AI vs Human Instinct: Who Should Decide What Gets Made? | Greg Petro

Introducing to the stage Greg Petro, CEO of First Insight. Thank you. Okay, um, I'm going to try to make this as interesting as I possibly can as far as presentations go. Uh, the first question I'll ask you is a simple one. Do you want me to give you a presentation that is kind of interesting, that you go away with and say, "Oh, that was interesting," or do you want it to be one of the most thought-provoking presentations you've had recently? You get to vote. Number 1, for those number 1, raise your hand. Number 2? Okay, good. Good, good choice. I told my team, I said, we were talking, I said, I'm feeling a little ornery this morning. So, I'm going to share with you some thoughts. First of all, I'm going to give you a concept about AI. Before I even do that, when do you think the first AI— oh, I got an air conditioning vent right here. When do you think the first AI was invented or used? 1950s? 1990s? Over 2000. Those in the '50s, raise your hand. Those in the '90s, raise your hand. 2000. 1956. First AI. It's a long time ago. In my career, I used AI in 1999 with neural networks. I won't bore you with that detail, but I was also a merchant, so it's kind of weird. Here's the interesting thing before I get into the whole AI piece and the concept of humans with computers. That term is called human computational modeling. Humans and computers, human computational modeling. As I'm driving over here in my car, I'm looking out in Manhattan, Midtown, and I'm looking at people. I'm an observer of people, and I'm noticing, just like I am here in the audience, How many of you are wearing the same exact thing? Very few. How many of you carry the same bag? Same pair of shoes? If anything, one of the biggest trends in our industry, meaning retail, brands, manufacturers, that has come about recently is micro-segmentation. Meaning people have, because of information and their own personal preferences, fragmented themselves into different things. In the '80s, when I was in retail, you could dictate trends, you could determine things, you could make a lot of the product, and there was demand for it. That has changed dramatically. In fact, Silicon Valley is raising money based upon companies that are looking at markets and figuring out how many people are within the market, It's individual-based. Why is that important to people in development and design of products and manufacturing is because you can create whatever you want and people are going to decide like they always have. The issue is, do you have a large enough market? So when we talk about how AI is used in retail, in manufacturing, in design, I want you to think of of AI as a weapon, okay? It is a weapon. It will be used with you, against you, or by you. But make no mistake, it is a weapon. And I'm gonna give you examples of where you can compete with it on a human level and where you're incapable of competing with AI. Okay? I don't care who you are on a human level, you're not capable of competing. So where can you win and where can you lose against this weapon? And how can you use it as a competitive advantage as opposed to something that's going to be used against you? So here's the other piece about the current environment now that we frame the situation and the problem. Consumer preferences not only are fragmented, they change constantly. What I like today, I don't like tomorrow. Sustainability is an interesting thing. We have the largest data set around sustainability, but consumers don't want to pay for it. So the— they want it, but they don't want to pay for it. So that becomes a cost to you, and it also comes at the cost of some of the challenges that you face related to what they're willing to spend. Supply chain disruptions, we're seeing that now with the tariffs. We saw it with COVID It's only going to increase rapidity of the supply chain. It's a very fragile ecosystem that you have to manage on an ongoing basis. It can be as simple as a plane, a boat like they had in Panama that got stuck, if you remember that event, right? Shutting things off. Sorry, I didn't shut my phone off. Increased competition. The barriers to entry right now from every angle are very low. And then economic uncertainty. Who would have thought the stock market's up to nearly a record high in the US? And then the last piece is technological advancements. I have in my family every generation except for Alpha, the newer one, and all of the children say they can't keep up with tech, and they're all in tech. And the reality is, as fast as you come up with it, it's beyond you. So where do you set your sights, and how should you think about things? At the same time, much hasn't changed about retail. We used to have the 80/20 rule, or the Pareto rule, right? 20% of your products give you 80% of your sales and margin. So you're in the business of trying to understand consumers, and the question that I would ask is, how often do you engage with them? That's the question that you need to ask yourself, even if you're in the design room. How often are you actually engaging consumers to understand what they want? What if you could poll them, which is why I polled you? I'm going to start there in a minute. If you could only have one piece of data that you had to make all your decisions off of— and this is not a loaded question, it's not predetermined, it's— let's just be honest. What would you pick right now? Would you pick historical data, trend data, competitive data, consumer data, or AI data? So let's do a show of hands. I could tell you who— I've done this presentation before and I can tell you what the outcome is. Anybody want historical sales data? A for your information, if you can only have one? Trend data? Hmm, maybe. Come on, you think you all want A. Competitive data? You do want to know what your competition is doing. Consumer data? Consumer data, okay. And then AI data. Yeah, well, you don't— do we know what AI data is? Come on. Okay. All right. I want it, but I'm not sure what it is and how it works. We'll talk about that in a minute. Well, AI transforms retail data. Here's what