I argue that this is just an AI agent fully autonomously. There’s all different evolutions of how you can get agentic commerce to support the consumer.

Let’s go to the second question. “What’s the number one barrier preventing you from an agent transacting?” Worry the agent will get it wrong. Is that right? 32%. So mistakes is a concern. Don’t want to give up control, 20%. So some of the factors we talked about in our consumer research, it applies obviously for Sessions crowd here as well today. There’s going to be a number of factors, but I think as you design experiences, you need to address how customers feel, because ultimately they will not trust an agent unless you give them comfort that you have the right controls in place to support them in that journey. With that, let’s go to some of the Q&A. I think, Gaurav, we should have maybe a few questions potentially.

GAURAV MALHOTRA: Yes. We do have a couple of questions.

ALEXANDER PADDINGTON: All right. First question. “Do you expect agentic commerce be purely a mix of ecommerce or accelerated ecommerce volumes as a whole?” It’s a great question. I think ecommerce has huge tailwinds and has also been growing very strongly, but I would argue that agents can also potentially assist you even in the offline world, right? As you think about that car discovery process, I still want to go try the car in person. I still want to actually do the research maybe online, but then actually go in person to a store. So I do think that ecommerce will continue to grow with all the tailwinds. If anything, it might make it easier for consumers to do more and more online, but it’ll still be in-person purchasing that doesn’t go away.

Okay. Here’s another question. Two likes. “Do you think that legacy companies should lead AI agents adoption or should be made by another type of actor?” That’s a heavy question. So I would argue that we have to look at this by product category. We have to look at in terms of what is actually being done, whether it’s through LLMs or through the merchant. If it’s something like groceries—high repeatability, low tickets—like maybe that one is going to be one that can actually go faster with companies like ChatGPT and Claude. But I would argue that merchant agents have a real role to play, right? We talk about Lowe’s, we talk about Instacart. They’re building agents in their ecosystem, so I think that’s going to proliferate across the industry. And the question is, how do you differentiate your agent versus your competitors?

All right, here’s a third one. How long BCG signals agentic commerce beyond fixed… consulting.” Okay. I like that, into consulting. Oh, this is largely limited to the transactional use cases. It’s a great question. I would say that agentic commerce, in terms of high value in services and some of the more traditional ways of commerce, I see less of agentic commerce driving that, but I do think it’s really just changing the way you work and the way you actually think about output. This is an interesting question. How is this going to impact even consultants? You could argue maybe you might need less consulting and help in the future. But I would argue with agents, with better intelligence, we can get answers faster, higher quality, and better value output with assistance of agents. So argue, I think we’ll have to see how that actually transpires.

“Do you recommend companies to get their brands cited on ChatGPT?” Okay. I would say that what’s critical, and we’ll get to it in a couple of minutes, it is critical that you are able to have your products and services discoverable across all channels. It could be ChatGPT, Claude, Gemini, it could be for your own agent interfaces. I think you need to be building capabilities and experiences that will allow you to open up your aperture to all consumers based on how they want to transact. It’s hard to predict the future, but I think ultimately, hopefully, this gives you a sense of where things might be heading.

So we covered a lot. Thank you for the polling results. Hopefully that was fun. There’s four things I wanted to close as we think about today. The first thing: we think consumers are ready for agentic commerce, but trust is fragile. And I don’t think the unlock says I need the best AI model in the world. Yes, better AI is helping and it’s making a big difference. Everyone loves Claude, for example, but I think it comes down to better payment infrastructure, better fraud controls, better authorization logic and tokenization. Maybe my credit card details aren’t being given to the merchant. I’m actually just given a token that the agent can use. I think the industry needs a safe autonomy layer to actually then adopt agentic commerce. So that’s the first one.

The second is we talked about, okay, products and services, the discovery is moving to AI. You need to make sure your data is actually going to be machine-readable. So you need to change in terms of how you really open that up. So no matter how consumers want to shop, it is exposable to agents and LLMs in the future. Third, there’s a bit of a reality check. Agentic commerce, there’s lots of multimodel architecture. You need to really have openness in terms of standards and protocols you adopt, but you have to do that without slowing down your velocity of engineering and development. So how do you make sure you future-proof and open up across all the standards and protocols that have been rolled out? And the fourth and final thought for you: agentic commerce is not going to just happen in one big wave and one tsunami. I would argue it’s category-specific and so should be your strategy. Thank you all.