AI in Action: Learn from Amazon’s Playbook to Drive Innovation in Your Business

Hey everybody! Justin Honaman here. I lead Worldwide Retail and Consumer Goods Go-to Market at Amazon. I actually sit within Amazon Web Services. And I get — my team and I, well, coming out the industry and we have the opportunity every day to work with some of the biggest food and beverage [00:05:00] brands in the world, and actually some of the startups that are launching, and growing and really challenging their category. And so it’s a really exciting time to be in our industry.

And I’m excited to spend some time with you today. We’re going to be talking about one of the hottest topics of the year in AI. And I’d say it’s now over a year, but I’m telling you that this has just been a topic that’s on fire. And so I’m excited about it. If you don’t like technology, or you’re scared of technology, do not fear. The next 20 to 25 minutes, we’re not going to go too technical.

But I am going to give you a sense for what we’ve got here in the AI space. I’m going to give a little bit of background on Amazon, and then we’ll talk generative AI, and what you can do about it, so. And we’ll specifically put the lens on this of food and beverage, so it should be very relevant to all of you. And quite frankly, this topic should be relevant to all of you. Whether you know it or not, it will be something that influences your daily lives, in many ways. So [00:06:00] with that, I just appreciate you taking the time. But my question for you, as we get started is, what are you doing today, so you’re not obsolete tomorrow? Like, it — this is such a big deal.

Like, we here at Amazon, we’ve been looking at this generative AI, I mean, we’re thinking about this as big as laptops, as big as like the Internet being launched, as big as the lightbulb being invented, as big as the printing press. We’re thinking of not only AI, but generative AI is that significant. And we’re early days. And so I challenge you, again, if you’re not leaning in, please lean into this topic. You’re doing that here with me today. But just over time, stay informed of what’s happening here. A lot is moving very quickly. So many of you probably know Amazon, and thank you for those of you that shop on amazon.com. Thank you for those of you that are customers of ours, whether it be in Amazon advertising, or AWS, or beyond. But thank you.

[00:07:00] Our mission here is to be Earth’s most customer-centric company. In retailing consumer goods, actually at AWS, we actually have those under one industry vertical, if you want to call it that. So you can see the big pillars that we’re focused on here within CPG. Number one, helping you grow brands consumers love. Helping you innovate beyond the product, like how you go to market, how you do pricing, how you do planning, and then the supply chain, which has been super critical, especially the last couple of years.

On the retail side, we would like to say that we were born from retail. And as some of you may know, AWS was actually born out of retail, out of amazon.com in 2006. And for six or seven years, AWS was the only cloud platform out in the market. And so we’d like to say we’re born from retail, and much of our innovation comes from retail. A big focus on frictionless and Unified Commerce, especially over the last 12 to 18 months, as many of the retailers we work with have been progressing their platform. And then also helping retailers worldwide with their planning allocation, [00:08:00] store operations and whatnot flow.

Now, many food and beverage brands work with many parts of Amazon. So you can see here, many food and beverage brands sell through the marketplace, amazon.com. They run their technology business, or ecommerce business on AWS. They might be Amazon Fresh, they may be Whole Foods, Amazon, Whole Foods, our go stores and whatnot.

So if you think about our — the brands or portfolio of companies that sit within Amazon, you can imagine that a number of them come together to really enable a differentiating factor for our customers. And it’s a, it’s a big deal, when you think about how we can help accelerate innovation with a retail brand, or a food and beverage consumer brand.

So just a little bit about what we do here at Amazon. Some of you probably familiar with what you just saw, and some of you, maybe this was new, so. But either way, you all walked away with a little bit more knowledge on how we’re set up. Now, food and beverage is interesting, right?

So if all of you are in food and beverage, then this page will make a lot of sense. But these are the challenges, or even the — not even challenges; these are some of the topics we have to be aware of every day when working with the food and beverage brand. Everything from like shelf lives and expiration dates, to label requirements, to an alcoholic beverage space. Like, how do you go to market? DFC. Three-tier distribution through alternate routes to market. We’ve got beverage brands with bottlers, and maybe they also manufacture product. We’ve got cold chain considerations.

And on the innovation side, you’ve got a consumer that’s shifting right, and preferring certain products over others. So it’s an interesting space to be in, in food and beverage, because it’s added, I don’t know about complexity, but added considerations as you’re thinking about route to market. Right? And this is the world that I play in every day, and you all are in to get — and we’re all in this space together, navigating this.

