Latest / The Retail Podcast / Generative AI in Retail: Insights from NVIDIA’s AI Leaders

Transcript

Generative AI in Retail: Insights from NVIDIA’s AI Leaders · The Retail Podcast

0:00
Embrace AI because especially genital AI is bringing so much
0:05
intelligence. And you know, I think I quoted
0:08
our CEO that said, you know, AI is not going to take your job,
0:11
but someone using AI could take your job.
0:26
Hello and welcome to the Retail podcast.
0:29
Now I'm it's really graced because Azita Martine, who is
0:35
the lead or president or vice president, Vice President, I
0:39
always keep promoting people. The CEO of NVIDIA directors, the
0:44
vice president of retail, global retail for AI and NVIDIA was the
0:49
keynote kick off speaker at NRF on day one.
0:53
And as some of you know, I've already given you my thoughts on
0:56
where we're going with AI. Basically, I position the fact
0:59
that last year was about the use case.
1:01
This year we're hearing about AI transformation and Satya has
1:05
already spoken about how the SAS models are going away and
1:08
there's an AI layer that's coming on top.
1:11
But hey, you're not here to listen to me, we're here to
1:14
listen to you. So if you don't mind taking us
1:18
through your because at your keynote, you sort of set the
1:22
scene about what's happening in the world of AI and the industry
1:27
pretty much from a global context.
1:29
And then you sort of board it to what it means to NVIDIA.
1:32
So I was just wondering if you don't mind sharing that with the
1:35
audience? Generative AI is now at a stage
1:39
where it's moving from pilots to become real.
1:42
And the best way I would describe it is it's it's almost
1:46
like an AI factory that takes your data, which is like the raw
1:50
material in that factory, and it's generating digital
1:55
intelligence in the form of AI agents.
1:58
And these AI agents are making your employees more productive.
2:03
They're delivering intelligence that just enables your company
2:08
to be much more productive. So now what customers are
2:13
companies are doing is fine tuning these models with their
2:17
data, whether it's PDFs, documents, customer data and
2:22
images and videos and be able to create these digital agents that
2:29
are helping your employees become more productive.
2:32
And so an example would be shopping advisor, which we're
2:35
going to show to you a little bit later.
2:37
And this shopping assistant is almost like replicating the
2:42
knowledge of all your best sales associate and having that
2:45
available 24 by 7 at scale on your e-commerce side to help
2:50
your customers find exactly what they're looking for.
2:53
What's been top of mind for the retailers that are coming to
2:56
talk to you? You know what?
2:58
What are they saying as like their top of mind for 2025?
3:03
You know, it ranges from everywhere where you know you
3:06
have early adopters and you also have companies that it's like it
3:10
has to be proven by 5 other companies or 10 other companies
3:14
before I would do it. And I would certainly say that
3:18
this is so much more real than e-commerce was in the year 2000.
3:23
And the companies that are going to adopt it early are going to
3:26
absolutely have a competitive advantage over the ones that are
3:31
going to be the laggards. Absolutely.
3:33
And so, and the people that came to to to see some of our
3:38
technology during NRAS, I think range from, you know,
3:43
commitments from the very top that, you know, we want to, you
3:47
know, in the stores, prevent shrink or have a, you know, more
3:52
optimised layout to make it easier for customers to find
3:56
what they're looking for and be able to check out faster.
4:00
And that's all computer vision technology all the way to how
4:04
can you help us improve throughput in our distribution
4:08
centres and how to make fulfilment a lot faster in our
4:12
fulfilment centres all the way to, you know, what are the top
4:17
generated AI use cases that are going to bring business value
4:21
to, to our to the companies. And one of the areas that I
4:26
highly recommend, even though 20%, less than 20% of retail
4:31
revenue comes from e-commerce, I think generative AI has
4:36
completely reinvented e-commerce and there was an opportunity to
4:40
leverage it to improve search, to improve to have these
4:44
shopping assistance on the website and to even create
4:49
advertisement that retailers can make revenue from by selling it
4:54
to to the consumer packaged good companies.