I would tell you. There are different types of AI. So if you take nothing away from this conversation today, understand that all AI is not the same, and I'll explain it. There's only 3 types. Very simple, very direct. It can help you do forecasting. AI can do that. It can help you prescribe actions, and it can also identify things. So summarizing consumer consent, sentiment, like, hey, what do they think about the product? We've all used that. You can go on Amazon, they have Rufus. Here's commentaries or what AI does, or Google does it. Refined designs, because it can actually look at your product and understand it and then suggest changes. And then, marketing assets. Let's say you're a design house and you didn't want input from consumers. You understand the consumer, you feel. Okay, my suggestion to you would be, Engage the marketing side of the equation because at least it can help you figure out what to sell and who to sell it to. So the 3 types of AI are right here: predictive AI, generative AI, and agentic AI. And you could read for yourself what they do and how they do it. Here's the first thing I will tell you. AI is far more powerful on a computational level than a human. There— it is not humanly possible for you to take all the data streams that you have and all the information and come to a conclusion better, faster, or more accurately than AI. It's statistically improbable, no matter how smart you are. I mean, you gotta let that sink in. I'll give you just a great example of a product developer. He created a product, this one right here, and when he did, Steven Jobs said, "No one's gonna want a camera on a phone." Look it up. Well, that was not correct. Was he a bad designer? Obviously not. The question is, you'd say, well, maybe consumers didn't know that they wanted a camera. Well, maybe they did. The question is, when you're looking at decisions that you're trying to make, you need to think very broad, and then you need to use technology or capabilities to narrow that decision in. Think of it as just simply guardrails to making bad decisions. Missing something. So the way we think of collection of data and AI, and we've been doing this for 17 years— AI native, I call ourselves— is collecting information from consumers and then modeling AI on top of it and giving you something called a value score that ranges between 1 and 10. And you're going to hear a little video about what it does. Not a pitch. I'm not here to share with you how and what we do. You're welcome to go to our website and learn more, but I will give you an example of use cases of how this works so that you can understand how companies around the world are using it. So this is a video, and they told me if I click it, it'll play.

So what exactly is a value score? A value score is a predictive indicator of buying intent. If the store If your Value Score is low, your product or messaging isn't connecting with customers. If it's 4 to 6, customers see some worth, but there's still a disconnect between how you value something and their perception of that value. But if it's a high score, that means you're speaking their language. The best part? Anything can have a Value Score. From car displays to water bottles, from women's dresses to Super Bowl commercials. Just imagine the art of the possible. Value Scores are a pulse on what consumers really want. Agentic AI with human insights. That's the value of First Insight.

Okay, so after you have your eardrums burst and we got your attention, you're wide awake. You have this thing called a value score. Now, uh, have any of you seen the new Meta sunglasses with Ray-Ban collaboration? I want you to picture that, and as you walk around the world, everything has a value score. Our team members, I tell them, when you go into a meeting, you get a value score. And a couple are in the audience today, they're going to give me a value score of how I did this presentation between 1 and 10. It's a feedback tool. It's an AI feedback tool that gives everything a score. And it does it before you bring the products to market. It does it before you set the price on a product. It does it before before you place the ad. So you heard an example. We have companies in the world that are testing Super Bowl ads and trying to determine which one's gonna resonate with consumers. Why? It's a multimillion-dollar decision before they do it. Same thing with product design. Same thing with pricing models. Same thing with what you're offering up to anyone. Essentially, we're gonna place a value score on everything, and the value score changes. So I had a bottle of water downstairs where it was nice and cool, wasn't worth that much, or a glass of water up here right now, probably worth a little bit more to me. The value score went up because it's hot. If I had stands or lights, it would be a lot more. Think about it as it also changes when you walk into a store or a location, you're trying to buy something, and whether or not they want it changes. How you position it, how you advertise it. That's the concept of a value score developed off a series of inputs from consumers modeled in a way with AI that makes it very simple to use and understand, at the same time very powerful. So it applies to all these areas right now around the globe. So we do about 25% of the global retail market. In 4 continents, the largest retailers, brands, and manufacturers around the world, right? 