So a pretty exciting space. Now, I don’t know about you, but our world, the last 12 to 14 months, has been just all day, most of the day, many days, generative AI, and all the talk about AI. It’s been incredible. I mean, like I was saying, once we were coming out of COVID, what a great time it was to be back in retail and consumer goods, and seeing the businesses grow.

And man, what a great time to be in technology in our industry the last year, year and a half, even before that if you count clouds, cloud computing and whatnot. But this has really been a gamechanger, an eyeopener. You can see the types of news that’s out there, the size of market here from McKinsey, the investments being made by Amazon, for example, in companies like Anthropic, and I’ll talk about what that is in a few minutes.

You’ve probably seen lots of news around, like, how quickly this technology has been adopted. I mean, just amazingly, how fast we’ve gone to 50 million users in the large language — a public large language model. And I’ll explain that a few minutes. And there’s also tension. Guess what? Many [00:11:00] food and beverage brands, and CPG brands, and retail brands, don’t make it. In fact, they don’t innovate. They get stuck in their own ways, and they can’t move quick enough to stay relevant. And 50% of the S&P 500 companies will be off the Index in 10 years.

I mean, that’s crazy, right? So there’s a lot of tension in the market, even though there’s excitement around new technology. Eighty % of companies, you can see for our partners at Accenture here, did some recent research with us. They’re accelerating their efforts around Gen AI. Nine % are outpacing because of it around revenue growth. And it is what most of you, as the number one cause of change. And that was in 2023. I would just say that will, and is continuing in 2024. And many of you are realizing that. What else are we all experiencing together? Right? This whole concern or question around jobs changing, or upskilling needed. Fifty percent of employees will need to be rescaled in the next five years. I think it’s actually higher than that. In fact, it’s probably more closer to [00:12:00] 70, 80, 90%, based on what we’re seeing every day with our customers.

Roles are changing, and jobs are changing. Right? And on top of that, if you think about it, there’s a big need around this space for data. Once again, data is important. And if you’re not in technology, and you’re not in the data space, you should consider yourself soon to be in technology, and need to understand the data space, because it’s going to be critical to be able to do some of the things that we’ll talk about in a few moments. If you’re not familiar with the definition of AI, and this is — again, we’re not going level 200 or 300 here, but this is a good page to screenshot, or write down, if you want to, in terms of just some definitions.

But you know before ChatGPT came out, and everyone started talking about large language models and generative AI, you’d hear a lot of bantering around, we’re using AIML. Hey, we’re using AI or ML, but nobody really understood what that was, except for a few people. [00:13:00] And so the definition here is very simple. AI is all of this. Okay? Machine learning is a subset of this that focuses on patterns. There’s deep learning, which we won’t spend much time on today, but that talks about neural networks. And there’s generative AI, and that’s where you — you’ve heard some of these phrases like large language models, or foundation models. And we’ll jump into that shortly.

So the questions that our team, with you all, many of you are customers, have been addressing regularly over the last year, almost year and a half now. Or, hey, why is this important? Last year, every day, I was on calls with customers, and those that wanted to work with us, that hadn’t been working with us, with their CMO’s, their chief supply chain officers, their chief legal officer. So not just the CIO and CTO that you might expect talking to AWS, the board of several major brands. Like, incredible. Right? Think about how big this is.

But what is Amazon doing about it? What makes us different? What do we offer? What can we do with images, and text, and coding. And what are these models, and how do we even differentiate? And I’ve been living that with you all. There’s been new roles defined over the last couple of years. Right? And there’ll be new roles needed. Some of these roles didn’t even exist five or ten years ago.

But the real tipping point for generative AI has been a couple things. Number one: large volumes of data. Some of you been around a little bit. You remember big data, when that was a big deal about 10 to 12 years ago. Well, that has scaled. Cloud computing, right? AWS was the first in the market. Others have now entered. But that’s what we call “scalable compute”, so you can spin up capability quickly. And then machine learning, and the knowledge from people, and also the capability around machine learning has really scaled.

Now, it’s also interesting, another thing that we’re all facing together, and you guys are probably living this, right, your companies, like how do we control this? I mean, you can lock down my laptop, but I still got my phone here, and next to me here, a personal laptop. Right? Like, you can’t lock those down, and I can use those with public large language models. So what’s interesting here is the technology is moving faster than all of us, as individuals, can keep up with it. It’s moving faster than businesses can keep up with how individuals are using the technology.

