Nerd Alert: How AI Reveals Hidden Consumer Preferences

Summary

---

Episode Overview

Episode Title: Nerd Alert: How AI Reveals Hidden Consumer Preferences
Podcast: The Marketing Architects
Date: October 2, 2025
Hosts: Elena Jasper (Marketing Team), Rob Demars (Chief Product Architect, Misfits and Machines)

This episode explores how artificial intelligence—specifically the Mixture of Experts (MOE) model—is transforming marketers’ ability to uncover and react to hidden consumer preferences. Building on recent marketing research, hosts Elena and Rob break down how traditional segmentation often misses the nuances of actual human motivation and how smarter models can reveal actionable insights for brand leaders.

---

Key Discussion Points & Insights

1. How Do We Really Choose When We Shop?

• Opening Thought Experiment: Elena challenges Rob to consider what drives his own purchase decisions: price, brand, or features. Rob immediately admits, “I'm a sucker for brand... Tylenol just gets me because I just like. It's Tylenol. It's Tylenol for a reason. I have to buy Tylenol.” (01:08)
• They humorously reflect on Rob’s early-adopter purchase of Apple’s Vision Pro goggles as another example of brand loyalty.

2. Introducing the Mixture of Experts (MOE) Model

• Traditional Models vs. New AI Approaches:
  • Old models (like multinomial logit) “assume we all make decisions the same way, or at best, that we can be grouped into a few fixed segments. But reality is way messier.” (02:20)
  • The MOE model is “a machine learning framework that uses expert networks to dynamically sort consumers into behavioral segments… based on actual purchase data.” (02:45)

3. Four Hidden Consumer Segments Revealed (03:15)

Based on a study of 100,000 retail transactions over 5 years:

• Price Sensitive Shoppers (35%): “A 1% price increase, just 1%, caused them to buy 2.3% less.”
• Brand Loyalists (25%): “They barely flinch at price changes.”
• Promotion Driven Buyers (20%): “Discounts of 20% or more triggered a disproportionately big sales spike.”
• Feature-Oriented Consumers (19%): “These people care less about price and more about product attributes like quality, certification, sustainability, or tech specs.”

Notable Quote:
“Unlike traditional models, MOE lets people have partial membership access across segments. So you don't have to just fall into one... which makes a lot of sense because we don't always make decisions the same way.” – Elena (03:55)

4. MOE Model Outperforms Old School Segmentation (04:20)

• MOE “predicted consumer choices with nearly 79% accuracy. That's compared to 64% for the old school way…and 71–73% for more advanced logit models.”

5. Rethinking Consumer Buckets

• Rob realizes he’s not just a “brand loyalist,” but is also intrigued by product features:
  "It's sort of like when you take those personality tests… it's not just one particular letter… I lead personally with brand, but I do get pretty intrigued by features and start to correlate those with value as well." (06:16)

• Elena affirms, “Segmentation doesn't have to be static anymore. AI can adjust as consumer behavior shifts with income, with trends, or with market shocks.” (07:08)

6. Marketing Takeaways

• Old Models Over-Simplify: “We assume everyone reacts the same way to price or that people can be split into fixed buying groups. This research shows that's not how real consumers behave.” (05:00)
• Different Levers for Different People:
  • “Price sensitive shoppers need discounts to move while brand loyalists will pay more for their favorites... Promotion driven buyers can be swayed by temporary deals… and feature oriented buyers want quality and attributes. They're not going to be impressed by a price cut.” (05:25)
• Threshold Effects Matter:
  • “A 20% discount acted like a switch. Anything smaller barely moved the needle, but… it triggered a big sales response.” (05:44)
• Dynamic Segmentation:
  • “AI... can give us a more flexible, realistic picture of who their customers are and what's really driving their choices.” (07:08)

---

Metaphors, Quotes & Memorable Moments

Fun Analogies: The Consumer “Zoo” (07:54)

• Elena’s summary using animal metaphors:
  “Most marketing models treat consumers like a herd of cows. You set one price and the whole herd moves the same way. But the mixture of experts model shows that we're more like a zoo. You've got bargain hunting hawks circling for discounts, loyal golden retrievers who always come back to the same brand, deal chasing squirrels stuffing their carts when the sale hits, and picky pandas who only care if the product is premium enough. If you market to everyone like they're the same animal, you're going to miss the show.” (07:54)

• Rob pushes back: “Isn't the whole point that you're not just one animal… you're like one of those man horses or whatever, where you're like centaur, horse on the bottom and the human on the top or whatever.” (08:24)

Application Example: Mattresses (09:02)

• Elena notes: “Maybe you'd look at your customers like they're all those bargain hunters, but really you got this zoo. So you don't want to only rely on that strategy because you've got different types of people who might buy different ways within the group.” (09:04)

---

Timestamped Highlights

• 01:08 – Rob describes his susceptibility to brands (“I succumb to the power of the brand”).
• 02:20 – Elena explains how current models miss the nuance in consumer preference.
• 03:15 – Four hidden consumer segments introduced.
• 03:55 – Importance of partial membership: consumers blend across segments.
• 04:20 – MOE’s 79% accuracy compared to older models.
• 05:24 – Key takeaways for marketers; thresholds and different levers by segment.
• 06:16 – Rob relates segmentation to personality tests.
• 07:08 – AI enables dynamic, flexible segmentation.
• 07:54 – Memorable “consumer zoo” metaphor for dynamic segmentation.
• 09:04 – Mattress example: Don’t assume all customers are deal seekers.

