Retailgentic | Consumer Behavior & Retail Trend

What happens when consumers start trusting AI agents more than friends for shopping advice? On this episode of Retailgentic, we welcome Professor Luca Cian from the University of Virginia’s Darden School of Business.

Luca brings a unique perspective as a psychologist turned marketing scholar whose research spans visual persuasion, AI adoption, and consumer trust. We dive deep into how shoppers perceive AI in utilitarian vs. hedonic purchases, why trust in AI depends on “augmented decision-making,” and how brands should prepare for an AI-first retail future.

From the paradox of choice to the emerging world of “GenAI Engine Optimization (GEO),” this conversation is packed with insights on how AI is rewriting the rules of marketing, loyalty, and consumer psychology.

Tune in for:
  • The psychology of trust in AI: why we accept AI for practical tasks but resist it in creative/hedonic domains
  • The paradox of choice and how AI alleviates decision fatigue
  • How brands can adapt to the rise of agentic commerce and GEO (GenAI Engine Optimization)
  • Future of loyalty programs when shopping decisions are mediated by AI agents
🕒 Timestamps:
00:02:05 — Guest intro: Luca Cion, UVA Darden professor, studies consumer psychology and AI in marketing.
00:09:36 — Early AI research with GPT-3 preview; skepticism then rapid shift in perception.
00:17:12 — Survey: 60% trust AI purchase suggestions more than friends → ties to “paradox of choice.”
00:21:01 — Word-of-Machine Effect: AI trusted for utilitarian products, humans preferred for hedonic ones.
00:28:53 — Trust grows when AI is augmented by humans and shows transparency (black box effect reduced).
00:36:11 — AI Mistake Generalization: consumers forgive humans but not AI; one AI error erodes trust broadly.
00:46:06 — Loyalty as a decision shortcut; in the AI age, human touch and anthropomorphism become vital.
00:50:40 — People blame AI less than humans for unfair treatment (e.g., firing decisions).
00:53:32 — Executive MBA discussions: AI displacing middle management but creating future opportunities.
00:58:00 — Managers should lead with human strategic vision, then use AI/data to validate decisions.

🧠 Want to stay ahead in AI commerce? Subscribe and follow along: 
📰 Subscribe to the free Substack: retailgentic.ai
📺 Watch episodes on YouTube: youtube.com/@retailgentic
🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Creators and Guests

Host
Scot Wingo
CEO and Co-founder, ReFiBuy.ai

What is Retailgentic | Consumer Behavior & Retail Trend?

Retailgentic, presented by the ReFiBuy Media Network and hosted by Scot Wingo, is where Agentic Commerce meets retail innovation. We help retailers and brands prepare for the future of agent-driven shopping with actionable news, expert insights, dxis, forward-looking predictions, and practical guidance on Agentic Commerce Optimization (ACO).

Loyalty will be very important.

Because loyalty, if you think about,
is another way to short-circuit, to

shorten the, the decision-making process.

Like, I already know a product,

I'm loyal to that brand.

I d- I will not do the
awareness, interest.

You know, I will not do the search again.

I will just go and buy the product again.

And- I also believe, I don't know how, if
this is true or not, I mean, this is just

my intuition, that in an age where we will
use much more AI and everything, we still

need even more a human touch in some.

Welcome to the Retailgentic Podcast,
where we explore the intersection

of retail, e-commerce, and AI
agents, which we call Retailgentic.

Our mission is simple.

We are shining a bright light
on the new agentic AI shopping

mega-trend, which will disrupt the
trillion-dollar digital retail market.

The stakes are high.

Google, OpenAI, Microsoft,
Perplexity, Amazon, Apple, Meta, and

Anthropic are all fighting to be the
consumer AI shopping agent winner.

There are gonna be some tough decisions
for online brands and retailers to make.

What happens to checkouts?

How should payments work?

Should you continue to invest
in retail media networks?

What does the world look like when
over half of your traffic is not

from humans, but from AI agents?

These existential questions
are the tip of the iceberg.

On this podcast, we're talking to industry
experts, analysts, and influencers

to get answers to these questions,
identify trends and best practices.

This is Retailgentic, your
podcast partner in navigating

our AI shopping agent future.

And now here's your host, Scot Wingo.

Welcome back to the Retailgentic Podcast.

On today's episode, I am excited
to host our first academic

on the podcast, Luca Ciahn.

He is a professor at the University
of Virginia's Darden School

of Business, where he teaches
MBA students about marketing.

Luca studies the intersection of
creativity, social psychology, and

marketing, which has led him to
our favorite topic, how consumers

use and think about AI when they're
making purchasing decisions.

Some topics we cover that I found very
interesting that all have implications

on the adoption and strategies around
agentic commerce are word of machine,

utilitarian versus hedonic purchases,
trust and the black box effect, the

AI mistake generalization problem,
and AI versus human blame attribution.

These were all relatively new to me.

I have read some of his papers, so I
did kind of know going in, but it was

interesting to hear from him first,
uh, person how he thinks about these

and what it means for our world.

That's really just the tip of
the iceberg on this far-ranging

discussion that I really enjoyed,
and I wanna thank Luca for coming on.

Uh, I know it's, uh, unusual for
a academic to pop on a podcast.

He was very, uh, open and, uh,
we had a really fun discussion.

I hope you enjoy this conversation
with Luca as much as I did.

Luca Cian, welcome to
Retailgentic Podcast.

Thank you.

Um, I'm very excited to be here with you.

Yeah, I…

What, what, uh, led me to wanting to
have you on the podcast is I was reading,

uh, you know, I'm, I'm always searching
out there for interesting content about

what's going on and what people are
writing, and you had a Darden piece.

Um, you know, you're, you're a professor
there at the University of Virginia,

uh, Darden School of Business, and there
was a really good piece that talked

about some work you've done around,
uh, what I call agentic commerce.

You may have a different
academic phrase for it.

So I was like, that's someone…

We haven't had a, an
academic on the podcast yet.