67 language currency combinations, 180 countries. So you think about a company right now that's that large, and we're working with them, and in 24 hours we're running their entire product line and giving them feedback about what consumers are suggesting is going to sell and what isn't. And what are they doing with that? They're dropping products. They're narrowing their assortments. They're doing SKU rationalizations. Right now, today, we're doing pricing tests for tariffs around the globe in multiple currencies and then giving them a global view of how they ought to change their brand image related to pricing or product segmentation. We're also doing ad testing, influencer testing, marketing segmentation information so that they can understand which products that they are. Here's an example of how we would break down the Starbucks offer for a segment of consumers and what they're saying about it, what new categories they're going to go into, how they should price the product, what they should feature on the cup, all the attributes that you're developing and designing in those products. Here's a wheel design where we're taking a Volkswagen wheel, and you would think to yourself, "Hey, that's interesting." And one of our employees says, "Well, I don't really care about the wheel." But in fact, it matters, just like the stitching on the blouse, or a handbag design, or whatever you're doing. It's no different. We're in fashion and apparel. We do it for furniture. We do couture designs. Think about that for a minute, right? Fashion apparel couture designs, high end. Hmm. This is an example of generative AI that actually allows you to pick the tone of how you want the generative AI to describe the product online that's SEO generated to give you the highest conversion rate. It has proven itself to be better than any other copywriter. So if you're in that role, it is likely this is going to help enable you to become more strategic, or— and if you don't engage it as a competitive weapon for yourself, it's going to replace you. I hate to tell you, that's the reality of the situation. Not in every case. I speak about the topic quite often. AI is not here to replace humans. It's here to enable you to become more strategic at a scale level that you've never had before. On the design side, there's lots of recommendations that we make, but here's an example of— AI taking a bunch of competitive analysis and coming up with a new idea all by itself. And then it can write the copy and script for it. It could look at the production for it and the pricing model for it and make the recommendation of who you should target it. Marks Spencer. Anybody familiar with that company? Uh, they're based in the UK, uh, one of the largest companies that do, you know, men's, women's, kids, grocery. This is public statements that they made for us in December. We've been helping them in their turnaround process, going quite successfully in the products that they're bringing to market. So when you think about, hey, how AI shifts retail decision-making, not here to be provocative in suggesting you should be worried. I'm here to enable you to think about how you can leverage this. Smarter assortment planning, smarter pricing strategies, increased conversion rates of offers that you're going to make, and then revenue and profits. I was talking to somebody earlier downstairs and I said to them, even if I believe that AI could develop products I'm not sure I support that. Designers are wonderful people with very creative imaginations, including myself. That's one thing humans have. But every product we bring to market isn't a perfect success. What AI can do is help you figure out which ones to invest in and which ones not so that you can be more successful on a financial side. Or even if you're stretching it, on the actually very provocative designs, where to take that, but then buy less, make less, so that when you make that manufacturing decision, you're not overinvested and then having to do all the markdowns. And humans and computers have been working together for a very long time. Who doesn't want collision avoidance in their car? If you've never had it, it's amazing. It just beeps, right, when something's around you, or a camera pops up and tells you what's behind you. Who doesn't want that capability? But you just want to feel in control. So when you think about AI, it's about understanding it and appreciating it more than it is fearing it. Okay, so that's it. I'm early. I got 13 minutes. I told the team I would be early. Any questions? Oh, there they are. Anonymous. Building trust in generative AI is a challenge, and so with Agentic AI. Share a unique approach you've taken to overcome organizational resistance. Great question. Wasn't my idea. Wasn't our team's idea. It was our customer's idea. About 9 years ago, we started with this company, and they had used our technology to put value scores on products all the way from the design process— a vertically integrated retailer— all the way from the design process to market. And what they were feeling was that resistance. So what they did was They would have town halls and they would feature a product that was called a Perfect 10. That's what they called it. I don't know why. Think about 1's the worst or the least performer, 10's the best performer. And they featured it and they made the merchants and the team members come up to the town hall and they featured the 10, talked about the success. They went to market, how successful the product was. And it reinforced that behavior. Actually, over a period of time, the individuals that had the most amount of 10s were getting their bonuses, and pretty much everybody wanted to do it. A very organic way to do that. That's the way our customers have approached that, is you've got to get, you have to build the momentum internally in the organization. It has to be organic. You can't force people to do something they don't want to do. So the way we featured it is once you just correlate success with something, people want more of it. And so that's what they did. Price elasticity is very dependent on consumer group, you question. How do you ensure consumer groups— great question— on a predictive model? Okay, so this is where AI is really, really interesting. By the way, humans can't process this much data. Like literally, this comes in a second. The model can produce these outcomes. The way I explain it is if you did a focus group or you asked a bunch of consumers what they thought, imagine, well, I'll give you a different example. We all have friends, hopefully, that we actually engage with, not on Facebook, in real life, where we talk to them, and let's say we want to go out to dinner. And there were 5 of us in the group, and we said, "Who— where do you want to go to dinner?" And we asked everybody where they wanted to go to dinner. And