4:58
So I think there's a lot of opportunity to increase revenue
5:02
leveraging generative AI on e-commerce and mobile sites.
5:06
So. It sounds like there's a real
5:07
mix of generate new revenue for retail media, help the customers
5:11
have a better experience through the automation and bringing, and
5:15
you said this earlier on, bringing all of that knowledge
5:18
of your sales to the customers. In terms of Nvidia's focus in
5:22
the retail space, I'm curious about where you see it going on.
5:27
NVIDIA, most companies know as far GPU.
5:29
So our GPU is, is what companies like Open AI perplexity, Meta
5:35
are using to train these real large language models and so
5:40
forth. But, and there is actually an
5:42
accelerated computing company. And what that means is not only
5:46
our hardware, which involves our GPUs, our networking, even our
5:52
own CPUs, but on top of it, we have a whole set of acceleration
5:57
libraries and our application frameworks.
6:01
And basically those acceleration library and application
6:04
frameworks are the acceleration software that developers use to
6:09
build these AI application and to build their, to build and to
6:14
deploy, right. And so we're really a platform
6:18
for building AI applications. And the blueprint that our CEO
6:24
talked about doing his CES Kenu is basically step by step a
6:29
workflow of how do you build a particular Gen AI application
6:35
using our platform with open source models.
6:39
So it's got step by step instructions.
6:44
You know, how do you do fine tuning, how do you use RAC, How
6:48
do you do the different components of building a general
6:53
application? And it even includes the code.
6:56
And so that's what helps developers build these
7:00
applications a lot faster and ensure that their performance
7:05
from an inference perspective, which is the responsiveness of
7:08
the generative AI application is the fastest and the most cost
7:12
efficient. So when you look to the future,
7:14
when you look to the next two or three years, what fee do you see
7:19
in from what you've taken from the show?
7:22
What do you think will be the the sort of winning focus areas
7:26
in AI? Is it in the business?
7:29
Is it in front of house? Is it back?
7:32
I'm just curious on on your opinion.
7:34
I mean, I would say it's almost everywhere, but an area that's
7:38
benefiting the most from AI is supply chain.
7:41
Got you. And there NVIDIA is really
7:44
pioneering two things. One is called physics AI.
7:48
And physics AI is AI that understands physics.
7:52
It understands dimensions, volume and so forth, but more
7:58
importantly, it understands how people and objects behave in a
8:05
digital trend of a real space. So it allows you to create a
8:09
physically accurate digital representation of a store or
8:14
distribution centre, simulate different layouts and be able to
8:19
predict accurately how people and objects behave.
8:23
And so that's being used for optimising throughput and
8:27
distribution centres in your filming centres.
8:29
It's used for optimising layout of stores for higher revenue.
8:34
And so that's physics AI. Second thing that I talked about
8:38
was physical AI. And physical AI is basically AI
8:44
for the physical world. And it's a combination of
8:48
computer vision, which is driven by smart cameras, robots and
8:54
models that understand physics. So it's a combination of our
8:58
Omniverse simulation software, our robotics technology, and our
9:03
computer vision technology. And it's all used to be able to
9:07
train robots in a simulation environment.
9:11
And the reality of it is that that in the physical world, you
9:15
cannot even estimate how many situations can come up.
9:21
But in the, in the, in the digital world, you can actually
9:25
simulate thousands of scenarios and be able to train those
9:29
robots to behave a certain way during different situations.
9:35
So I think we showed a video during the keynote that showed
9:39
that an incident happened in one of the aisles and a bunch of a
9:43
boxes fell. And so there what what was
9:48
happening is the smart cameras were providing perception.
9:53
So it's like you're looking down, the AI is looking down to
9:57
see what is going on. And then you have an agentic
10:00
agent that is now going to has seen this and it's instructing
10:04
another agent to reroute the forklift that was coming to go
10:08
through a different aisle in order to ensure that, you know,
10:13
no safety situation happens, right.
10:16
So that's an example of physical AI, which is for the physical
10:22
world. And I think that's a good way to
10:25
describe this. NVIDIA actually announced our
10:28
Cosmo Nematron foundation model, and that's basically a
10:35
foundation model that has been trained with billions of hours
10:39
of video. Yeah, and it is to the physical
10:42
world what tragedy BT has been to text and language.