Usually, we have people that are,
you know, uh, thought leaders from

consultancies and other businesses.

So I thought it would be interesting
to have a new voice on the podcast.

Oh, that's exciting.

Thank you for hosting me.

Yeah.

Let's start, uh, with, uh,
go through a background.

So even though you're based out of
Virginia, I imagine, being from the

University of Virginia, you don't
seem to have a Virginia accent.

Let's, let's talk about that.

Right.

I have a very subtle accent.

Yeah, yeah.

So origi- originally, I'm from Italy.

I know, I know.

Surprise here.

I don't know.

And, uh, I, I was born in a little town
in the mountains, very, very close,

uh, with the border with Austria.

The town is called Cortina d'Ampezzo.

Actually, we will have the Olympics
soon, in a few months over there

between Cortina and Milan, the Winter
Olympics, so that would be fun.

And then, uh, yeah, I- I am a first
generation college student, so, um, I,

I spent some times there in, in Cortina.

I did actually my bachelor and my
master in, uh, Trieste in Italy.

And then I started my PhD in Italy
at University of Verona, where

I actually finished there, but I
also spent some times as a visiting

scholar at University of Michigan.

And after that, I decided to
take a postdoc at University

of Michigan in marketing.

And so overall, I spent, like, five
years at University of Michigan.

I had a great time.

And then after that, I moved to
University of Virginia, Darden

School of Business, and I've been
there, wow, almost now s- 10 years.

So I became assistant professor
there, and then associate professor,

and now recently, full professor.

So I did the, the whole trajectory.

Ah, so you're … Are you tenured now?

I know that's, like, the big,
the big goal is to get tenured,

so have you been tenured?

Is that-

Yes, yes, yes.

So, so now I kind of laughed.

Congratulations.

That's awesome.

I know that's a big mark in your career
as a professor, so congrats on that.

That's true.

So being from Northern Italy, I know
Milan is the center of luxury goods.

Is that how you got interested
in this concept of marketing?

Is that kind of a, from, from what
goes on in your part of the world?

Oh, such a, such a good question.

No, actually, I didn't know what to do.

How, you know, when I was, when
I was a kid, I didn't know what,

what to be, what, uh, as, when I
was going to be a grownup, right?

So, uh, at the beginning, back
in the day when I was a small

kid, I want to be a painter.

And, uh, since then, I
lo- always loved art.

And actually, um, I think my passion
for marketing is more like, um, uh,

intersection of three passion I have.

One is the, uh, love I have for
art and creativity, maybe because

since I was a kid, I was exposed
to, uh, painters, creative people.

So I don't know, I always
loved them and the creativity.

And of course, being in Italy,
you're always exposed to art,

so I always loved that part.

And then when I was in college, I took
a class in, uh, uh, social psychology,

and I fell in love with that.

I said, "Oh my God, what is this thing?

I love social psychology."

And, uh, so actually my PhD is in, uh,
in psychology, in social psychology.

And, uh, um, and then marketing for me is
sort of, uh- A conjunction of psychology,

and I do love psychology, my PhD is
there, but I always love something that

is a little bit concrete as well, right?

Something that I can see the results.

And sometimes in psychology,
not all the time, but sometimes

psychology tend to be at times
theoretical, at least in academia.

And, uh, I thought that applying
psychology to marketing was going to

be great because was understanding
consumer psychology, but also see--

Uh, I was, I was able then to see,
uh, results in the real world.

And so that was very exciting for me.

And at the beginning, uh, my research
was not on AI actually, but was

mostly on, um, visual persuasion.

And so I was, um, bringing my love for
art and visual art into psychology, into

marketing, you know, like combining all
these three, uh, three things together,

and that's, that's where I actually
start doing research at the beginning.

And then slowly I changed my
interest into, uh, AI and,

um, the psychology of AI.

So, um, you know, I love to
work on AI, but I work from a

psychology, psychological point of
view, like how people react to AI.

That's, that's my, the, my interest point.

So I kind of started at your most
recent pr- publication and went back

in time, and I don't have access to
all the cool academic libraries and

stuff, but I noticed you've been writing
about this for, for quite a while.

So, uh, pretty early on, like right
around when ChatGPT came out, like

late '22, early 20, you know, uh, '23.

Um, was that just luck or you saw
the ChatGPT coming out and you

thought, "Hmm, this is gonna be…"

Like, what was your thought process
that led you to really engage with that?

You know, it's, it's, it's so true.

Like, I remember I was working on a
piece on, uh, AI, uh, at that time

working, like able to create, um- news
pieces, like, you know, like being

able to b- to write piece, piece,
a piece of news for, uh, journals.

And at that time, ChatGPT-3
was not yet public.

We had, um, Academic Preview, and I
do remember that, uh, in my, you know,

the first times I, I was presenting,
uh, those job, like, that work.

Very few people believed that, oh, ChatGPT
can actually write like a human or can

actually, uh, do decent writing jobs.

So my point there was to convey
that actually the, uh, artificial

intelligence and, uh, uh, large language
models were able to do these things.

But at the time, specifically, ChatGPT-3
was only a preview in the academic

world, and so very few people knew the
real possibilities and the real impact.

So it has changed a
lot in those few years.

But, uh, going back to your question,
um, it was a mix, to be honest.

Uh, I always been in love with technology.

So I, I say that I love art, that's true,
but always been, uh, a geek and a nerd,

and so I always loved to be I always
love, uh, you know, uh, technology, so

I was always interested in technology.

And having a psycho- background
in psychology, I was in- always

interested in sort of, uh, human
And technology interaction.

And then I cannot deny that a big
part of, uh, my, um, research in AI

was also due, uh, to meeting, uh, a
co-author of mine, um, another professor,

her name is Chiara, Chiara Longoni,
and, um, we were working on another

paper, and then she told me, "Luca,
let's start working on AI together."

I said, "Oh my God, that's--
I always loved that.

Let's do it."

