Why 95% of companies stall on AI, and the behavioral science that fixes it

MichaelAaron and Richard unpack a new Behavioral Insights Team report on why most companies fail to get employees using AI, covering how social proof lifts adoption, why removing small points of friction beats motivational pitches, and why framing AI as something to lose outperforms framing it as a gain.

Transcript

MichaelAaron Flicker: [00:00:00] Welcome back to Behavioral Science
for Brands, a podcast where we bridge the gap between academics and
practical marketing. Every week, we sit down and go deep behind the
science that powers great marketing today. I'm MichaelAaron Flicker.
Richard Shotton: And I'm Richard Shotton.
MichaelAaron Flicker: And today, we're diving into adopting AI and the
human insights that will help organizations do it better.
Let's get into it.
So Richard, as we were preparing for today's episode, we said virtually
every organization is challenged with how do they bring AI into their
workforce? How do they bring AI into their marketing departments, their
product teams across the organization? How do you get AI adopted in
the organization?
And there's some real human hesitation that is going on here, and that
we [00:01:00] thought if we set up an episode all about adopting AI, it
would be really valuable for folks listening, and we think there was just a
lot of really interesting studies that we could bring to it. So we felt it was
a pretty fertile topic- Yeah
to bring to a- all our listeners.
Richard Shotton: It's very easy to fall for the Kevin Costner "Field of
Dreams" myth: build it, and they will come. It's very easy to fall for the
idea that you pay for a premium version of ChatGPT or Claude, and then
everyone starts using it in the company, but that isn't normally what
happens.
So the brilliant point about all these experiments we're gonna talk about
is they give you some really simple ideas to make sure that the
investment in AI that your company's making is harnessed as much as
possible, and that actually your staff use this amazing technology.
MichaelAaron Flicker: And there's a research report from MIT that said
over 300 AI initiatives [00:02:00] that while over 40% of organizations

have piloted general purpose-use LLMs, only 5% have implemented
embedded or task-specific generative AI, and they call it this gen AI
divide.
And the larger learning is you have these AI models, you have this
amazing technology, and so little of it gets used in organizations today.
You and I actually found this amazing report from the Behavioral Insights
Team called Adopt, and we'll put it in the show notes. It's a free
resource, a 29-page report from the Behavioral Insights Team that really
looks at getting AI adopted inside firms, and we use that as a lot of the
inspiration for the material we're gonna talk about today.
Richard Shotton: Yes, and it's probably worth stressing that we're
gonna go through an awful lot of [00:03:00] really useful studies and
insights, but it's just skimming the surface. Worth looking at that report.
There's plenty more in there.
MichaelAaron Flicker: Yeah. And that's actually part of a larger series
of reports that they've put together.
Can you tell everybody a little bit about the larger series, and then we'll
get into, we'll get into, Ah, okay ... what we want to talk about. So
Richard Shotton: they have done four, and there's how to encourage
people to adopt AI. There's how to, I think align AI with the kind of
interest of the company that people are doing the thing that you want.
I think there's also augment, and then there is a fourth A which I cannot
remember. I always have a problem with these- Adapt ... acronyms.
MichaelAaron Flicker: A-adapt.
Richard Shotton: So we know we've done, we know we've got adapt.
Okay, adopt, adapt- Adapt ... align, and augment. Is that the four? Okay.
I think so. Good. Glad someone's paying attention to these proper-
MichaelAaron Flicker: I didn't mean to put you on the spot.
Yeah.
Richard Shotton: Yeah.

MichaelAaron Flicker: Which is the final. Yeah ... But really interesting,
and so we're choosing one of the four- Yeah ... adopt for today's
conversation. Yes. And and from within [00:04:00] that report, they go to
the the inspired by the COM-B model. They look at three core pillars for
driving AI adoption in the report: motivation, capability, and trust.
And maybe we can start driving into those. Motivation being providing a
reason for using AI. Capability, enabling people to use AI confidently.
And trust, assuaging people's worries around using AI as the three major
pillars that they wrote the report against.
Richard Shotton: Yeah, and I think that would be a good way to split
the episode.
So maybe we start with motivation. Now, how can you encourage people
to want to use AI? And one of the principles that they discuss in the
report is social proof. So this is the idea that [00:05:00] when people
make a decision, they don't behave as discrete individuals, but instead
they are very influenced by what they think others are doing.
So if you can make AI adoption look like it's popular, look like lots of
other colleagues are doing it, then more people will start using AI. Now,
the brilliant thing about the report Is that it doesn't just use generic
studies to argue this point. And we've often talked about studies like
Cialdini's, which show that if you stress most guests in a hotel reuse
their towel, you make reusing that towel much more likely to happen,
often much more effective than stressing the environmental benefits.
So there are lots already of studies and experiments into things like
towel reuse or menu choices or doctors prescribing particular medicines.
Social proof has been shown very widely. But what's really interesting is
even though AI is a reasonably [00:06:00] recent technology, there's
already studies out there showing that social proof affects AI adoption.
So there's a brilliant 2024 study from Jan Judek, who's at Ostfalia
University in Germany. He recruits 254 people, and he gives them a
reasonably simple task. They are shown data about a stock price for 20
weeks, so they're given this information, and then the task is to try and
accurately predict what's gonna happen in the next 10 weeks Everyone
takes part in this task.

