AI Expectations: Hype, Reality, and the Road Ahead

I think the market might treat it like a bubble and it may fluctuate up and down like a bubble. And there's this quote and I don't know who said it but the idea is that we always um overestimate the effects of technology in the short term but then underestimate the effects of technology in the long term. Um think about when uh the internet came out, right? So like you could call the the the.com bubble a bubble, but now what are the most valuable companies on the planet? Is there internet companies or companies that are a part of the internet? And so I do think that AI is here to stay for sure. I think of one reason why people think it's a bubble is it's the same mentality that you might have if let's say like let's say your kid has a friend who's an idiot and you know that they're an idiot and then 20 years goes by and you go into your doctor's office and now that kid is a doctor like any rational person would recognize that that that child that teenager or whatever has grown and learned over time. Imagine a world where like that idiot kid was like I'm going to grow up to be a doctor and you're like you're not ever going to progress. That's kind of the mentality of these AI bubble folks is they look at what AI can and can't do now and they just presume that it's going to stick to its same capability in the long term. But if AI is growing and learning, then you should look at how humans grow and learn. And hopefully, you know, if it was to save your life, you'd trust that doctor, right? There are some credentials. Now, of course, what that means at the pattern side is like testing everything uh the same way you would test the doctor. So, I don't really think AI is going to be a bubble. There are some things I don't know if people are going to shop with AI or not. I don't I think it's going to be like I think it's going to be like a self-driving car. Every once in a while you might get into that futuristic Tesla or BMW that's self-driving and you might want it to drive for you, but every once in a while you might choose to drive. You might uh the the AI might choose the most efficient route between point A and point B, but every once in a while you might choose the scenic route. So I kind of think of it more like uh something that's learning that you as the person who owns the car should always have the opportunity to get behind the wheel and also like it's going to make mistakes but the rate of mistakes should be much lower than what humans do. So I think a lot of people are trying to figure out what a good AI experience looks like. In fact, you can go and see interviews from the head of product at chat GPT and they're like, "We never thought it was going to be a chatbot." Like, "We always thought it was going to be something else." And the reason why is because people don't love chat experiences. Imagine if you had to interact with like if you had a a cleaner come and clean your house and if you had to interact with them via chatbot and you're like, "All right, do this room now. Do that room now. Do this room." Like, that's not a good experience. You don't want that. Um, and so at Pattern, we've been really thinking a lot about what does this great AI experience look like for our brand partners and the people who want to work with pattern. And what we realized is that there's thousands of years of history behind what makes a great humanto human relationship. And if AI is to model human intelligence, actually the best humanto AI experience looks a lot like the best humanto human experience. So for example, why do people work with pattern today? Well, they love our data. They love uh our automation. And then they also love that whenever they need to or want to, they can talk to one of our pattern e-commerce uh brand managers. Well, the idea here is that the reason why they talk to those brand managers, as nice as they are, is because they're informed with our processes and data as well. Uh that's why they're getting on the phone with them sometimes, but also it's why they're getting into predict our uh dashboarding software because they want to see how their business is going. And so as we think about AI and uh human interactions, we think a lot about what is the best pattern brand partner experience to date. And what's crazy about that is like sometimes you show up during like a weekly call and there's a deck and you're like skip to the end, tell me the good part. And sometimes you're like, give me a five-page paper to explain this solution. And so the the the weird thing is is like this flexibility. So think of it as like cognitive load. So, if I was your employee and I said, "Hey, boss, we need to go ahead and spend this much money on advertising uh next month, okay, and here are the four or five reasons I've written you a 40page paper. I need you to read it." Well, I just made your cognitive load harder, right? If I were to say, "Hey boss, we're out of stock." And then just sit there, right? If you're like, "Well, what are we going to do about it?" So, if I go, "Hey, boss, we're out of stock. Here are three options to fix it." Versus, "Hey boss, we're out of stock. here are three options to fix it. I think we should go with option one. By the way, I' I've already done option one and I'm going to make sure this doesn't happen again. And so, as you think about what makes you as a human, you as a manager have the uh best experience, that's really how we're trying to model AI is to make it where the cognitive load doesn't all of a sudden become on you. And this is the problem people always thinks about dashboards. They're like, "Oh, I want a dashboard for everything." But the truth is is like you don't. Most people don't. They they don't want the AI writing a 50-page PDF that they have to answer for everything. They want a more human-toum experience. And that's what we've got to model. So people look at our financials and they're like, "You guys buy inventory?" Like, "Do you really use AI?" And it's like, it's kind of a funny question for us internally because not only do we use AI, but we think we're pioneering some of the best practices when it comes to AI. Um, it's possible that we get some things wrong, but I mean we have our team, some of our team members, we tell them that they need to be on Twitter like an hour a day because what's cool on Twitter is that you're getting new updates every single day uh in terms of what people are learning u how they're changing how they work with AI. In fact, I saw one I saw one professor of business management say that they noticed that managing AI a lot of the same management principles that they try to teach their MBA students apply in terms of like managing it the right way. So anyway, at pattern I think we're I think we're using AI very very well. Now what exactly does that mean? So like let's kind of break AI into a couple of things. So, um, at its very core, AI is a form of a machine learning algorithm that predicts the next token in a sentence. So, if I say, "My daughter is allergic to peanuts. It's too bad because kids her age usually love to eat peanut butter." And you can kind of guess the next word, jelly sandwiches, or something like that. So, AI is predicting the next word. Well, that's machine learning. And for years, we've been using machine learning for things like advertising. So if I take this product and I advertise it on this search term and I'm getting this cost per click, what is my ROI and which search term should I advertise on? So we've been doing that for years. We call that destiny. Uh the whole idea being is we forecast or we predict uh what your uh product's destiny should be. How good should it be performing on any given