And since then, um, it was, uh,
love at first sight with this topic.

So, um, yeah.

So it was a co-author pushing me into
that, but I was always interested and,

uh, and then I thought it was very
interesting because, uh, it is moving so

fast and there are so many implication in
terms of society, in terms of industry,

uh, that I found it very fascinating
because it keeps me on my toes.

Like, whatever, maybe even a research
paper I'm, I could have done three or

five years ago may be still relevant
today, but maybe not, because so many

things have changed in five years, not
only from the, uh, programming point of

view, but also, like, in terms of how
we see artificial intelligence, right?

So in schools, in…

But also how, you know, uh, the lay
person, the lay people view of and

perception of AI has changed dramatically
in the last three years, right?

So, so that's, that's,
that's what, uh, make…

You know, it's, this is one of the things
that makes this work very interesting.

Yeah.

It's, it's interesting, there's a
very binary reaction, especially

with creative-minded people.

Um, most of them, I would say, in
my experience, like 70% reject it

outright 'cause they're like, "This,
this is a threat to human creativity."

And then a- another set kind of embraces
it and starts thinking through the

other side of the, the, the equation.

So, so right there is a, an
interesting psychological split that

I've seen, and I'm not in any means
a, a psych-, you know, thinking

that way, but it's interesting
to see people have a very binary

reaction to it at a very high level.

I agree, and, uh, that's
very fascinating, right?

And, uh, indeed, even in general, if
you show some, uh, a piece of art, for

example, to some people, and then, uh,
uh, you ask people to evaluate the piece

of art, and then- If you say that that
piece of art was made by a human versus

an AI, in general, when people know that
a, that piece of art is made by an AI, the

evaluation of that piece of art are always
lower than the same piece of art but when

you tell them that it's made by a human.

Independently if the piece of art was
made by an AI or by a human, I think, uh,

people really want or hope that humans are
still the stronghold of creativity, right?

Uh, but, but AI is very good
at, uh, at being creative,

especially recently, right?

So AI is able to create music, uh, pieces
of art, often, uh, indistinguishable

from creations made by people.

And so I think, uh, clearly
that generate a lot of fear.

Uh, you know, we always thought, "Oh,
what, what distinguishes us from machines?

That's creativity," right?

But if that goes away, then we're screwed.

And so I think we're really,
really afraid of that, right?

But even in academia, if you think,
um, I remember when AI ChatGPT came,

uh, soon after we had all, you know,
like s- many students using it, and

we all, all immediately realized that,
uh, many times ChatGPT could have…

Like if you, for example, for a
midterm or, or sometimes even for a

final exam, ChatGPT could, uh, answer
very well to most of the questions.

And so at the beginning, there
was a complete rejection.

Like, students cannot use ChatGPT
in any format, you know, like

sort of complete avoidance.

And now you can see that many people
in academia are shifting towards the

other way that is, well, students
are going to use ChatGPT outside, uh,

when they're done with their degree.

And so- Right … if we want to provide
an education that is relevant, we

need to, uh, use ChatGPT in school.

And then we have just to rethink how to
teach and how to grade and how to evaluate

students using generative intelligence,
uh, artificial intelligence and, uh…

But, uh, without telling them not
to use it, because that would be

a little bit, um- You know, uh,
out of, like, would fill a bubble.

Like in academia or at university
you cannot use ChatGPT, but as soon

as, uh, you're done with college,
yes, you will use it every day.

And then honestly, it's very difficult to
implement a system where we can monitor

when people use or not use ChatGPT.

So I think it's, it's rather, it's better
to ask student to embrace them, but

clearly requires from professor an effort
to rethink the way we teach, the way they

evaluate students, et cetera, et cetera.

Yeah.

The take home final is, uh, no,
it's not gonna work anymore.

Let's talk about some of your research.

So I'll let you, uh, summarize,
like in the, you know…

So, so you're, you've kind of started
gravitating in your research to

some interesting consumer behavior
that's in my world of, you know,

how is this gonna hit, hit the way
people shop online and buy things?

Uh, what are some of the areas
of that that you find interesting

that you've been researching?

Oh, that's, uh, so, so interesting.

Uh, I think there is so much there, right?

Uh, because-

If you think about the way people
shop is changing very quickly.

And I was actually recently, uh,
commenting some, um, survey data

that showed that I think 60% of
people, um, trust more artificial

intelligence in providing suggestions
for their purchases more than friends.

Yeah.

Right?

So clearly there is a shift in, in the
way we shop, uh, as consumers and in,

uh, in the way we also trust ChatGPT or
any other LLM, artificial intelligence,

generative artificial intelligence,
whatever you want to call them.

But, uh, clearly there is a shift in
the way also we trust them, right?

And, uh, and I think, um, it's because
w- there is a big offload in the

terms of, of the mental burden, uh-

When we were to buy something
before, we had to look to so

many different alternatives.

And in psychology, we know there's
a fact called basically when you're

overwhelmed by too many choices, right?

So-

The paradox of choice.

Yeah.

Right.

The, the paradox of choice.

When there are too many choices,
people tend to be, to feel overwhelmed,

and then they don't like that.

They don't like the situation
where there are too many choices.

We like choices, but when there are
too many, that creates a lot of stress.

And if you think about right now in
any retail environment, there are

so many brands about everything.

Even the most common thing, like
water, has dozen of brands, right?

Yeah.

And then- Toothbrushes, you got,
like, 800 different toothbrush.

Yeah, it's crazy.

Right.

Right.

So how do you choose them?

And, uh, and that's hell on earth
if you're like me, that is like,

um, maximizer, meaning when you
start a research, you want really

to find the best thing possible.

You know, like sometimes psychology,
we call maximizer and optimizer.

Sometimes people say, "Okay, as soon
as I find, I find a option that is

good enough, I stop researching."

That's optimizer, and they live a
wonderful life that I envy very much.

And then there are people like me,
more maximizers, say, "No, I need

to do the best possible choice."