Greg— they get to Greg, and Greg's like, "I want to go here. I'm sure this is going to be a great place." Well, Greg never picks a great place, right? That's what happens when you survey people. Everybody gets an equal weight. The reality is they shouldn't listen to me. I'm not a good picker of restaurants. So when you do surveying of people, it's not a really good idea to count everybody equally. So the way we've used AI is a way of filtering and weighting people to those who understand your brand and your products 6, 12, 18 months in advance. And you'd say, how does it do it? And the answer is, I was a merchant. I bought tablecloths. My average customer at a department store 100 years ago, which is when I started in retail, like, I was demographically not her. But yet I was making that decision. The way I did it was I asked them, but I knew which ones to listen to more than others. So we've built AI models that actually filter and weight people based upon brands. And then what it does is it starts to accumulate enough responses relatively quickly, all digitally, and within 24 hours it can render you a price elasticity curve that has no history. And you'd say, well, that's great, except that we correlate it to sales data in every single item every single time after it comes to market. And it learns on itself. Not only that, at, at an item level, at an attribute level, but it also has demonstrated it to be more successful. Okay, so that's how it actually works. Causal AI is designed to identify cause and effect relationships rather than just correlations. And why do you use AI in the microfragmentation? Oh, interesting question again. Yeah. Hey, so I remember a famous study that was in convenience stores. Men— this was done by 7-Eleven— would go in the US, would go to a convenience store. And when they bought diapers, they were correlated to sell beer. 6 packs of beer. And they thought to themselves that was a causal event. Obviously guys that have baby wanna drink beer when they pick up diapers. Incorrect correlation. Some things can be correlated for different events for different reasons. Just because you have the most number of broken elbows in Minnesota in the month of December doesn't mean it's because of weather. It can be a whole host of other reasons, and you can be drawn down the wrong conclusion. So what will AI be able to do for you? So AI, if you put the right data streams in, can look at a series of events. Remember I said to you, you have to look very broadly. The government's been doing this for a number of years when they look at gaming theory, and AI is generally designed to be able to do that. So for us, here's what we do. We look at very broad, diverse groups in order to get input for companies. And you'd say, I wanna listen to my customer, and the answer is maybe not that's the best input. If you were going to design a women's dress, let me give you an example. For a 45-year-old woman with an income level of $200,000, Right? Who would you want to survey on an input of the designs that you have in order to give you an output related to success or failure? Theoretically, you'd start with 45-year-old women in this demographic group, blah, blah, blah. I would say I would want to talk to and find people like Giorgio Armani, Ralph Lauren, and a whole bunch of other people that demographically don't fit that model at all. Interesting, right? So you need to think about things differently about who you're talking to and why you're talking to them. Can segmentation done by assortment? Simple answer, yes, 100%. At a scale you can't even imagine. Uh, your tool impacts retail assortment process. Are you integrating with specific PT— all the vendors in PLM, all the planning vendors, and all the marketing. Think PT— not to name names, but just examples— PTC or whatever, O9, Oracle, and then Clavio and those folks. Okay. Can you give an example of agents you use successfully? Pricing for tariffs. That's running right now like constantly across the globe. It hasn't been settled yet. We just did a report. You could download the report if you want the aggregate report. Um, we don't share information by each company. Um, but we did our own survey around that and studies around that so that you could take a look at it for the industry. But that's where we do it. What are you investing in for 2030? Exactly what I said in the beginning. The vision that we have is that you'll be able to use AR or whatever device you have— glasses, phone, whatever— and you'll wand it over things. And everything will have a value score so that you can make decisions around that. OK, I got 5 minutes. Any other questions? If not, we all get to go to lunch early. Oh, one more. How can you use AI for licensing? Ha, great question. Think about— I'll give you a very simple example. What could you put Mickey Mouse on where people want to buy it? You could think about anything, and then you could figure out which ones have a higher value score. It's that simple. And you could do it at the speed that humanly is not possible, right? And then you can get an answer. And the answer is, it all sounds great until you see it work. So you got to get the sales data and see it correlated. Last more. Last question. Do you see customization taking off? Well, I think we're already there. My son bought me a pen a couple years ago. There was a 2-year wait line, wait list for the pen. It was off of, I forget that site where you all invest in it. Do you know? You invest in it. What is it? Kickstarter. Yep. Kickstarter pen. It took any kind of Montblanc pen insert or anything. It was made out of titanium. And it was created by a series of engineers. He's an engineer. I mean, and then he started to show me how much business was being generated. That single pen generated like $800,000. I was like, what are you talking about? And that's just one little pen. It's happening under our feet. And that goes back to my earlier point about microsegmentation. It's happening everywhere we are. By the way, there are 7 million permutations of how you can configure a Mini Cooper car. All right, thanks everybody. Hope you have a great day.