search term? And we spend to let your product get there. Um we also use it just for forecasting in general. So, we do buy inventory. Um, and part of buying inventory means that we need to forecast uh supply and demand. Uh, we need to forecast lead times. We need to make sure we order appropriately. And you know, if last year, this week, Kim Kardashian tweeted about one of our products, that means that we have to recognize that that was a blip that may not repeat this year unless she's AI, I don't know. But the idea being that like things happen and you want to go ahead and forecast appropriately. We also use AI today to um model competitors. So, in the same way that uh you were able to guess uh the next token in a sentence, like my daughter loves to eat peanut butter and jelly sandwiches, right? Uh we're actually modeling what our competitor's performance is. And that's really uh necessary if you want to do a SWAT analysis. If you want to look at strength, weaknesses, opportunities, and threats, the idea is you need to look at what your competitors are doing and how they're performing. And that's how we help our partners uh use AI to see what's going on in the market. And then also we've been using AI for things like uh causal analysis. So think about correlation versus causation. Okay. So if I'm trying to look at the shoes category, um is it the brand name being Air Jordan what moves the needle or is the color scheme of your shoes being red and black? Like which of those are actually moving the needle? And so using AI to analyze that is really important. Now obviously over the last couple of years we've gotten what we can call like big AI. So think of uh Grock uh Anthropics Claude uh Chad GPT Gemini uh Meta Llama model all of that. So what have we been using those big AI for? Well, we're not naively saying that we're going to go train our own uh chat model today, right? We're what we actually do is we're agnostic. We use the biggest and the best models and all different ways depending on what they're better at. So the first thing we use big AI for is labeling data. Now, when I say labeling data, let's go back to that shoe example of the Air Jordans. Well, imagine I've got all of the products within the shoe category. Because of our own data, we've modeled how well each of those products are doing. So, then let's label things like what are the reviews saying? Um, do the images show people playing basketball or people looking cool at school? Is it showing male like women or men? Is it showing um uh people wearing these Air Jordans in an urban environment or a more rural environment? So the idea is that you can have AI go through your title, your bullets, your description, your image stack. This these big AI companies, you can use those to then create labels for your data, which you then go ahead and pair with our performance, which is how we start realizing, oh, like it turns out in shoes, if I'm talking about basketball shoes, it's not all about playing basketball, it's also about looking good in high school or something like that. So that is like labeling data. Now, another way we use big AI is people have been talking about using these chat GPT- like models to um model personas. And so, what the research is saying is that I can tell an AI, you are a soccer mom in her 40s with three kids who lives in Oregon. Here are three products. What do you love and hate about each product? And it does a pretty decent job. Well, you still just kind of made up the anecdote of a soccer mom in her 40s. So what if what you do then is you take the concept of using AI to see how uh these different personas would react to your products and then you pair that with our demographic and psychographic data. So like for our products we know the demographics of the customers on Amazon. We have a good idea on if she really is a soccer mom or if she's a hockey mom. And so the idea here is that if you pair that labeled data with the pattern data then you can create these real personas. Now obviously we also use AI to write code uh with various levels of of success and I think the market as a whole is realizing that AI is good at some things it's not good at other things. Um we have not yet found that AI is good at architecting code yet but if you're very clear with instructions it can do a lot of really great things and honestly that's changing every week. So if like if you watch this video in January or February of 2026 versus in April or May like what AI can do with code might be completely different. Um, we also do AI for just like the I mean the standard stuff, responding to people, customer reviews. We try to keep it as real as possible. Um, but uh, you know, you can think of anything chatbot related. But one of the other things we're doing with AI is we're capturing knowledge and ideulating. So this is what's really cool. If I were to go to chat GPT and I were to say, give me a step-by-step guide for creating a brand strategy. Well, you go ask that question seven times, you're going to get seven different answers. you might get like six different answers or five different answers. There's some randomness in the output. So at pattern, we've got a whole bunch of experts and these experts, we have information that's not yet on the internet, hasn't been trained into these large language models as a whole. So what we also do with AI is we capture information. So imagine me explaining a step-by-step process as well as four or five of my peers who have slightly different takes on these processes, transcribing that and then putting that into a knowledge management system. And then when somebody who uses our software says, "Hey, pattern intelligence, I want you to go ahead and uh help me create an image generation strategy." Well, we can reference our frameworks and our data and the biggest and our persona information and the the biggest and the best AI models to create a strategy. And then I'd say like last way we're using big AI, maybe last is not the right word, is ideiation. So, what's really funny is that while we've been capturing some of these like step-by-step processes, uh, let's say I say, uh, I'm capturing a step-by-step process for determining if a product I sell in the US can legally be sold in Canada. Okay, that's a step-by-step process. Well, what I might do is go interview the internal expert who understands that. And then I might go, "Hey, chat GPT, how would you determine if a pro if a product is able to be sold in Canada from the US?" And it might be that like there the what my guy has said or what my gal has said and what chat GBT has said is like a 90% overlap. But sometimes the AI gives us a better answer. So in addition to ideulating, we're improving our own processes so that we get the best of AI. There's this joke where AI is good at everything you're not good at, but the second you start interacting with something you're an expert in, it has no idea what it's doing. Well, what we're doing is we're making sure that where the answers are better than us, which in some cases they are. It'd be naive to think otherwise. We use the real the better answer. And in some cases where we're better, we store that knowledge. Cool. You combine all of that. You can write copy. You can create image strategies. You can do pretty much you can run an entire e-commerce business across the globe.

What should brands really expect from AI? In this video, we sit down with our VP of AI Transformation to discuss how the market views AI, “bubble” debates, and practical examples of how Pattern uses AI today. Learn about the real impact of AI on e-commerce, how our teams blend technology with human expertise, and why flexibility and smart adoption matter for business success.