But there are so many, uh,
choices around adoptions that

makes things very difficult.

But what AI helps a lot is basically
creating a lot of helping, offloading

a lot of this burden of choice, right?

Able to look at thousand of
reviews and giving you a set

of, um, potential choices.

That clearly changes a lot in terms
of how business work and how, um,

humans, uh, or I call humans, but
you know, people, consumers do their

choices, and that's so, so interesting.

And, um- In particular, one, one paper
that I thought was very interesting is

what, uh, a paper that I did with, uh, my
colleague Chiara, that I mentioned early.

Um, we found an effect called
word of machine effect.

We call it that way.

Thought it was catchy.

And the, the main idea there was
that, um, for utilitarian products,

think things like, I don't know,
insurance, financial services or basic

household items, consumers are quite
comfortable letting AI take the lead.

So our research show that in general,
people prefer AI recommendation

in this context more than human
recommendations because people believe

that AI is more competent at processing
functional utilitarian attributes,

things like price, efficiency,
performance, et cetera, et cetera.

However, when we think about things
like hedonic products, think about

experiences, luxury items, I don't
know, creative pursuits, consumer

tend to resist AI involvement.

And so in those cases, people prefer human
recommendations over AI recommendations

because especially here, because
these purchases are more based on

emotion, taste, subjective experience.

So every time there is a hedonic
choice, and hedonic, it means something

based on senses, something based on
emotion, subjective, then people tend

to prefer human recommendations over AI
recommendation because we think human

are more competent than AI in doing
these things And there is a caveat

here, an interesting twist if you wish,
if you want, because here, uh, is more

like a bias than real AI capabilities.

Meaning people think that humans, that
AI is better than humans in terms of

what I call hu- utilitarian abilities,
but actually humans are pretty good

at utilitarian abilities as well.

And then we found that people have
a bias for which people think that

AI is less capable of humans when
there are things related to things

that are subjective or hedonic.

And that's not really true, because if you
think about AI successfully creates flower

arr- arrangements for flower companies,
develop new flavors for Starbucks, uh,

creates art that, you know, uh, is valued
thousands of dollars, is able to recognize

human emotion and respond accordingly.

And so this goes back to what we
discussed before with creativity, right?

So, uh, is actually able to be creative,
but people still, um- think that humans

are better off, are better than AI in the
more sensory, hedonic, creative things.

And so for those kind of products, people
believe recommendation coming from people

more than recommendation coming from AI.

Do, do you think that's set in stone,
or do you think over time as people

get a little bit more comfortable with
AIs and they're, you know, we're--

this is happening so fast, you probably
did your, your survey before they,

they've really started to develop more
personalities, which we saw with the

release of GPT-5 when they turned off 4.0

and everyone was freaking out because they
had, they had like a relationship with 4.0

that they felt they killed their friend
and just replaced it with a new, a new

entity, and they had to bring it back
because people had this, you know, they

had a bond with that model and the way
it talked and interacted with them.

Some, some of it is in a bad place.

But I think overall it's just people were
like, "You know, I prefer the way that

one chatted with me versus this new one."

Do you think those type of attachments
will help people get over that bias, or

is this gonna be kind of a lizard brain
with us for a long time kind of a thing?

No, I totally agree with you.

I, I, I think things will change.

I don't know how long they're going to
take, but things are going to change.

And if you think about sometimes
it's just a question of even framing.

Like for example, think
about Netflix, right?

Netflix provides, um, movie
recommendations, but it-- and then

movies is something very hedonic, right?

Based on taste and everything.

But Netflix doesn't say, "Oh, our
algorithms is, um, recommends this

kind of products or movies or shows."

They say, "People like you likes,
liked this movie and, uh, or show

and shows and everything," right?

So sometimes it's a question of framing.

You can use AI, but frame it differently.

But to better answer your
question, I think, yeah, things

are going to change very quickly.

So it could be that in few y- in few
years or even now, I d- this, this

was a research that we did in four or
five, four years ago, three years ago.

So maybe right now people actually
are, um, are more accepting, at least

some people are more accepting of AI,
uh, in terms of being competent at,

at hedonic, artsy, creative- things.

And, um, for sure the
newer generations, right?

That are using AI as a, sometimes
as a companion, as a, almost like a

friend or sometimes as a therapist.

So clearly when AI changes personality
because there is a upgrade, like you

mentioned, from ChatGPT 4 point, I don't
remember, 5 to ChatGPT 5 now, people are

going to be upset saying, "This is not
the-" Persona I was talking before, right?

So what was happening?

So, and it happened even before
with other platforms that they had

to basically go back to previous
models because people were upset.

It's like if a friend of yours all
suddenly change personalities and you

don't recognize them anymore, right?

So-

Yeah, I want the old you back.

Yeah.

Andreessen Horowitz came out with a
piece, you probably haven't seen it

yet, and it was kind of a framework
for the amount of consideration and

how you use AIs in a different way.

Um, and they basically said for these
utilitarian things, more of what I would

call an agent, where you just say it,
you know, it, it, it says, "Hey, uh,

hey Lucia, here's, here's three things.

Which do you want?"

And you say, "I'll take choice B," and
then it just goes and buys it for you.

Um, so, so in that, in that low
consideration model, what you're

calling utilitarian, it would
be almost like fully agentic.

So it would go-- It would have the
agency to go do things for you.

But then for these things that get
higher consideration, it would be more

of a companion and kind of, you know,
along the way, maybe help you look at

new offerings and then offer to, to, to
help take care of more of the process.

And, uh, so I think that's interesting
that, you know, there, to your

point of framing, um, a lot of
times with these new things, we,

we look at them through one lens.

But I think as, as this gets more mature,
it will also have these shades where if

you're doing something high, you know,
a high consideration, a luxury item or

something, it can be on the side like a
GPS kind of a thing when you want it more

in the background versus in the foreground
for something that's low consideration.

So that was a, a new framework that,
that I thought was interesting.