All 254 participants, they are offered the use of a AI system that can help
with that forecasting. But the twist in the experiment is sometimes social
proof is used to encourage adoption of the AI tool, sometimes it isn't. So
for half the people, they are given the reasons why this AI forecasting
[00:07:00] tool is brilliant, but they're told that few of the previous
participants use the tool.
In that setting, 52% of people decide to use the AI tool. 52%. Other half
of people, they are given exactly the same information about the
forecasting tool, but this time they are told lots of previous participants
used that AI forecasting tool. And in that setting adoption rates go from
52% in the control to 65%.
So you've got this 25% uplift in adoption. The key point to emphasize
here is the supposed strengths of this forecasting tool, they do not vary
by group. All that changes is what other people have done. This, I think,
is a really important experiment because it's so easy to fall into this trap
of thinking, "This time it's different.
Oh, [00:08:00] AI is such a revolutionary new tool. We can't use all these
old experiments to increase adoption. We-- You know they won't work.
Everything's different." What this is brilliant at doing is showing that in
the specific realm of AI, if you want people to use it, you should use one
of the most well-proven tactics, stress what others are doing.
MichaelAaron Flicker: And if you lean back, it makes sense that- The
principles of influence and persuasion are going to be the same when
humans are trying to deal with anything that's new. And what we've said
many times in behavioral science is especially effective when there's
uncertainty, especially effective when people don't feel that they know
the right context of how to choose.
Yeah. These things like social proof can be very impactful. So giving so
leaning back a little bit and looking more broadly at how can we get
humans to adopt AI effectively? These [00:09:00] same drivers of human
behavior that have always existed makes sense that they would be
effective here as well.
Richard Shotton: Yeah, absolutely.
Your point about uncertainty is crucial. In situations where people don't
have past experience to draw on, in situations where they don't quite

know what the outcome's gonna be, they are particularly influenced by
the behavior of others. I think people fall into a slight logical error of
thinking other people must have some information.
They must be drawing on insight to make their decisions. So we are
influenced by the crowd. So AI, if anything, will be a better area for using
social proof than many others.
MichaelAaron Flicker: Social proof was one of the very earliest
episodes that we did. We had Aperol was an example. It had nothing to
do with AI, of how you can show popularity in a bar just by the color of
the [00:10:00] liquid of Aperol.
But something we talked about in that episode that's ringing true to me
now is if most people are doing something, let's make that more popular.
But if pe- if the majority action is not doing it, you wanna be careful not to
make that the thing you focus on. And so being creative about how you
use social proof in this instance makes a big difference.
So if most people in the company are already using AI maybe that's
what you focus on. But if most staff haven't started using AI yet, or
maybe they're only using it occasionally, then you gotta look somewhere
else. Like- Yes ... maybe tell them about competing companies or
people outside your company and how they're applying it, right?
'Cause you wanna get that momentum. You don't wanna focus on the
wrong thing.
Richard Shotton: Yeah. The, and the other things you could do, you
could discreetly message the bottom 49% of users and tell them
honestly that they are using it [00:11:00] less than the norm, less than
the peers. That's another way of harnessing social proof, discreet
messaging to the bottom half.
Or you could, again, honestly, and it has to always be honest- Of course
... you could talk about the increase in popularity of using AI within the
company. That's known as dynamic social proof. So stressing a change
in usage, a positive change in usage harnesses social proof very well.
'Cause what tends to happen is if it's a a minority behavior and it's being
used a little bit but it's growing, people tend to extrapolate outwards, and
they imagine that growth will continue forever and it's gonna become
popular.