10:46
Yeah, I've been in technology all my life and I'm like
10:49
struggling to keep up because my brain is exploding.
10:52
Where what NVIDIA have announced and where you're taking it.
10:55
If you could go back in time and tell a younger you to come into
10:59
this world, I think you've mentioned your career.
11:03
Yeah, it was an aerospace. Engineer, that was.
11:04
It you were an aerospace engineer and then you sort of
11:07
shifted to NVIDIA and you you weren't really meant to end up
11:09
in rhythm. But anyway, my question being if
11:13
it's OK to ask you about empowerment in terms of I'm
11:15
looking at my career, what would be some sort of words of wisdom
11:19
that you would if you could go back to a younger self?
11:21
I think that really generative AI has democratised AI in many
11:27
ways. So I want to encourage, maybe
11:29
you don't need to code, but you can still use generative AI to
11:35
become so much smarter. So I want to encourage the
11:39
younger generation to embrace AI because especially generative AI
11:45
is bringing so much intelligence.
11:47
And you know, I think I quoted our CEO that said, you know, AI
11:52
is not going to take your job, that someone using AI could take
11:57
your job. And so I want the younger
12:00
generation to embrace AI and use it to, to be a lot more
12:05
productive. At the end of the day, I would
12:07
say even if you're in science, you got to love what you do.
12:11
So my first recommendation is really look around you, look
12:16
inside, really understand what your passion is.
12:19
Learn as much as you can about what's happening in that area
12:22
that you're interested in. Choose the top companies in that
12:26
and aim high, aim big. Don't let your insecurity or
12:31
your fear hold you back and get out of your comfort zone.
12:37
Volunteer for roles that maybe you don't need 100% of the
12:41
qualification, but you know, if you even read 60% of it, I'm
12:45
sure you're smart enough to learn the rest of it and
12:48
surround yourself by super smart people.
12:50
Because the more you get out of your comfort zone, the more
12:54
you're going to grow and the more you're going to learn.
12:57
I love it. And then what a beautiful way to
12:59
end the interview. Thank you so much for giving me
13:01
your time. Thank you.
13:02
Thanks for having me. Hi Alex, I'm Cynthia Contours.
13:05
I'm the Director of AI for retail CPG at NVIDIA as is it
13:09
was talking to you about our AI blueprint for retail shopping
13:12
assistance. Yeah, I'm here to show you a
13:13
little bit about it and and answer some questions.
13:16
So yeah, we're demonstrating the AI assistant for a number of
13:20
different categories. 1 is fashion, the other is furniture.
13:24
Two really interesting categories, but also I can talk
13:26
to you about grocery as another example.
13:29
So let me go ahead and start with, let me go ahead and start
13:31
with fashion, which is as an example.
13:33
Now one of the things that consumers have really been
13:36
struggled, struggling with is it had to change their behaviour to
13:39
an e-commerce site. I need to think about what are
13:41
the keywords I need to search. And the reality is that's not
13:44
the way we work, you know, as you mean this, you know, what
13:47
are the upcoming fashion trends for this spring?
13:50
Hey, do you have any purses or shoes to go with this skirt?
13:53
That's really the way we work. So I'm going to do an example
13:56
here of I have an upcoming event that I'm going to a lunch and
14:00
I'd like a recommendation for a dress or skirt so.
14:20
OK, Alex, see, as you can see, I asked for a dress and skirt
14:23
appropriate for an outdoor event and it brought back a dress and
14:26
skirt. But since it gave me a skirt, it
14:28
also gave me some blouse ideas as well.
14:30
Now typically in a search, you come back with just the products
14:33
and then I need to start digging in, double click on the product,
14:37
go and see more information, and then come back out, double click
14:40
again. So then the beautiful thing with
14:42
the shopping assistance is I can now just ask questions.
14:45
And so with that, let me go ahead and show you how that
14:48
works. So I like the skirt, but of
14:55
course the natural question is does it require dry cleaning?