That's super interesting, and made
me think of another, um, insight

that I had throughout my research
that I think could be relevant here

that I think is very fascinating,
is that, um, sometimes we find that

people have what we call AI aversion.

Like people don't like AI for some aspect.

Think about in the previous
research was like, uh, when there

was a hedonic purchase, people
think AI was less competent.

Or sometime we did-- we had another paper
for, um, um, that I mentioned before.

We ask for accuracy of news pieces
written by AI versus humans, and

people believe AI-written pieces
were less accurate in general.

Uh, but- In general, every aversion,
every bias against AI in my research

disappears when there is something that
we can call augmented decision-making.

Meaning when there is a human plus
AI, so maybe the AI does something.

But when you tell people that they
can … There is like they have, um,

um, the possibility to, to take the
final decision Or to interact with

AI, then the bias against AI reduces
significantly or disappears completely.

So it's basically the
best of the both worlds.

So in the moment you allow people
to inter- like, to take, to, to have

some control over the decision-making
of the AI, to make the AI agent not

completely independent, then people
trust AI much more than when they

think AI is completely independent.

That's what we found in general, uh, as a
way to, uh, diminish aversion versus AI.

Like, the idea that you can put some
human elements in the decision-making

process that usually, um, alleviates
the doubts people can have about AI.

So if you're building an AI, you wanna
make sure that, that it feels like

you're, you're augmenting what the human
is doing and having these checkpoints

where you're asking for permission and
you're not going too far down the process

versus just running, running rampant.

Ab- absolutely.

Two points there.

One, exactly like that.

Like, if people feel they have
some, how can I say, interaction,

but not interaction, but like some
power over the decision-making that

the AI is doing, then they feel
more confident in the final output.

Especially, you know, if I buy a product,
for example, and I know that behind, or if

there is, I don't know, if the government,
for example, use AI for, you know, review

my taxes, I could be afraid of that.

But if I know that there is a human
somewhere that reviews what the

AI does, then people tend to be
much more, uh, welcoming the AI

implementation in the, for example,
in the governmental, uh, space, right?

But as long as there is some human
overlook and some human ability

to take the final decision.

That's number one.

And number two, uh, what other colleagues
of mine found is this sort of called

black bloc- black, sorry, black box
effect, meaning, uh, people have the

impression that they can understand other
humans, but they cannot understand AI.

Now, I agree that most
people cannot understand AI.

I have some doubts that we can
really understand what's going

on in the mind of other humans.

Yeah.

I don't know that.

But in general, people have this
bias that, oh, for example, if

you go, if you think about a human
doctor versus an AI doctor, right?

I can think or I can understand
what is going on- In the mind of a,

a human doctor, but I have no idea
what's in the mind of an AI doctor.

That's what in general people think.

And there's-- and therefore, there
is less trust, uh, usually towards

AI when they think it's a black box,
meaning when I- they think that the,

they have no idea what's going on.

But if you leave-- if you give even
the illusion of transparency, like

if you help people understanding
what's going on behind the process

guiding the AI, then the trust
towards the AI increases tremendously.

So these are the two things that in
my research I found being able to

increase trust towards AI tremendously.

One is, uh, augmented, uh,
decision-making, like there is a

human in the process somewhere.

And number two, uh, the ability to see
through the black box of AI, even if

it's something just perceptual, right?

So just giving the impression that
people can understand what's going

on, but just giving little insights
behind the decision-making process of

the AI gives much more peace of mind.

Yeah.

Anthropic writes a lot about this
on the geeky side around in, in,

uh, let's see, what's the word?

Interpretability.

So they're trying to understand how
are the LLMs learning, and then why

are they coming up with the answers.

And if we could, if we could, you know,
decode that better, then maybe people,

A, from a safety perspective, it's always
good to know 'cause we all wanna avoid

the, the Terminator outcome, right?

Um, but, uh, you know, the-- But then
also, if we knew more about that,

people would trust it more if they
kinda understood why is it saying

the kinda things it is and, and you
know, and why, why is it developing

these kind of personality traits that,
that are, you know, kind of unusual

sometimes, things like that nature.

So there's a whole nother set of
scientists researching that from a,

you know, helping explain it, but
just even, even helping consumers

understand, oh, you know, it's
been trained on this set of data.

That's how it knows what it knows.

Some of that transparency
can always help as well.

Yeah.

What are some of the other, kind of in
the world of consumer behavior as it

meets AI and, and especially anything
on the retail purchasing world, any

other interesting insights from your
research that, that have surprised you or?

The interesting thing that I, I found,
um, very interesting, especially from

the, uh, retailing side, is that, um, I
feel that people Are welcoming AI when

the predictive personalization feels
useful, helpful, rather than creepy.

And easier said than done, I
understand it's a very subtle line,

but it's a very important line, uh,
for brands and retailers, right?

So, you know, uh, if AI agents can
really feel that they're useful,

you can feel that they're useful
and they're personalized to you,

but not creepy, then people really,
uh, are willing to use them.

But in the moment they feel creepy,
maybe because they are too much

of a black box, or maybe because

they know something that they were
not really allowed to know about you,

then people lose trust very quickly.

And another thing that we discovered
in our research that I thought was

interesting, uh, we, we, we did this not
with traditional retailers, we did this

research in terms of, uh, government, uh,
agency, but I think could be applied in

general, uh, is that what we found there
is that, um, basically when AI makes a

mistake, people generalize that mistake
much more than when humans make a mistake.

Meaning, uh, say that there
is a person making a mistake,

you give them a second chance.

When AI makes a mistake, people
say, "Well, the AI is not able

to change, is not able to learn."

That's not true, but you don't
give them a second chance.

Not only that, say that a, a doctor
makes a mistake, then if you meet

another doctor, you say, "Well, every
human is different, so if a doctor

makes a mistake, it's not that any
doctor will make the same mistake."

But with AI, we found that people
overgeneralize mistakes made by AI,

meaning if one AI makes a mistake,
people tend to think all the AIs, all,

you know, all the, uh, system that
use AI will make a similar mistake.