So that's another way of using social proof.
MichaelAaron Flicker: I was thinking about I was reading a news article
about the difference between Anthropic and OpenAI and their
accounting models. And f- OpenAI uses a more classic accounting
model where they recognize [00:12:00] revenue when they get it.
Anthropic, and I'm-- I if I'm not getting this 100% right, I'm 90% right.
Anthropic realizes the future revenue of the commitments that people
are making now, which helped it get a higher valuation than OpenAI. But
it's that interesting idea. And so the whole conversation in the article was
which one is actually a more valuable company? Is it the one that has
the money that's coming in right now, or is it the one that's foreseeing
the momentum of the money, the commitments it's getting, and should it
be valued higher because of that?
And it's very much like what you're saying here. The way that they're
talking about their accounting structures is the way we're talking about
how to use social proof. Should you get credit for the momentum of
people continuing to come on the platform? How can you use the
momentum of people adopting to get people to feel that there's a lot
more [00:13:00] social proof happening there?
There's a lot more success evidence of success just by talking about the
increase in the percentage of people coming on the on using AI more
heavily.
Richard Shotton: Y- yeah. So th- this idea there's some lovely
experiments. It's not just us speculating. There's a lovely Sparkman and
Walton study.
It's essentially arguing that if you can't claim absolute popularity then
turn to dynamic social proof, emphasize your momentum, and then it'll
make the same behavior that much more appealing. So I certainly think
it's an effective way to get people to adopt AI. When it comes to the cold,
hard reality of what that company might be worth, to me, that would be
s- sounding alarm bells that someone is trying to persuade you about
future money coming in.
There's a proverb, and I never know if these straddle the Atlantic, but
what is it? A bird in the hand is better than two in the bush. Two in the
bush. And that [00:14:00] is something I would be using in this scenario.

MichaelAaron Flicker: Yeah. But luckily, no one is listening to us for
investing advice.
Richard Shotton: Yes, exactly.
Yeah, we should probably have-- There should be some big, Okay. ...
health warning. Yes. These two people do not know anything about
investing. Exactly. Do not listen to them.
MichaelAaron Flicker: Exactly. Yeah. But Otherwise, I would not be
sitting in this small shed in my garden. I'd be in a palace somewhere.
Yeah. Okay.
But when it comes to social proof, almost like you and I are doing right
now, it is best to tailor that social proof claim specifically- Yeah
to where you can make the b- the most impactful claim. So if we are
talking about getting adoption amongst insurance companies say how
many other staff members from other insurance companies- Yeah ... are
using AI. The-- Where-- And that's f- from the Cialdini-Towel study,
right? Where you can get- Yeah
more nar- narrow. That's where it's gonna make the most the biggest
impact. [00:15:00]
Richard Shotton: Yeah. The argument from Cialdini and others is that
we are not equally influenced by the crowd. We are more influenced by
people like ourselves. Even quite tenuous links will boost the impact of
social proof. You're from New Jersey, I'm from Essex.
I will be more influenced by knowing what other people from Essex do
rather than other Brits. You'll be more influenced by knowing what other
people in New Jersey do rather than other people in, in, in America. So
absolutely, if you want maybe new starters to use AI, tell them what
other new starters are doing.
If you want the board to use our AI, tell them what other board members
are doing. Yeah. The at- You don't get the maximum impact by using the
largest number. You get the maximum impact by using the most tailored
number.
MichaelAaron Flicker: Perfect. So- We've talked about this first area

Richard Shotton: Motivation. Yeah. Social proof. Yep.
MichaelAaron Flicker: Let's move to this next area, which is [00:16:00]
capability.
Richard Shotton: Yes. So this is essentially the ar- the pillar of do your
staff have the ability to use this software? And I think we can cover this
one reasonably quickly, which is... 'cause there's some more interesting
brand new studies that I think might be better to focus on. But when it
comes to capability, probably the biggest thing you can do as a company
is focus on: How do I make using AI easier?
Now, I can imagine there'll be many people rolling their eyes and saying
MichaelAaron and Richard are just telling us something we know. Of
course, we know we've got to make the interface and the setup as easy
as possible." But I would stress, and I don't know how I can stress this
any harder, but it's not just a statement of the obvious.
The argument from people like Daniel Kahneman, Nobel Prize winner in
2002, Richard Thaler, Nobel Prize winner in 2017, is that again and
again, people [00:17:00] underestimate the importance of making
something very slightly easier. It tends to have an outsized effect. So
what companies do again and again is they do a cost-benefit analysis,
and because they think removing friction will only have a small uplift,
they don't do it enough.
There's an awful lot of things that they think just aren't worth the
investment because they massively underestimate how much of an
impact removing those little bits of friction would be. All sorts of things
that companies can do if they buy into this. Let's have a look about
there's the idea of defaults.
If you want people to use, let's say, an AI note-taker more often, default
to that happening automatically. If someone has a meeting, the AI note-
taker kicks in automatically. It's easily overrideable, but if you set it up as
the default, it will happen a lot more. You might want to think [00:18:00]
about what's known as a sludge audit.
Now, go through the AI user's journey, think about all the little things
they have to do to get thing, get their various different systems in place,
and if there is even the most inconsequential barrier, put more efforts
into resolving it. It'll have that surprisingly large effect.