It’s moving faster than public policy. And so hence, why literally, daily, for those of you that follow the markets, and do reading of news and whatnot, you know that that’s why there’s angst and some concern around AI, and how do you put guardrails around it? Or is it too late. Right? The technology is out. So an example of this is like GDPR over in Europe. People were collecting personal information, then they were using that personal information to market to individuals. Businesses started marketing to individuals because of that information. And then the policy got put into place. And the situation here is moving much faster, and many different entities are trying to figure out what they should actually be doing.

So, but we’re all in this together. I hate to say that phrase, because it sounds cliché. But literally, like, whether I work at Amazon, or you work at major food and beverage brand, we’re all what I just showed you has been swirling around last couple of months or years. Now, there’s some interesting things to think about, as you think about the — what we want to call “the AI-driven organization”. Now, I think this is aspirational for most. Right? And there’s reasons why we’ll get to in a few minutes. But these are some of the things to be thinking about, as we — as you think about your own strategy and vision.

But data is being a big part of this. Being about to innovate and move quickly. That’s a challenge for many big brands. Right? Personalization at scale, for those of you that have a mechanism for engaging directly with customers. And then how do you move the needle with people, and reskilling? So we’ll dive into that in a minute. Okay. But before we get there, a couple datapoints. AI is not new. Like, it’s been around since the 50s. And it’s actually been used for being used by many of your organizations today. Right? And we’re using it, for example, in product recommendations, and managing fraud-free commerce, and contact center solutions, in image analysis.

This is not new at Amazon. Actually, AI is built into many parts of our business. I mentioned it’s built into amazon.com today. If you’ve been to one of our fulfillment centers, it’s built into the robots that move product around, like those fulfillment centers. Pretty incredible. Alexa devices, and many different machine learning models running there. Our “Just Walk Out” technology, I’ll talk about in a few minutes, where we’re using computer vision and sensor fusion, which is an AI model to build your virtual cart.

So it’s part of a lot of what we do. A couple of other places, where you see AI, traditional AI, okay, we’re talking just traditional AI, like live streaming. This is amazon.com/live. You can go there now. There’ll be a different curator than the one you see on the screen here. But it’s an example. It’s powered by that. Virtual Try-on.

This is already live, not only on our platform, but also many others in terms of the ability to check out and try different products before you buy. AR and VR, of course, powered by AI image search. If you like my shirt, you can take a screenshot of it, and actually use it to find others out on the Internet that are similar. And that’s actually something that’s evolving quickly with generative AI.

But these are all traditional AI examples.

Virtual stores. CPG brands are actually looking at some of the virtual store concepts. “Just Walk Out”. So if I if you were in the room, I’d say how many of you been into Amazon Go or Amazon Fresh store? Some of you would raise your hand. The way that it works is you see the guy here walking in. But in your Amazon app on your phone, at the top of the app — if you’ve never been in one of these, you look right now, it’s cool.

If you look at the top of your app, there’s a thing called “a store code”. And you get this code, and this is the code you see him scanning as he’s walking into the store, right, as I walk off the camera. He picks up a product, and just walks out like he didn’t — it looks like he’s not paying, but he is. There’s cameras in the ceiling throughout the store. We’re not using facial recognition. We basically track where you are in the store, and what you pick up off of the shelf, or what you put back, and that creates your virtual cart.

That uses AI. Traditional AI. The same thing with Amazon One. This is basically you use your palm for payment. And so AI is built into the machine learning algorithms behind the scenes that enable that. Again, you enter just like this person, pick up what you want, and leave the store. Alexa, in grocery stores, for wayfinding powered by AI. Retail media, of course, a hot topic. We could spend a whole hour on this if we wanted to. But powered — many of that powered by AI. And in the last example, just traditional AI is heat mapping a store. For those of you that work around the store execution, to be able to get to use data to know who’s shopping — or not who’s shopping, where they’re shopping in the store. This is a great way to do that.

You can use traditional cameras in a store, and power that through an AI model, and get some good insight. So, well, what about generative AI? Right? That’s the big hot buzzword. I have a bell here that I rang like every day. Every time that someone wanted to have a meeting on generative AI last year, it was like, yeah. And then some days it was like, I mean like crazy. So generative AI, right, it’s — hey, here’s some examples, right. It’s built into our platform today on amazon.com. If you go, you’ll notice that there’s sometimes thousands of reviews for products.

We’re using a text model to summarize those reviews, and allow you to go and dive deeper. Right? It gives you insights on what is happening, or what people are saying about your product. Now, you can do this today, on your ecommerce platform. I have many customers that have hundreds of reviews on their own ecommerce platform, and they can’t read through 400 reviews, let alone 4,000. You can have a model running, and accept data immediately, like today, right? You don’t have to go buy something or download something. It’s available now. Right?

