An interview with one of my favorite e-commerce analysts, Part 2

Good morning.

Today’s my birthday and I’m feeling major gratitude, including for this community right here. I’m grateful for all of you who have jumped on this ride with me over the last three-plus months and I’m as invigorated in my work as I’ve been at any point in my career. So thank you.

If you’ve been on the fence about upgrading to a paid subscription for full access to my articles, my archive, and other perks, today feels like a pretty good day.

In AI commerce news, The Information reported late last night that OpenAI is already pulling back from one of its biggest AI shopping launches: the checkout feature within ChatGPT chats related to shopping. The company had onboarded products from Walmart, Etsy, and some Shopify merchants, but ran into issues involving product data and more.

“We’re evolving how we approach commerce in ChatGPT,” an OpenAI spokesperson said in a statement. “We’re prioritizing making ChatGPT search and product discovery great, with the Agentic Commerce Protocol serving as the infrastructure that connects users to merchants… Instant Checkout is moving to Apps, where purchases can happen more seamlessly.”

I’m skeptical that the average person wants to shop via these specialized ChatGPT apps created by retailers like Target that live within ChatGPT, versus just shopping in Target’s app. But it’s early days.

This means if you want to purchase something you’re researching in the generic chat experience, you’ll be redirected to the retailer or brand website to complete the purchase.

Is this the failed social media “Buy button” era all over again? And should Google and Shopify now feel more or less confident about their efforts with the Universal Commerce Protocol meant to build checkout capabilities into Gemini and Google AI Mode? Let me know what you think.

Now on to the good stuff…

Yesterday I published Part 1 of my conversation with Juozas (Joe) Kaziukėnas, the founder of Marketplace Pulse and one of the sharpest independent observers of how AI could reshape commerce. 

In Part 2, he argues the industry may be obsessing over checkout integrations and protocols while overlooking a more basic question: what shoppers actually want from AI tools. 

We also discuss why TikTok Shop and Whatnot deserve more attention today, the early economics of ChatGPT ads, and why the most interesting action in agentic commerce might be happening quietly on GitHub.

Plus, how to enable some ChatGPT features that haven’t fully rolled out yet. Enjoy.

You’ve posted about Alibaba’s AI shopping features and how the company is actually putting marketing behind them unlike the big AI platforms here.

Juozas Kaziukenas: Yes, but the experiments going on in China are very similar to what we have here as well. They haven’t found a different take. It’s really the same thing, just easier integrations because Alibaba owns the whole stack. So it’s much easier to roll it out.

What are we not focusing on in this space that’s actually real today?

JK: I think for us, meaning the Western world or the U.S., we always make the same mistake. We continue to talk about technological capabilities, and we miss what users actually want.

It’s been very messy to get any sort of data about how user behavior is evolving. How many people are using ChatGPT? How are they using it? Are they asking shopping questions? Are they interested in making transactions on these platforms or tools?

So much of the conversation seems to be about payment integrations and checkout and protocols. But ultimately, to quote Steve Jobs, you have to start with user behavior and work backwards to the technology, not start with the technology.

Just today I saw a post that Meta AI might be adding some shopping integration. And I’m like, yeah, but it’s now the same thing everyone else has. None of this is new.

When Google rolled out UCP, obviously it’s a completely different technology and it’s now on a massive platform that is Google. But ultimately it’s the same blocks of products we had already seen on GPT a few months before that.

So when Copilot adds it, when in the future Claude adds it as well, this isn’t new. This behavior and this type of integration already exists at a surface level. What’s ultimately much more important is how people are using these tools.

To me, the actual shopping functionality of these chatbots has not evolved as quickly as the integrations and the payments and everything else. I feel like we’re still missing a lot of UI and UX elements that would actually make the experience delightful and mind-blowingly good, where you would choose to use it, as opposed to something you do accidentally while using ChatGPT.

So I think as an industry we’re spending way too much time on integrations and not enough time trying to think of ways to make this an incrementally better shopping experience for many people.

Are you confident or optimistic that any of the LLM companies, I’ll include Google in that, are going to make the experience so enjoyable that people want to transact there?

JK: They have nailed the product-research type of shopping. When you need to research things, ask follow-up questions, dig into features, and compare products, that works well.

For example, on Amazon you can ask Rufus questions about products, and that may be a more native experience than clicking through reviews or browsing descriptions.

But there are many other types of shopping that are much more focused on discovery, impulse, and visuals. That is severely lacking from the current AI chatbot experience.

So I think it’s just very early days.

Many of the use cases people talk about are still very fantastical. When people talk about the ultimate agentic commerce experience, where a system checks out completely on its own by its own decision, it’s not even clear that demand exists.

But in the meantime, we can make the product-research shopping experience on these tools better. And through checkout integrations there will probably be an incremental source of transactions.

This weekend I read Derek Thompson’s post, “Nobody Knows Anything,” critiquing that viral AI doomsday Substack post. And I feel like it could also critique a lot of the commentary around agentic commerce.

We’re discussing science-fiction ideas about what it could be or what it should be, and saying maybe by 2030 it will already be 20% of e-commerce. But we’re making wild guesses based on no data and no meaningful change in user behavior.

But there’s still a lot of excitement, no?

JK: I try to hold two thoughts together. I cannot remember a time in online retail when there was so much excitement about something that could be done in the future.

Even brands that see no incremental revenue from these channels are still deploying resources and people to solve for it. That has never happened before.

That excitement is what’s needed to actually make change. That’s why I remain bullish and wary at the same time.

What else is on your radar right now?