MichaelAaron Flicker: It's so interesting when you track the usage of
different LLMs.
Google's Gemini is jumping nearly 30% a quarter when they embedded
it inside the Google search. So when you ask Google a natural language
question, it gives you a AI-enabled response that you can then go
deeper into. To me, that's just like a brilliant example of just how you can
set the default. Yes, you could have opened a separate browser window
and went to ChatGPT- Yeah
but the fact that your already natural [00:19:00] proclivity is to search in
Google, Google recognized that and decided to start inserting its AI
answers right into your Google search thread ahead of all of their paid
search results, which arguably it's what's funding all of Google's
revenue. There's this, the, that's just one example of how a change at a
of a default can really- Yeah
make a big i-impact.
Richard Shotton: And I think that is a brilliant example that could act as
a, as an analogy for a company that's listening. They might think, " we've
invested heavily. We've got premium licenses for everyone. We've got
these amazing AI. Surely, because it helps people do such a better job,
they will click three or four times to get to it."
But actually, the evidence would suggest you will get far more uptake,
even if what you're offering is amazing, if you can remove some of those
clicks. Maybe embed it into people's-- into the internet site, [00:20:00] if
that's what you're expecting everyone to go every day. But the simpler
you can make it, the more it will happen.
I think that's a really important one that just gets ignored so often.
MichaelAaron Flicker: And just final thing, I'd love to underscore this
idea of a sludge audit. To me, I had never heard that term until you said
it. And to me, you're just doing what any SaaS product normally does.
You're saying like, "What are the barriers to getting more usage, to
getting better better engagement?"
And then you're trying to break those down. So the same thing, if you're
trying to get AI adopted in your organization, very smart to do an audit to

really identify and then prioritize the biggest areas- Yeah ... which is
holding people back from engaging with AI.
Richard Shotton: Yeah. And it-- I would say what people often do is
they have the question in their head: How do I [00:21:00] motivate
people to want to use AI?
Now, of course that's important, but the bigger question, the one that will
have the m-more sizable impact on usage, is what is stopping people
using AI? That often leads you down to this road of what are the little
barriers, and let's remove them. So you ask the question in slightly
different ways, you end up with very different answers.
MichaelAaron Flicker: Yes. I think that's great. Yeah, absolutely. Okay,
so we started with motivation, then we moved to capability. Now we
wanna move on to the trust pillar. And as you said- Yeah ... some more
deep and interesting studies here.
Richard Shotton: Yeah, the... Yeah, there's some brilliant stuff here
around, I think people being deeply nervous about what the impact of AI
might be on them.
So there are some really interesting studies that suggest, for example
people can often be perceived as lazy if they use, if they're using AI.
So there's a study [00:22:00] from 2025, very recent study. It was done
by Jessica Rye for Duke University. Nice big sample size, 1,215 people.
And she gives this group a paragraph about a lawyer, and the lawyer
has produced some arguments.
Sometimes the participants are told the lawyer drew on the work of a
paralegal. Other times they're told that the lawyer used AI. Now, the
quality of their argument is exactly the same in both situations. But when
people are asked how lazy the lawyer is, you get a very clear difference.
So this is all on a seven-point scale.
The higher the number, the worse. If the lawyer had used a paralegal,
average rating of laziness 2.16. If the lawyer had used AI, that rating
goes up to 2.5. [00:23:00] So exactly the same output, but 16% higher
rating of laziness. Now, this is something, I think this is where the trust
pillar's all about, is people f- rightly judge that all things being equal, they
might be marked down for using AI.