As one of our customers, [00:21:00] you can just log in and start using it. And you can upload all those reviews and start writing the model against it, and get some insights into your data. Like, if you’re not doing that missed opportunity, like available right now. Okay? Another good example of how on Amazon we’re using, it’s called “Diffuse-to-Choose”. And you can see here, another way of doing virtual try-on. But if you circle the spot you saw on the bottom, all it is taking products and putting them onto a model, or your own picture. Kind of cool.

There’s lots of work happening on images and the ability to manipulate images. This is all very real, very accessible today. The ability to take furniture and put it in all kinds of different environments. Done. You see this word “prompt” over here. And see this the words after it? For those of you that may be new to generative AI, this is a function of doing generative AI. So prompting, or prompt engineers, or like the ability to come up with really good ways of giving the model clarity on what you want to have happen. That is [00:22:00] essentially what this is. And I’ll show you some examples. But just know, you’ll be hearing more about that.

What else at Amazon? We rolled out Rufus. It’s a conversation way — conversational shopping experience. And you’ll see this continue to grow, not only with us, but I think more broadly as you think about new and different ways to interact with products. Now, many of our customers are already way — are well into testing out generative AI. These are, these are just some reference examples. But one of the customers wanted to build concert photos of people at a concert holding their beverage products. And these are not real people. Using our image models, you can do that immediately today.

Again, if you’re marketing, you’re in brand management, you’re in category planning, you’re in shopper marketing, shopper insights, you’re using an agency to do all this, you could be doing it — a lot of it yourself, and a pretty quick rate in terms of being able to test it out and try it out. And then the quality, look at [00:23:00] the quality of the image here, the eye in the middle.

In terms of product innovation, some of our customers are using this as a way to come up with new image, potential product, new images of products that could be potentially sold, and so, and then driving that into the pipeline. You can take a manual, for example, this Coca-Cola freestyle machine, and upload that. So your contact center could provide support for that, that manual or other equipment that might be out there, versus having to go through a thick manual. I mean, our models can run through that in seconds, and then be able to answer questions about what’s wrong with the product or manufacturing machine.

Planograms, we’re doing some interesting work there. I think this is super interesting. For those that are not familiar planograms. It’s a very manual process today, and there are some tools that you’re able to create the planogram and push it out to the market. But they’re not very dynamic. And so imagine if you had — were able to take the inputs from sales data [00:24:00] and out of stock data, and sell through data, and then be able to provide dynamic planograms, like, on the fly individually to source that. It’s very possible with generative AI.

I can keep going on the list for you, but just powerful use of images, texting and coding. And here’s where our customers are finding value today; these four areas. So I just gave you a couple of examples on the previous page. But these are the four big buckets. So I covered some, like, on the creative side Like, listen, if you’re doing ecommerce listings, bullets, AdWords, descriptions, titles, you can be doing that today with our text models, and it’s super accurate. Summarization, I showed you the example available now. Coding. It’s unbelievable. If you’re a coder or programmer, taking one set of code, and doing QA on it, or moving to another set of code in seconds. I mean, that’s incredible.

And then document [00:25:00] processing it. So for many of you, you got sourcing and procurement teams. You’ve got contracts. Like, they’re all text, right? So these models can run through those very quickly. The use cases start to become very amazing when you think about them. We have several customers looking at field sales applications for leveraging generative AI for recommendations when you’re walking into a store. I mean, the list goes on. So if you’re thinking about use cases, here’s some things that you can take and leverage yourself.

Because sometimes — actually, not sometimes like how would you even know what would be a good use case? You wouldn’t. So here’s some examples, where our customers and retail are finding a lot of values in the center section here: marketing, sales and digital commerce. And on the, on the consumer goods side, similar, but also in the route to market, retail execution, execution and market, and sourcing and procurement. The areas of supply chain, like logistics, planning, and [00:26:00] manufacturing, there are some good traditional AI use cases, but the generative AI elements have yet to catch up. I know many of you are in, in food and beverage, so like I mentioned some of your roles earlier.

But if you’re in category management, or shopper marketing, or allocation and planning on the store side, like, you’re — you’ve got massive spreadsheets, right, pivot tables, that you’re manipulating every day, or dropping like Nielsen IQ data into and whatnot, right, for leveraging these models now. Like, why not try it? I think you’ll be amazed at how quickly you can upload a spreadsheet and then run these models against that spreadsheet, or the data, and start mining it for information. And then you become better and better at asking the questions. So something to think about.