---

Summary: Actionable Implications

• Move beyond one-size-fits-all marketing: Recognize—using AI models like MOE—that customers can simultaneously fit more than one segment and that their buying triggers evolve.
• Get granular with triggers and thresholds: Especially when planning promotions, discounts, or product feature messaging, understand which segment you’re speaking to and what actual change moves the needle.
• Embrace dynamic segmentation: Real-time models let you see your ‘consumer zoo’ as it shifts, letting your marketing grow with your customers, not chase after static categories.

---

Memorable sign-off:
Rob, after a fun, analogy-laden discussion: “At least I listened.”
Elena: “That’s it for this episode of the Marketing Architects. Now go forth and build great marketing.” (09:18–09:44)

Transcript

[A]

Nerd Alert. Learning is important, right?

[B]

Yes, exactly. But a bunch of nerds.

[A]

Nerd alert.

[B]

Marketing Architects. Hello and welcome to the Marketing Architects, a research first podcast dedicated to answering your toughest marketing questions. I'm Elena Jasper on the marketing team here at Marketing Architects, and I'm joined by my co host, Rob demars, the chief product architect of misfits and machines. Hello. Hello. We're back with your weekly Nerd Alert. Every week, I'll take a deep dive into academic marketing research and translate its complex ideas into simple, understandable language for Rob, and of course, for all of you. Are you ready to nerd out, Rob?

[A]

I just ran a diagnostic, Alina, and you're still a big nerd. So good. Good news.

[B]

Perfect. Wouldn't have it any other way. Let's get into it. As always, we'll link the research we cover in the episode notes. This week, I read a study titled How Do Consumers really Choose? Exposing Hidden Preferences with the Mixture of Experts Model by Diego Valarino. This was published this year, 2025. But before I get into things, Rob, let me ask you this. When you buy something, do you think your choice is more driven by price, by brand, or by product features?

[A]

I would like to say that I am a savvy shopper that's discerning of price and features, but I'm not. I'm a sucker for brand. And how many times have I stared at the shelf at Target and have seen the regular Tylenol versus the generic version? And the generic version has the same ingredients and it's cheaper. But Tylenol just gets me because I just like. It's Tylenol. It's. It's Tylenol for a reason. I have to buy Tylenol, so I just. I succumb to the power of the brand.

[B]

I think another good example of that for you is the fact that you were one of the first people to buy those Apple goggles. What are those called?

[A]

Yeah, the Vision Pro.

[B]

Vision Pros. Yes. Don't you think?

[A]

I mean, maybe, but it's also a superhuman technology you strap to your head. So.

[B]

Yeah. When's the last time you wore those?

[A]

When I first bought them, but they're still really cool. But.

[B]

I just read a diagnostic. I think you're a bigger nerd than me.

[A]

All right, well played.

[B]

So most of us would like to think that we could identify this like Rob did. Robber. You're very confident that your choice is driven by brand. But this new research suggests our real preferences are often hidden and only show up when we look at actual shopping data through smarter models. So for decades we've used econometric models like the multinomial logit. Some of us have used those models. I have not. And some of us have to predict consumer choice. But these models assume we all make decisions the same way, or at best, that we can be grouped into a few fixed segments. But reality is way messier. Enter the mixture of experts model or the moe. This is a machine learning framework that uses expert networks to dynamically sort consumers into behavioral segments not based on assumptions, but based on actual purchase data. The researcher, he tested MOE on a massive retail data set with 100,000 transactions over five years. And the model uncovered four hidden consumer segments that explain most purchase decisions. The biggest was price sensitive shoppers. That's about 35% of consumers. A 1% price increase, just 1% caused them to buy 2.3% less. Then there's brand loyalists, which, Rob, you'd put yourself in this bucket, that's around 25%. They barely flinch at price changes. Then we have promotion driven buyers. That's about 20%. Discounts of 20% or more triggered a disproportionately big sales spike. And finally we have feature oriented consumers. That's roughly 19%. These people care less about price and more about product attributes like quality, certification, sustainability, or tech specs. Unlike traditional models, MOE lets people have partial membership access across segments. So you don't have to just fall into one. If you're usually price sensitive but occasionally swayed by a favorite brand, they the model will pick that up instead of forcing you into just one bucket, which makes a lot of sense because we don't always make decisions the same way. And we looked at performance. Moe predicted consumer choices with nearly 79% accuracy. That's compared to 64% for the old school way. We used to do this and compared that to. And that's compared to 71 to 73% for more advanced logit models. All right, Rob, does that change at all how you thought about yourself falling into the brand bucket? Or do you still. Are you still pretty confident that. Or maybe this could be a fun question too. If you were in a second bucket or like a mix of buckets, where do you think you'd land?