There is this perception that it's a bit of a cheat. So what a company has
to do is tackle this head on. If you don't address it, there will be a very
strong self-interest for people to want to avoid using AI.
MichaelAaron Flicker: And I think an important part of the study is that
when they were rated the quality of these, they were the same.
Yeah. Exactly, yeah. To me-
Richard Shotton: It's only the AI usage that's being judged, yeah.
MichaelAaron Flicker: Yeah, and that it's a perception issue. It is not
a... It the, it's the matter of do people trust that they're going [00:24:00] to
be judged the same way? And this study reveals that no they're gonna
be judged as more lazy.
And so it's incumbent upon organizations to say, "How do we change
that narrative? How do we fight back against that maybe natural sense?"
and you might have seen this in other technology revolutions in the past,
if you are able to work from home on a laptop, were you seen as lazier
than your in-office peers?
Yeah. If you were able to be on a cellphone rather than on an office
phone, were you seen as being lazier? I don't know. I think each round
of technology brings this lack of trust- yeah ... that, that comes with r- i-i-
in a technology innovation.
Richard Shotton: And that I think is a theme, that this is not an
unprecedented change.
Some of the fears and worries echo through previous interventions. So
then you start to think, " if that's the case, what worked really well last
[00:25:00] time?" I would say one thing to address this fear of being
judged by using AI, you've got to make sure that you're harnessing the
messenger effect. So it's not just about w- the rationale that you put out
as a company about why using AI is so important, it's who says it that's
important.
So you wanna be getting the most admired people, maybe the most
senior people, talking about how they've used AI. And then I think you'll
change some of these negative perceptions.

MichaelAaron Flicker: And you and I have talked about this maybe in
our last episode on AI I believe we covered here Was that you also
wanna talk about all the work that you did after the AI produced the initial
results.
That there, there was an illusion of effort, there was actual effort put in
after the work was done because th- after the work was done by AI
because that's where the human- Yeah ... benefit comes from, right? If it
[00:26:00] feels like it's just been automatically and immediately spit out
and copied and pasted, it's gonna be judged as less quality.
Yeah. But if you talk about the effort that went into perfecting it, then
that's g- can be judged much higher.
Richard Shotton: So there's some lovely studies by Andrea Morales
she's at the University of Southern California, and these are into the idea
of the illusion of effort. So what she shows is exactly the same service
from a professional provider.
It is rated very differently if the user thinks that the professional service
provider put lots of effort into it or a little bit of effort into it Exactly the
same service is rated differently depending on these stories of effort.
Now, that original study was done in the world of estate agents or
realtors, as you like to say, or realtors or whatever it is in America.
But there are new studies specifically about AI. There's a Cobi Millett
[00:27:00] study from 2023, who's from VR University in Amsterdam,
who shows people posters. Sometimes they're labeled hand-drawn,
sometimes AI-generated, and repeatedly people rate the hand-drawn
one higher. Exactly the same poster, you just change the label and
artistic merit, creativity, even purchase intent varies wildly.
And when I say wildly, I'm not exaggerating. Purchase intent is 61%
higher when people- Yeah ... think the p- the poster is hand-drawn rather
than AI. So you're absolutely right. The way to get round that is to firstly,
I think, talk about the efforts that you've been to set up the right AI
systems. If you have put together a kind of a new protocol, maybe you
could talk about all the efforts you, you've been to, to make the user
interface simple or the efforts you've put to feed the [00:28:00] right data
into the system.

And then once the output has been created, absolutely, these stories of
how the user has taken the original output and refined it, you need to
add those in if you want it to be valued.
MichaelAaron Flicker: And we're talking in this episode about how to
get greater adoption of AI in your organization. So on this one hand,
we're talking about giving the trust that it's going to be well-received.
And I think on the other hand what the behavioral insights team advises
against is this, quote, "do or die approach." The that you must adopt AI
now and this really hits on reactance, right? This resisting a new
behavior when it's forced upon us when we feel our freedom of choice is
removed.
That, that's an important part of all this.
Richard Shotton: Yeah. There's this idea of reactance. There's a
classic 1976 study by [00:29:00] Pennebaker, I think it was at the
University of Texas, and he is trying to stop graffiti in some university
toilets. And sometimes he puts up a sign that's very polite very kind of
soft language, "Please don't graffiti."
Other times it's quite... the sign is quite authoritarian. Do not graffiti!"
Exclamation mark, underlining. And what he finds is that the
authoritarian message actually doubles, I think, the amount of graffiti
versus the- ... more gentle one. And the argument here is one of the big
drivers of human behavior is a desire to retain a sense of control and
agency.
And if the persuasive tactics to adopt AI become too forceful, they can
backfire. So what I think companies should think about is how do you
emphasize that [00:30:00] users are in control, that your staff are in
control? And the best way to do that, I think, is to look at co-creation.
There's an amazing idea called the IKEA effect.
Dan Ariely and Michael Norton came up with it in 2012, and essentially
they argued the more effort people put into a product and their products
they use were things like IKEA boxes or origami birds. The more efforts
people put in, the more they appreciate them. So if you go and if you...
Let's say you create a new system for use or embedding LLMs into your
business. If you just expect people to adopt that because of your