And so there is a lot of general-
generalizability over there that is not

actually true, because every company
use- uses AI in very different ways

and trained in very different ways.

Yeah.

And I spend a lot of time in the
different models, and they have such

different personalities and behaviors.

They, you know- I guess I've
kind of gone through the uncanny

valley that I, I wouldn't expect
them to make the same mistake.

They, they have such different outputs
that, you know, they-- Maybe when

they were early, they were pretty
similar, but they've diverged quite

a bit now over the last 18 months.

Interesting, though.

I agree.

So it's important not to have a
hallucination 'cause then they, they're,

you're gonna, they're gonna really go
put up the wall of distrust, basically.

Yes.

Yes.

Yes.

And that's, uh, that's something I-
Very problematic in, in different ways.

If you think about the government,
if they use AI and then there is a--

Or even a company, if they use AI and
then there is a mistake, people tend

to distrust that company or, or that
governmental agency in the future.

Even if they say, "Oh, we corrected
that problem," but people say,

"Yeah, but your AI did a mistake.

I really don't trust it anymore."

Right?

So now everyone wants to deploy AI,
but what we found is that people

are very reactive to mistakes made
by the AI, meaning people are less

forgiving, let's put it like that.

You know, and maybe this will change
over time, but people think really

that the AI is something more stable,
and so if an AI makes a mistake,

they will always make a mistake.

And so that's, that's something to, to
think about when you are a government-

governmental agency or a company, right?

If you deploy it too quickly
and there is a big mistake, then

the risk is that people are not
gonna use it anymore, right?

So, um, hallucinations, as you
said, is a, is a big problem.

And I think even if now they say
that ChatGPT-5 has diminished the

hallucinations, I think that many,
many people will say, "Oh, you know,

back in the day, two years ago, I tried
ChatGPT, it had a lot of hallucinations.

I don't trust ChatGPT."

You know, like, it will take
some time to regain credibility,

um, in the eyes of many people.

Not all, but many.

For-- If you were, you-- Let's say a, a,
a big retailer or a brand, like a Nike

or someone like that came to you and
said, "Okay, uh, you, you understand all

this consumer behavior things going on.

What, what can we do to prepare for
the, you know, consumers using these

increasingly to do their shopping?"

These, these, you know, what I call
agentic companion for shopping.

Um, do you have any, any top-of-mind
thoughts for things they can do, um,

to get ready for, for this new world?

Yes.

I think, uh, there are, there are
many things that retailers could do,

and very interesting discussions here
in terms of what, in general, you

know, like, uh, at different stages.

Like, I'm thinking, for example, you
know, sometimes in marketing we have

this funnel that sometimes it's called
AIDA or in many different ways, but you

know, there is attention, then there
is interest, then there is desire,

and then there is action, right?

And so as a marketer or a company,
you have to first You know, create

attention towards your product
and interest, desire, and finally

lead the customer toward action.

Now, there are thousands of
different models, but more or less

they go through this funnel sort of
conceptually speaking, um, figure.

But AI is, is changing this, uh, a lot
because it compresses a lot the area

of, of interest, desire, especially of
the, uh, awareness because I can just

use generative AI to give me a list
of the best products to use it, right?

So as a company, yes, I can, I
can still do advertisement and

reach, increase attention, and, uh,
interest, uh, using advertisement.

But if then people use ChatGPT to
offload many of their choices, then how

impactful is that form of advertisement?

I, I don't know, right?

So I think right now we are, um, in a
moment where there should be a shift, at

least in terms of some, um, marketing,
uh, funds that instead of being invested

in traditional advertising or, you
know, whatever form it is, should be

more in terms of creating content.

Like what … How a, a large
language model is able to create

this short list of product that
it often suggesting to people?

Well, reading the internet, right?

So the more as a brand you can create
content on the internet that, that

feels authentic through, for example,
communities or sometimes it's called, you

know, like inbound marketing, but the main
idea is like creating content, creating

community, all this kind of stuff, then
those are the information generating a lot

of positive reviews, then those are the
information that most probably LLMs are

using to make their, uh, recommendation.

So it's very difficult to influence the
LLMs right now in terms of, oh, I cannot

pay them, I cannot pay Anthropic to put
me in the top five recommendations of

the best backpacks, for example, you
know, if I'm selling a backpack or if I'm

Nike, you know, the best running shoes.

But if I'm able to have enough good
content about my newest running shoes

and a lot of blogs talk about them,
and if I can create a lot of positive

content, then most probably LLMs are able
to include that in, in their shortlist.

So it's, it's, it's, it's a different way
to influence how cons- consumer, uh, take

their decision, especially if they're
now, they're using AI as a companion

in their shopping decision-making.

Yeah, there's a whole new industry being
born called GEO, which is kind of a,

you know, uh, you had SEO, which was for
how to optimize against the, the Google

out- the PageRank out- rank algorithm.

Uh, but now they're trying to figure
out how do you, how do you get the

content in front of the Gen AIs, thus
GEOs, Gen AI engine optimization.

So yeah, it's an interesting…

Sometimes they call it
answer engine optimization.

I think it's settling on GEO.

But there's a whole, there's
a whole new world there where

everyone's just beginning to
figure out how to make this work.

But you're spot on.

It, it ends up being authentic content.

That's a, that's the short circuit
answer to the whole thing, is how do you

get that authentic content out there?

Yeah, and it's difficult, but
I think ironically it could be

even better for customers, right?

Because if it could be more genuine
and really based on the reality

of a product and service rather
than pure advertising, right?

Yeah.

N- now that we're kind of…

Now a lot of retailers are thinking
about some second order things.

So one of the first order things
is like, how do you just show up?

But then a lot of people are starting
to think if, if the consumer is

finding products and kind of buying
them up here w- at the Gen AI level,

what happens to the loyalty program?

And then like a lot of the
dynamic pricing kind of mechanisms

that, that retailers have.