JK: I check GitHub every week to see what’s happening with the Agentic Commerce Protocol and the Universal Commerce Protocol. They’re getting comments every day from different companies for features that are not live anywhere yet.

That’s interesting because even though it’s not live anywhere, the industry is pushing it along as something meaningful. That hasn’t happened in online retail before.

For years, when I talked about commerce, I would open by saying Amazon has basically remained the same for 30 years. It’s a search box with product results, and it’s remained that way.

So this is a unique moment where a new technology has aligned everyone to experiment with the same thing. Google and Shopify announcing things that aren’t even built yet. That hasn’t happened before.

It’s kind of neat and kind of exciting for the industry.

For things like blockchain, the metaverse, or voice commerce, they were niche ideas some companies were excited about. This one has everyone excited.

That’s both a sign of a bubble, but also an environment that can be productive.

So where should retailers and brands be spending more time outside of this area?

JK: Should people put more attention on TikTok Shop? Should people pay more attention to Whatnot?

Absolutely. These are massive revenue-driving new takes on commerce that exist today. We don’t have to debate what TikTok Shop is. It exists and it works.

Yet we’re much more excited about agentic commerce even though we can’t agree on what it means. For most companies that are not Amazon or Walmart, my advice is that the AI boom enables you to do the things you’re already doing faster, cheaper, and easier.

AI in the short term should be about improving supply chains, optimizing ads, and reducing operational overhead. Those applications of AI can help businesses today. Then leave it to the Walmarts to figure out how chatbot shopping interfaces will work.

If you’re a small brand, this is not a strategic decision you need to make today. You’re not going to miss out by not participating.

Sure, you can check whether you appear in ChatGPT responses. But you shouldn’t do that while ignoring your performance on Google. This shouldn’t become the new north star that makes you ignore everything else.

What do you think about ChatGPT introducing ads and the form they took?

JK: So I’m not surprised that it is fairly basic. I’m not surprised that it doesn’t have meaningful tracking, or seemingly even good targeting, because they obviously had to start from somewhere.

But they seemingly have put themselves into this impossible puzzle of trying to introduce ads while also trying to maintain the neutrality and organic usability of ChatGPT as a product. And thus, the position they put these ads in feels like it’s not aggressive enough for these ads to be performative, or performing well.

So I think we’ll see the performance of these ads and how well they work. But over time they will have to get more aggressive in where these ads are positioned and how they look, for it to actually be a meaningful revenue driver for ChatGPT.

Because as this very basic unit at the bottom of the response, it feels invisible to me as a ChatGPT user. It’s hard to measure. It’s hard to imagine it performs well, given the high price companies are seemingly paying for it.

But many people have argued that for a mass-appeal product like ChatGPT, with a billion users, the only way to monetize at scale is advertising. So advertising was inevitable. And this will evolve, in targeting, in how it looks, and what it can do, over time.

I just think a company like ChatGPT is in quite the pickle, because it’s competing against companies like Google that can avoid putting ads in Gemini for as long as they want, because they have all these other revenue sources.

So I think the future of OpenAI is still undecided. They’re trying to please so many different sides: users, advertisers, and their vision of what ChatGPT is and should be. They’re facing a lot of difficult challenges.

My last one…there’s been a couple of cases where you’ve gotten ChatGPT to show features that were not yet public and posted about that. What’s the trick?

JK: What ChatGPT is doing, which is not that unique, and actually not unique for their products, is that they test different features for different users. I think OpenAI made an acquisition in this space sometime in the past, but it specifically enables targeted A/B testing of features.

But you can just enable them yourself. We knew from the ChatGPT checkout announcement that cart functionality was coming because they mentioned that in the announcement. And most browsers, like Chrome, have developer tools. You can search the source of the app or website. You can search for “cart” and see if it comes up anywhere.

If it does, you can perhaps find a feature flag that enables it. And you enable it, and it starts to work. That’s just a byproduct of how they developed it.

And this doesn’t allow you to access features that are restricted on the back end of OpenAI. For example, I can make it show the checkout UI, I can make it show the cart page, and you can add products to the cart and remove products from the cart. But I cannot check out on the cart because the checkout feature is still restricted on the back end, and you cannot impact that or change that.

But many features that are coming, or going to be released in the future, even future models, eventually appear earlier. Days earlier, or weeks earlier, in the code of ChatGPT. Most websites are somewhat public.

When ChatGPT rolled out their instant checkout integration, I believe in November of last year, I remember seeing in the source code that they had Shopify Shop Pay support in there. But it was never live.

And just today I was like, oh, I wonder if that ever went live. I don’t think it ever did.

But you can go to ChatGPT and add one little line of code, and all of a sudden Shop Pay becomes available. 

I still have not seen any ChatGPT ads myself, even though I keep trying. But you could inject fake ads into the front end of ChatGPT and it will show you how that looks.

For example, when they launched ads, I was like, is it ever going to have buy buttons? Is the ad integrated with instant checkout, or is it always going to link out?

And for now, I verified this: it can only link out. The ads action is always a link to somewhere. It has no integration, at least for now, with instant checkout or any other features of ChatGPT. And this is not speculation. You can just see that in the front end code of ChatGPT.

Ultimately what people are trying to do is build an understanding of how these systems behave. Because when you interact with them as a user, they can feel automagical, like they’re inventing things out of thin air.

But ultimately they’re not that magical. They often rely on web search and combining things into results. The more you dig into it and understand, the less magical they feel.

Thoughts or tips on the topics in this newsletter? Just reply to this email or message me at [email protected] or confidentially on Signal.

Feel free to forward this email to someone who might find it useful. And if someone forwarded this email to you, please consider subscribing to The Aisle here.