demands as the leadership team, you might get one level uptake. If you
ask people all their points of view, and even if that didn't change one iota
what you'd recommend, the fact that you've asked them, the fact that
people feel like they've input into it, they are much more [00:31:00] likely
to want to adopt it.
And that is a underselling the argument because, of course, if you go
and ask people, you will get some really good feedback. It will- That's
right ... change what you would've wanted to do. But even if it doesn't,
it's much more likely to be a successful piece of persuasion.
MichaelAaron Flicker: Because it gives them a sense of voice, it k-
gives them a sense of co-creation, and it removes that s- feeling that
they don't have any freedom in adopting this technology.
It brings them in and makes them a a part of building the system rather
than being forced to use it.
Richard Shotton: Yes, exactly. Exactly.
MichaelAaron Flicker: The Behavioral Insights Team has other
suggestions of what w- we should consider as companies trying to get
increased trust in building AI systems that gets used.
Richard Shotton: Yes. So another thing they talked about was how you
frame your messages.
What [00:32:00] most companies do when they are trying to increase
adoption of a positive new technology like AI, they will talk about all the
upsides, all the benefits. The suggestion from the Behavioral Insights
Team is you should flip that on its head and talk about what you might
miss out on if you don't use this amazing new technology.
So this is the idea of loss aversion. So experiments into loss aversion go
all the way back to 1973, 1974 work of Kahneman and Tversky. We've
previously talked about some wonderful studies by Elliot Aronson at
Harvard. He did this amazing study where he tried to sell people loft
insulation. And if he went round homeowners and said, "Take out loft
insulation and you'll save 75 cents a day," he got mediocre uptake.
But when he went out and said, "If you don't take out loft insulation, you'll
be wasting 75 cents a day," he got 56% more people agreeing

[00:33:00] to the loft insulation. So exactly the same financial incentive,
but when it was framed as a potential gain, did much worse than it was--
if it was framed as a potential loss if it wasn't done.
So lots and lots of evidence generally supporting loss aversion. But
what's so interesting is, again, there are specific studies about AI
adoption. In a way, this is amazing. AI has only been around a few
years, and there are already so many behavioral science studies giving
people advice on how to effectively get AI adopted.
So this is another 2025 study. It's by Joseph Buckman, who's at Georgia
State University, and he gave people a task in which they could either
ask a person for help or AI. Now, they had 20 [00:34:00] rounds of this
task, and sometimes this is the gain frame. People were told, "Every
time you get the correct answer, you will be given 50 cents."
So there's 20 rounds. Maximum you can win is $10. In that setting, and
remember these people had a choice, they could get the AI help or the
human help, most people went for the human. They asked for the
human rather than the AI. The other half of people taking part in the
experiment, they were given basically the same offer, but it was framed
as a loss potentially.
So what they were told is, "We are gonna give you $10 up front. Every
mistake you make, we take 50 cents." Now, realistically, logically, that is
the same thing. If you get everything wrong, you get zero. If you get
everything right, you get $10. It's the same as the previous group, but it's
framed- But it feels different
very much as a loss. If-- yeah yes. That's the key [00:35:00] point.
There's this amazing Amos Tversky quote where he says, "We don't
choose between options. We choose between descriptions of options."
So yes, the option is staying the same, but the descriptions are very
different. Now, in that setting, we go from human dominance, now AI
and humans are equally influential.
People are equally likely to turn to AI as a human for the help. So the
argument here is if you want to increase adoption, don't just focus on
what people can gain by using AI. You've got to talk about what they're
missing out on if they don't use it