Uh, this may be something you're not
thinking about because this starts to get

away from the psychology a little bit.

But loyalty is definitely a, you know,
a, a, a psychology element of, of

how do we, how do we make it sticky?

Um, any thoughts on what's gonna
happen to those types of programs?

Uh, that's a very, very cool question.

I think honestly, um

Loyalty will be very important because
loyalty, if you think about, is another

way to short circuit, to shorten
the, the decision-making process.

Like, I already know a product,
I'm loyal to that brand.

I, I will not do the awareness, interest,
you know, I will not do the search again.

I will just go and buy the product again.

And I also believe, I don't know how,
if this is true or not, I mean, this

is just my intuition, that in an age
where we will use much more AI in

everything, we still need even more
a human touch in something, right?

We are humans, we l- we love,
we love to think that…

We love o- we are social
creature eventually, right?

So that's why I think we
anthropomorphize everything.

Meaning, I, AI, like I want
to call it Claude, and I want

it to give a personality.

ChatGPT is my friend, right?

So we, we, we believed all this AI is
sort of like a person, as a friend, right?

So we clearly understand that most
people at least believe, understand

it is not a human, but we tend to
anthropomorphize things, right?

So give them sort of
human characteristics.

And, uh, and loyalty I think
is, is very similar, right?

So we give a, uh, the
brand some human traits.

I'm loyal to this brand because
I like their values, or I like

the way they stand, or the,
I like something about them.

And then loyalty is something that
usually humans have towards other humans,

and why to be loyal towards a brand.

Yes, there are some rewards.

Uh, could be, you know, points or
whatever, but it's also something deeper.

Um, we get attached to a brand, right?

And, uh, and I think it will be important.

What I'm trying to say is that in an
age where there will be a dominance

of AI, people will find, will always
need to fulfill their social needs.

And the more we became, we become
isolated, like we don't go out much

more, we don't have massive group of
friends, we need to fulfill these, uh-

needs of friendship and love.

And then hopefully we fulfill with
other humans, but sometimes we need

to fulfill it with AI or with brands.

And if you think about, or
AI made from brands, right?

So imagine that I am sad one night.

Yes, I can call a friend, but I can
also write to Claude and say, "Hey,

Claude or ChatGPT, I'm sad tonight.

What can I do?"

And so they're always there for me,
sometimes more than humans, right?

And so I think things like, uh, loyalty
and human touches on AI will be even

more important in the future because
people I think the more we go ahead,

the more people will feel alone in
the society, and they will rely on--

They will see brands and ChatGPT
as potential friends or companions.

Um, and so loyalty is absolutely
very important in that sense.

Cool.

Well, I don't wanna take
too much more of your time.

If, um, was there anything that you wanted
to chat about we didn't get to cover?

Maybe, uh, one thing that I thought was,
was interesting that we, um, we discussed,

we, we found, uh, recently, uh, that could
have some implication in the future, is

that imagine that you use, you use AI
to give bad news, or to fire someone,

or to misbehave towards someone, right?

So what we found is that because AI, I
mean, although I just say that people

want to see AI as a human friend, but
in general, many people know that AI

is, so far at least, you know, uh,
doesn't have, uh, agency, then we tend

to, uh, blame AI less than humans.

So what we found is that when an
AI treats you unfairly, you, you're

less upset in general, to simplify.

And, and that has some consequences
because if you think about, well, if I

have to-- If I have a company or if I have
to do something unfair, I can do it as a

manager, or I can use the AI to do that.

So your next layoff may come from an AI.

Correct.

I can use AI to fi- to fire you, you know?

And then, and then people will be
upset to be fired by an AI, but they

will blame less, and they will be less
upset rather than, "Oh, I am, Luca is

going to come and f- and, you know, and
say you're fired," basically, right?

There is less blame, and because of
that, in general, less maybe anger,

but for sure less, less blame.

And so I think that's, that's important
because the more we use AI in the

future, the more things like we will
use AI in substitute of even sometimes

take decision that are not fair.

And so to study how people react
to that could be interesting.

And we found this caveat that are
p- particularly interesting in, in,

in terms of blame that I think could
have some interesting Let me see You

could use it in the classroom for
the bottom half of the results.

You, you, you, you get to tell
the, the top half and then the AI.

Fantastic idea.

I will totally do that.

You gave me a great insight.

Actually, I'm grading right
now, so I will do that.

Could have LucaBot, uh, just write,
"Talk to LucaBot if it, uh, it didn't

go well."

Exactly.

If you have any problem, don't blame
me, uh, blame my LucaBot, absolutely.

Do you talk about these concepts
in your MBA classes or is this

still more on the research side?

No, uh, we, we talk about that.

What are, what are the … And you're,
you're mostly … Your students,

uh, you know, so like here at Duke
and UNC, the MBA, there's two.

There's some of them are, like,
just right out of undergrad.

They're still kind of, like, in
their early 20s, but then there's,

like, the exec program where they're

They've usually been in the industry
for three, four, five years.

Um, what, what are those … Do
you teach both of those cohorts?

Uh, and then what … How do
they react to some of this?

D- that younger generation, how are
they reacting to some of these concepts?

So, um, I traditionally, uh, taught
MBAs, but in the last five years I only

teach, uh, executive MBAs and, uh, and
so people with a lot of experience.

And we do use sometimes,
uh, AI in the classroom.

Like for example, I have a class
where they use AI to generate

some creative concept and then
they have to build over that.

But, but an interesting discussion
we recently had was mostly, that

is not marketing related, but was
mostly on, um, the idea of how

much AI is going to replace jobs at
mid-level management or even higher.

You know, like- If you think about,
um, was the CEO of Ford recently

that said, "Oh, you know, like AI
is going to replace, uh, 30% of…"

I don't remember, don't quote me on this,
but, uh, a big part of, uh, workers.

And Amazon even recently used a lot of--
laid off a lot of people and say that, you

know, was because of increased use of AI.