[A]

Boy, Elena, I've never needed the Rob GPT more than now. That was really smart and a lot of math. I was taking notes, but they don't mean anything.

[B]

It's okay. It looked like you break it down for me actively. Okay, so you don't. You don't want to answer my question?

[A]

Not really.

[B]

Okay. Okay, fine. All right, so what should marketers, what should Rob actually take away from this? First, models that many companies still rely on are oversimplifying consumer behavior. We assume that everyone reacts the same way to price or that people can be neatly split into fixed buying groups. This research shows that's not how real consumers behave, which I think Dale Harrison would agree with this study.

[A]

Yeah.

[B]

Second, different consumers truly respond to different levers. Price sensitive shoppers need discounts to move while brand loyalists will pay more for their favorites. Promotion driven buyers can be swayed by temporary deals, but they're especially reactive when to discount crosses a certain threshold. And feature oriented buyers want quality and attributes like sustainability or innovation. They're not going to be impressed by a price cut. So I think that's kind of interesting that if you could look at your customer base and see like where do most people fall? That could help drive your marketing strategy. The study also highlights how important thresholds can be. So in the promotion driven segment, a 20% discount acted like a switch. Anything smaller barely moved the needle. But when you cross that line, it triggered a big sales response. So that's interesting to know if you're running a promotion like knowing like where is the line where people are going to really start to respond. And I'm guessing we have some history testing that sort of thing. Rob?

[A]

Yeah, that was super, super duper helpful. And I would agree that, yeah, we're not just binary, right? There's different triggers for us, there's different blends. It's sort of like when you take those personality tests and it's not just one particular letter, it's like, yeah, but you're kind of going over on this side of the fence or the color or whatever. And I definitely think that I lead personally with brand, but I do get pretty intrigued by features and start to correlate those with value as well. So I can see that, yeah, we're not just one size fits the category.

[B]

Yeah, yeah, I know that can be a problem with some of those personality tests. Like we do the disc test here and sometimes you get sort of stuck in one area and it's hard to get out of that. Once you're framed, when you really look.

[A]

At them, you're like, well you're really not just that color. You actually veer towards this side.

[B]

And so yeah, nice to know, nice to know how it balances out. So what this does is this model, this mixture of experts model, it shows that segmentation doesn't have to be static anymore. AI can adjust as consumer behavior shifts with income, with trends, or with market shocks. That means that by using AI, marketers can get a more flexible, realistic pictures of who their customers are and what's really driving their choices. So kind of fun. Another AI. Another use of AI is to use it to sort of see where your customers are going and track it over time and allow them to exist in different categories. All right, now for a Rob GPT.

[A]

I think you just gave me the Rob GPT.

[B]

No, that wasn't the Rob GPT. That was just a. Oh, that was just more of an explanation.

[A]

That was the Rob GPT. No, that was really helpful.

[B]

No, lucky for you, we got. Okay, got something a little more.

[A]

That's getting even more clear.

[B]

Okay. Most marketing models treat consumers like a herd of cows. You set one price and the whole herd moves the same way. But the mixture of experts model shows that we're more like a zoom. You've got bargain hunting hawks circling for discounts, loyal golden retrievers who always come back to the same brand deal, chasing squirrels, stuffing their carts when the sale hits, and picky pandas who only care if the product is premium enough. If you market to everyone like they're the same animal, you're going to miss the show. That was fun.

[A]

Well, that was really fun.

[B]

But do you not like the Rob.

[A]

Does it really work? Because isn't the what wasn't kind of part of the point, though, that you're not just one animal. You're a blend of animals. Like, you're like one of those man horses or whatever, where you're like centaur horse on the bottom and the human on the top or whatever.

[B]

Yeah, no, you're right. That is part of it. But I guess you're right. Maybe this is a bad Rob GPT.

[A]

Well, at least I listened. At least I listened.

[B]

Yeah, you did. I mean, like, how people might think, okay, I'm selling. What's a. What's a category that's often discounted that we could use as a natural brand? Maybe. Aren't they kind of relying on sales?

[A]

Probably always. There's always a sale on a mattress. Yep.

[B]

Yeah. So you. Maybe you'd look at your customers like they're all those bargain hunters, but really you got this zoo. So you don't want to only rely on that strategy because you've got different types of people who might buy different ways within the group.

[A]

Right, right.

[B]

But I agree, it's not a perfect, perfect analogy. But you know what? We can't always be perfect. That's it for this episode of the Marketing Architects. We'd like to thank Taylor de Los Reyes for producing the show. You can connect with us on LinkedIn, and if you like the podcast, please leave us a review. Now go forth and build great marketing. All right, Ready?

[A]

Yes. You got to be hydrated for nerd alerts.

[B]

That had such a big echo that Waterfall Marketing Architects.