MichaelAaron Flicker: It really it really stands in my mind that fear of
loss that that this study is is exploiting it activates this desire to adopt
something different so that you don't lose it rather than that ability to just
[00:36:00] build up to the $10. So in what ways can we-- can companies
who are thinking about getting AI adopted can use this without instilling
fear, without instilling a sense of a sense of being scared while they do
it, I think is really critical
Richard Shotton: Yeah you're right to draw that distinction.
And just because a little bit of loss aversion is powerful, it doesn't mean
that y- you know, this extrapolates all the way to such a forceful use of
loss aversion that you scare people to their wits' end. So you're
absolutely right. Y- there's always gotta be a bit of art as well as science.
You want to be emphasizing that people could be missing out on top
performance if they don't use AI.
What you don't wanna be doing is, I don't know, taking that to the Xth
degree and stress that you might not have a job if you don't use this
[00:37:00] thing. Th- there's always a balance, and what we tend to find
is stuff that makes people feel really bad, like ashamed or scared or
worried y- you're straying into the ostrich effect here, and things that
really scare people, it tends to create an ostrich-like response, i.e.,
people stick their head in their sand.
MichaelAaron Flicker: And they ignore it, and so it's not- effective
anyway. Exactly.
Richard Shotton: Exactly.
MichaelAaron Flicker: So yeah, on top of it being not a very nice thing
to do it
Richard Shotton: may not be effective. Yeah. How lovely that a nice
way of behaving and an effective one overlap. And maybe- Yeah
as an example of that, in the situation we talk about, people were
given... There was $10 on the table. The gain frame, you can gain 50
cents each time. The loss frame, you'll lose 50 cents every time you
make a cent-- a mistake. Y- that I think is worth having for people to
have in the back of their minds.

That isn't scaring [00:38:00] people. It is, I think, getting the balance just
right.
MichaelAaron Flicker: Yeah. It's about activating that fear of losing
without creating a fear of the situation. It's- Yeah ... it's activating that,
that desire not to lose r- rather than just an equivalent gain. Yeah. Yeah.
Absolutely.
Exactly.
Richard Shotton: Exactly.
MichaelAaron Flicker: So we looked at all of these-
Richard Shotton: Yeah ...
MichaelAaron Flicker: areas, Richard, pr- primarily to help people think,
okay, AI is a big market-moving force that can change businesses, but
there's human insight behind whether they're successfully adopted or
not. And so today's episode was really challenging us to say, "Where are
the areas of human insight that we can most push on to get to the best
outcomes for adoption?"
Richard Shotton: Yeah. Yeah, exactly. That's a lovely summary. And I
really enjoyed this episode because I think one of the criticisms,
[00:39:00] albeit an unfair criticism, of behavioral science is, "Oh gosh,
this stuff yeah, I'm sure it worked when people weren't very sophisticated
in the 1970s or the 1950s, but oh, it's not gonna work now."
We have just talked about a succession of studies from 2024, 2025, and
these insights into human nature, they are eternal. The operating system
of the human mind does not change year to year, and I found this
selection of studies we talked about are, I think, a very strong proof point
for that.
MichaelAaron Flicker: Yeah. And you you briefly mentioned it when we
were deep in the studies, but shocking how many new studies in the last
two years to prove and give confidence to some of these more
evergreen concepts that we've talked about with 50, 70 years of study
history behind them. To see studies in the last 12, 24 months that prove
it through as it relates to AI [00:40:00] should only give us all more

confidence and comfort that we're on the right path using behavioral
science to help get AI better adopted.
Richard Shotton: I-i-it's such a popular field of study now, and in many
ways we've to- we've taken AI as an example, but you could pick any
question, I think. Whatever category, whatever unique challenge a
listener has, there will be loads of studies out there that could help them.
We have picked AI adoption, but we could talk about any other
challenge.
There's so much brilliant research out there now. There's this massive,
amazingly valuable set of studies. I think what marketers should be
doing, one of their key jobs should be doing, is identifying the challenge
they have and then just matching the right behavioral science study to
help them solve that.
And people should find that liberating. If you have a challenge, you
haven't-- you're not starting from a blank sheet of paper. There's already
some amazing work out there that can give you the best [00:41:00]
chances of success.
MichaelAaron Flicker: And if you've listened to this whole episode up to
this point and you're still hesitant about adopting AI even more strongly
yourself- I don't know what to do if you are.
Yeah ... It, it stands to me it's always a nice reminder to, to think about
that the best hands driving AI right now are the most senior people with
the most ability to make connections and bring extra- extrapolated ideas
to the table to guide and refine what AI's output is. So if you are a mid-
level, a senior person thinking about, "How can I get AI adopted?"
And maybe you have this internal fear that soon it will replace the
uniquely human thing that you bring to the table it, there's really a
counterbalance here that by [00:42:00] adding y- your insights and your
knowledge to what AI is uniquely able to do well, you have a really you
have a unique advantage.
And I started by saying that's the most senior people, but actually I don't-
- I think everybody who has their hands on the wheel for AI adoption has
that chance. Somewhere in the report they have this chart of all the
different ways to engage AI. They call it H1 to H5, and it's H1 and H2 are