There are a lot of companies now using
being AI first, meaning using AI more

than, uh, AI first, and then if AI
cannot do it, then we'll hire someone.

But there are a lot of news recently
saying that AI will eliminate many jobs.

That usually happens when
new technology is introduced.

But the interesting part nowadays is
that it's not the blue collar, right?

We are talking about the white
collars, the middle management.

You know, like AI will replace
programmers, AI will replace

middle management, right?

And so…

And, and the discussion we will
having is that, well, you know,

like if you think about Google, uh,
that is now one of the best, uh,

companies in, in the world and hire,
you know, incredible amount of people

working for, for, for it, et cetera.

Well, it didn't existed when
internet was created, right?

And so maybe right now, yes, for sure
AI will create some job displacement.

Absolutely.

Right?

Especially in the future.

But maybe there will-- um, we were
discussing maybe there will be some

companies using AI- That will be
like the Google of AI, but clearly

we cannot see it yet because their,
uh, AI was just introduced, right?

So, uh, we can see the displacement
because it's in front of our eyes, but

we cannot see the future, what could be.

So we are still optimistic in general
in my class in terms of, you know,

AI can also create new jobs and, um,
and maybe new companies over that,

that they don't yet exist, like
internet … Like Google, you know,

uh, was created a lot after- Yeah

uh, internet was created, right?

You have to look at both
sides of the equation.

Sure, there'll be some jobs
lost, but what, what's coming?

What are the created jobs?

Yeah.

Right.

Gotcha.

But clearly there is a lot of, uh, I
think in terms of, uh, executive MBAs,

there is a lot of thinking there, like,
how can I use AI to make my company

more efficient, like more agile?

That means trimming a lot of jobs, right?

But also a lot of discussion in
terms also of how can I use AI

in terms of management, right?

So, um, do I need a creative, uh, director
or I can use AI, for example, right?

Or, uh, do I need to take my
decision only based on AI or what?

And so we are in an interesting situation
for which we came from a time where there

was not much data and all manager wanted
more data, and now we have more data

than, than ever, but maybe too much data.

And so, and I, I think the discussion
we have in class is that what is

important is that managerial decisions
shouldn't be necessarily driven by

data, but you … ideally you want to
first have a strategic This, like a

strategic idea from a, a human point
of view, from you as a manager, right?

And then you use your artificial
intelligence, big data to

confirm or maybe disconfirm the
strategic insight you might have.

But if everything is only, uh, run by
AI and, you know, big data, that could

be problematic because there are so many
direction where big data can go, where

AI can go, and you may lose the sort of
direction, the vision of your company.

So a human element there
is still very important.

Uh, and so-- And that comes first, and
then AI can use, of course, to implement

decision to better, to, you know, like to
im- im- im- improve the decision-making.

But still, hopefully, the strategic vision
come first, and then big data is used

to confirm or disconfirm that strategic
vision, but not the other way around.

You ever heard, heard this, uh, Goodhart's
law where once you start to measure

something, it's-- you over-index on it
and its effectiveness kind of goes away?

Yeah.

There's a whole-- Yeah, there's a, uh,
the CEO of Shopify is big on saying

they, they purposely look at a, a, uh,
you know, a, a matrix or a, uh, you

know, a, a mosaic of KPIs it- together
versus so they don't over-focus on any

one, uh, because they've, you know,
they've, they've over-corrected on like,

let's say, conversions or something
like that versus the top of the funnel.

And so you have to look at…

Yep.

So just it is interesting that we're--
It's the same paradox of choice

when you have so many data points
that if you over-focus on one, then

that may not be the right answer.

Absolutely.

And then I-- Just to conclude, I know
I, I enjoy this discussion so much.

I don't want to spend too much
time, but I love this discussion,

and I'm really passionate about it.

But even sometimes, especially in the
last version of, uh, sometimes when

I use especially, uh, Claude from
Anthropic, I notice that it's very…

I don't know.

Sometimes I notice, you know, in terms
of personality, ChatGPT is always very

more like, "Oh, there are-- If you
do this, there are pros and cons."

Claude is more like, "Oh, I
would suggest you to do this."

But every time you maybe you add an
information, Claude recently will

say, "Oh, that changed everything.

Let me tell you something else."

And, uh, so basically, although, like even
AI sometimes has problems in incorporating

all the information they have, they may
over-index on something and then sort

of, uh, you know, they-- A-as a human
bias, they-- that is still present in

AI, they have cognitive dissonance.

Because they said something before,
they constantly build on that, right?

And so, and so- AI is not there yet, most
probably fortunately, in terms of being,

having a strategic thinking of considering
everything i- in the mix, right?

Um, so.

Awesome.

So this has been a, a fun discussion.

We, we've been all over the place,
but I think that's part of the fun

of, uh, talking to someone like you.

You're thinking, uh, in a, in a really
wide kind of, and simultaneously deep way.

If, if people are interested in
following your research, um, you

know, what's the best place to look?

You're not a big Twitter
person, um, I imagine.

I haven't actually looked.

But I, you know, I…

You have a LinkedIn page.

You're not super active there.

Is it, um…

There is a page on Darden.

Should they…

Uh, I'll put it in the show notes.

Is that kind of the best place?

It kind of…

I noticed it, it rolls up your research
and some of that kind of stuff.

That's probably the best place
where to follow what you're up to.

Yeah, I would say LinkedIn
sometimes is the best place.

Now, I sh- I know I should be better
on social media, but LinkedIn is still

the best place for, for the moment.

And in the future, who knows, maybe I will
be big on, on X or on other social media.

But right now, LinkedIn is the best.

Well, thanks for taking time.

I really appreciate it.

And, uh, let us know next
time you have research.

We'd love to talk about it and
go over what you've discovered.

Well, a lot of fun.

Thank you, Scott, for hosting me.

I had such a blast, so thank
you so much for everything.

Follow, rate, and share to stay up
to date with Retailgenic, your guide

to the future of AI shopping agents.