That’s not separating it, that’s summarizing it. I need you to take the original transcript and not change anything other thant break it up into 3-4 line paragraphs to make it easier to read on the page. Do not change a single word or letter of the original script.

You can scan this code if you’d like. There’s some other great use cases here. I telling a big group last week, I said, if you go back to your team and say, “Hey, team, give me some generative AI use cases, we’re ready to start testing.” I think they look at you like, “What do you mean? How would I even know?” Or they’d say, “Yeah, I mean, my sales reports wrong, or the data coming in is inaccurate.” Right? It’s not generative AI.

So these are some use cases, but just know you’ll probably find that they’re a mixed bag. And like anything else, in order for it to work, yeah, to get the data, right. So okay, so what can we, what can we be doing to help you? First of all, and I’d encourage you to get going, and get started on an experiment. And now, if you’re a line of business person, you’re like, “Yeah, that’s IT’s job.” No, it’s your job, and IT. It’s not IT’s job. And it’s not IT’s job to figure this out. It is purely a joint effort from both. So do not say, that down the hall, IT solves this. And I’ll show you why.

But as you’re thinking about, like, prioritization, guess what? Line of business is key from the business value. You got to be able to be involved in the process. You should put in place some tenets, and things around like how to think about AI, and, like, what’s AI going to — how it’s going to be set up within your organization, right, in terms of the structure. There’s some things to think about. There’s governance. There’s how should we leverage the data that’s available? How should we think about use cases, and prioritize those? How should we monitor the use of those? There are skill sets that everyone needs. I don’t mean like some people, but, like, everyone.

And look, a big part of this is going to be understanding data, and like prompt engineering, and design thinking. There are people in your organization that do not have these skills, but that’s going to force them to change or upskill. Right? And the tech side, there’s some things here.

There’s parts of the tech stack that don’t sit in your IT organization. But in order for you to use the generative AI models, or to apply them, you’re going to need some of the skill sets that you see on this page here. There’s also a lot of things to think about. Right? From an HR perspective, do you hire new? Do you recruit in? Do you upskill? Reskill? Move people around? It’s an interesting tension and challenge, and you’ll see as I think you find the use cases that can be leveraged. And as you think about upscaling, these are the types of things to think about, as you’re moving people through the model, right, or moving them through the model of leveraging generative AI.

Now, there are also some concerns, right, some of you’re like, “I know, Justin, this is interesting, but I’m like worried because I’ve heard about bias. Like, who owns the pictures that you showed me? What about hallucinations?” I hear these thing — hallucinates. I hear it’s wrong a lot of times.” Yeah. That’s all true. Sometimes. But it depends. And you may not know when it’s wrong. Right? It’s interesting. And so this is a moving space. It’s changing weekly, if not biweekly. In fact, the content — some of the content you’re seeing today, I only updated last week.

So we — and last year, it was changing like every other day. But yeah, there’s many things to think about. We can help you think about some of these things, as we’re already working on this with many brands worldwide. And for some of you, you’ll be able to move faster on this because of the access to data. Right? That — and for others, you’re going to struggle because you don’t have access to the data. But there’s things to think about, your — certainly your legal team will be involved in figuring out, like, what do you do with images once you create those, and who owns them? And how does it get leveraged. But there’s many parts of this. And so — and there isn’t really any stopping this. So you can say, well, it’s not a priority for right now, but it’s going to

One [00:40:00] will be just built in. So if you use Zoom, or Chime, or Teams, or pick your favorite platform, it’s already there, right, in your office applications, in Adobe, in SAP, in Salesforce, like built in. You may not even have to think about it, right? It’s just being a capability. Like, I didn’t know how this does this.

But it’s like, when this started filling in the blank for me, on my [inaudible 00:40:26], you guys notice how sometimes it starts filling in the blank for you on your text messages, like that’s generative AI actually running. You didn’t ask for it. You don’t know what model is running behind the scenes, but it does it, right. That’s one use case. The other is the, I’ll say more bespoke. I have a use case. I want to, for example, I want to revise every one of my product descriptions, titles, bullets and AdWords, on my own ecommerce platform.

I’m going to use a text model with — at Amazon to do that for me. I’m going to do that in seconds instead of hours [00:41:00] and days waiting for an agency to come back and pay the millions of dollars. Like, I can do that in-house. So, like, that is a specific use case with a model, with data, and then an outcome or output that you can then use.

And so that’s — you see both of those playing out right now. And I think we’ll see that continue to accelerate in the next couple of months.