where AI agents drive task completion, and H4 and H5 are where
humans drive task completion.
K- H3 falls in the middle where it's an equal partnership. But to me, this
is the future of work. This is how we're thinking about how we're gonna
get work done differently when we think more broadly about how we're
gonna use these tools to get to better-
Richard Shotton: Yeah ...
MichaelAaron Flicker: faster outcomes.
Richard Shotton: So y- your point about [00:43:00] it is the interplay of
the human and the computer that, that really matters.
I, I completely agree with that. One of the earliest analogies I heard
about AI stuck with me 'cause I love playing chess, and it was a chess
playing analogy. And it went through the kind of history of thinking about
the best players in chess. And for millennia it was humans. The
grandmaster was the pinnacle of chess playing.
Then, I don't know, in 1980s sometime, not maybe early 1990s
MichaelAaron Flicker: IBM Watson?
Richard Shotton: Yeah. I thought it was Deep Blue, but what- whatever
the IBM computer was. Then people think, "Oh, that's, that beats..." I
think it was Kasparov. It's computers that are the ultimate chess players.
But now the thinking is actually it's the combination of the two.
A reasonably good player and a reasonably good chess computer
system will be either the world's best chess computer or the world's best
player. So the [00:44:00] phrase they have this amazing phrase for
player plus computer, which is essential, and I think that's a lovely way
of thinking about AI.
That it is this combination, the strengths of both systems that that, that is
the ultimate develop- the kind of ultimate player of chess or coming up
with a-
MichaelAaron Flicker: Thanks for the great engagement-

Richard Shotton: Don't
MichaelAaron Flicker: know where to go after that slightly dubious
analogy. Yes. I know. I like I like it. I like it. How about we wrap up
today?
Richard Shotton: Yes. Yes, let's wrap it up. Yeah.
MichaelAaron Flicker: Remind everyone- Yeah ... Of our topics that we
hit today.
Richard Shotton: So we have been going through the a- Adopt report
by the Behavioral Insights Team.
And it's amazing catalog of brilliant experiments that can help you
encourage people to adopt AI. We talked about boosting motivation, and
there we talked mainly about social proof. So the [00:45:00] amazing
Judex study, which showed if you stress honestly that lots of other
people are using AI within company, you'll make it more likely that
happens.
Then we quite briefly talked about the capability section, mainly focusing
on the underappreciated effect of friction. So normally when people are
trying to encourage a behavior, they focus on making that behavior look
appealing, motivating people what, to want to do it. Of course that's
important, but the argument for behavioral science is the key question,
maybe the ultimate question is what is stopping people adopting the
behavior you want?
Do your sludge audit, go through the employee journey, identify even
small little barriers to adoption, and put more efforts into resolving them.
And if you do that, these small changes in ease will massively boost
uptake. We then moved into the trust area and there we talked about
that amazing study by Reif or Reif, [00:46:00] apologies, not sure on the
pronunciation there.
That was the study in which she showed there is this unfortunate
perception that users of AI are lazy. That's something companies need
to address if they wanna boost trust in the system. And then we talked
about, in a slightly nuanced way, the Buckman study into loss aversion.
You are more likely to get adoption within kind of reasonable limits if you

stress what people are missing out on if they don't adopt AI, rather than
stressing what they can gain if they use AI.
And then there was- And with that everyone ... a slightly long-winded
explanation about chess, which I just try and force in on any podcast,
and this felt like my one and only chance this year. Yeah.
MichaelAaron Flicker: This would be the most human of things to, to
counteract the AI efficiency of it all, Richard, is to make sure our
passions get in here too.
It's all good. And with that we say thank you to everyone for listening
today. And if you found today's [00:47:00] conversation engaging and
interesting, please share it and and follow our pages. It helps us reach
more people just like you. And until next time, I'm MichaelAaron Flicker.
Richard Shotton: And I'm Richard Shotton.
MichaelAaron Flicker: Thanks for listening.
Advertisement: Behavioral Science for Brands is brought to you by
Method1, recognized as one of the fastest-growing companies in
America for the third year in a row, featured on Inc.'s 5000 list. Method1
is a proudly independent creative and media agency grounded in
behavioral science. They exist to make brands irresistible, helping
people discover products, services, and experiences that bring moments
of joy to their lives.
As behavior change experts, Method1 creates emotional connections
that drive true brand value for their clients, focusing primarily with
indulgence brands in the CPG space. Find out more at method1.com.