14:59
So typos and all.
15:17
It's AI. It understands.
15:20
So here we've got some great results coming back.
15:24
I didn't have to go in and look and hunt for that information is
15:26
able to go and search for that information, bring it back makes
15:29
it, it transforms it from my interactive with the e-commerce
15:32
site too. I'm working with your stylist or
15:33
your store associate just really naturally the next piece that's
15:38
great is then being able to ask multiple questions, all right?
15:42
Or being able to say, oh, I saw these pair of shoes online.
15:45
Do you have something like that? So let me go ahead and show you
15:47
what that looks like. It's really an image search.
15:50
So let me go ahead and transition and show you what
15:52
that might look like. So I saw these pair of shoes
15:55
online. They were really fabulous.
15:58
And I want to see if you have something similar for me because
16:01
of course they were a little expensive.
16:02
So do you have any shoes like these?
16:12
And so in the background, we're going and saying, OK, what are
16:16
the attributes of this image? Let me see what other products
16:19
in the catalogue are similar to this?
16:21
It could be red shoes, it could be strappy shoes.
16:23
It could be shoes with embellishments.
16:25
And in fact, you can see that's what we've got.
16:27
We have one pair of shoes. We've got something strappy.
16:30
Now this this shoe has an ankle component, has embellishment
16:34
components. So I have ankle components and
16:36
embellishment components as well.
16:38
So it's not a really nice job of finding similar things that also
16:42
go with that skirt, right? I can then say, OK, I like these
16:45
shoes, but how high are the heels?
16:47
What's the price? What's the comfort level?
16:49
What are how do people review them and rate them?
16:51
So all of those in all that information, I can just ask
16:54
naturally, the next piece that I probably would like to show is
16:57
the ability for integration, because this is a nice
16:59
experience. But the next piece is, well, I
17:01
want to be able to add to carts, right?
17:04
So it's really simple the. E-commerce website.
17:08
This is an e-commerce website. Website that I would have and so
17:12
the images there are they catalogue images I as the
17:16
retailer provides. These are catalogue images that
17:18
are in your retail catalogue right now.
17:20
So we're keeping it very simple. I can do an example though of
17:23
visualisation, right? And I'm happy to show that.
17:26
And I'm doing that in a furniture example.
17:29
So the other piece is being able to search for multiple things at
17:32
once, right? You can go ahead and search for
17:34
I want purses and shoes to go with a skirt.
17:36
Do you have any tops? And to go with this, Kirk, all
17:39
of that is just very natural. So Azita may have mentioned a
17:42
furniture demo where you're visualising furniture in your
17:45
room because furniture is obviously considered purchase.
17:48
There's challenges to oh, I like it, but how's it going to look
17:51
in my room? The price point in that is
17:53
lower, right? But I'm not sure about it if
17:56
it's going to work and being able to play and actually
17:58
visualise it. So let's go ahead and let's
18:00
transition to that next. OK, so as a consumer, I used
18:03
your mobile app and I scanned my room to create a 3D room.
18:06
So we're going to go ahead and ask to pull that information up.
18:10
So I'm going to go on site. So what it's doing is it's
18:21
saying, OK, this is Cynthia, she's got a living room.
18:24
Let's take that 3D skin of the living room and we're putting it
18:26
into Omniverse and VS 3D collaboration platform.
18:31
So this is a, this is a 3D scan of my room.
18:33
This is not a 2D photo of my room.
18:35
Yeah, right. And so I have objects in here.
18:38
I have a couch, I have a table, I have a chair, for example.
18:43
And so the first thing is obviously I don't have a case of
18:46
luck, right? I want to, I want to now bring
18:48
change my room and make a case. So do you have any modern
18:53
couches? Let's start there.
18:58
And very similar to the fashion example, it's going to go back.
19:00
It's going to bring back some recommendations for me.
19:03
It's going to give me information about those couches.
19:06
And so here we have 3 couches, right?
19:09
I've got something that's very architectural.
19:11
I have something that's a little softer.
19:12
I like that one better. And then I've got a curve couch
19:15
that's interesting, but I'm not sure about it.
19:17
But the one I like that modern sectional is on the high end for
19:21
me in price. The one that I don't like, but
19:23
it's outside of my price range. And then there's the curb caps.
19:25
That's much more affordable for me.
19:27
So let me go ahead and explore it.
19:28
I might not necessarily explore it or be deterred that, you
19:31
know, because of price points at this point.
19:34
So let me go ahead and just explore replacing my couch with
19:41
the curb couch. Now notice I'm not giving the
19:45
product name or anything like that, right?
19:48
The generatory I, the agent is understanding it and then it's
19:52
going ahead and giving command back a tape.
19:55
I've got a database right, of the 3D assets for my furniture.
19:59
So I'm replacing the council was in there with this 3D asset of
20:02
furniture. So I can see it.
20:03
What is it doing? Oh, now it's really transformed
20:05
my room. It's making it more curved
20:06
instead of squared, but it's still not crazy about it.
20:09
It was right. We can do better, right?
20:11
We can do better. So let me try that other.
20:13
There's a couch that's kind of on the high end, but let me let
20:16
me try that instead, right? So here's something that that I
20:26
may not have explored it because of price points, right.
20:31
And here we go. Oh, I like that.
20:33
Yeah, OK, I really like that. But I am going to need something
20:37
round in the room. And I did see a table that I
20:40
liked again, you know, I'm using examples outside of my price
20:43
range or I just want to find something like this fast.
20:46
OK, OK, so I've gone ahead and I've I've uploaded the image.
20:51
Do you have a table like this? No, I'm not asking coffee table
20:58
like this nesting coffee table like this marble top coffee
21:01
table like this. So I think as you may have
21:03
talked to you about about the ability to search on images and
21:07
not only did it give me the coffee table, but it also gave
21:10
me some recommendations for chairs right as a part of this.
21:13
So love it. It came really close to what I'm
21:15
looking for so. And there we go.
21:24
All right, I love it. That gives me some of the
21:26
roundness, you know, for the room.
21:29
And you can imagine going further.
21:31
So let me go ahead. I think I talked about doing a
21:33
multiple query search. Let me show you what that was
21:35
like. Because now that I got my couch
21:37
and my table, right, maybe I want floor lamps, right?
21:41
And alternative chairs or something of that nature, right?
21:56
Now, more than likely in this scenario, it's going to give me
21:58
similar chairs back, but let's see what it let's see what it
22:01
does. OK, I came back with three floor
22:05
lamps as different examples that would work in this environment,
22:10
as well as one chair that it really highly recommended.
22:14
I'm not so crazy about the floor lamps.
22:17
I'm not so sure about the chair, but let me go ahead and explore
22:19
it, right. This is something that's you're
22:22
putting control in the hands of the consumer to explore and
22:25
visualise, to then say, oh, yeah, I do love it, or yeah,
22:28
that didn't work out. So I hope this is giving you a
22:30
sense of just the power of generative AI and AI agents for
22:34
retail shopping assistance. So I'm sure the next question
22:37
is, you know, how do I find out more?
22:38
So you come to nvidia.com and there's a full page on retail
22:42
shopping assistance. So we create an AI blueprint
22:45
that includes an AI agent. It includes Nemo components for
22:50
guardrails, a pointer to Llama 3.370, BA large language model,
22:56
being able to rank, being able to understand imagery, a an
22:59
accelerated database, vector database.
23:02
All of this is available in the blueprint for people to explore.
23:05
Try it out, Import your own product, catalogue information
23:08
and imagery, play with it, Start to change out components that
23:11
make sense for you, and then ultimately get to the point
23:14
where you might want to deploy it for yourself.
23:16
Where do people? Start a good question.
23:19
The place to start is sign up to be notified.
23:21
We have an upcoming package, Rev launchable blueprint, so it'll
23:25
be available for your teams to click a button.
23:28
It'll be unpacked and put on a clouds environment for them to
23:32
start starting with immediately. They don't need to know how to
23:34
download and install it. We're making it as simple as
23:36
possible for for your teams to get started.
23:39
You can also work with a number of our partners as well.