ChatGPT: Do These Transformative Consumer Goods Use Cases Outweigh Concerns?

Generative artificial intelligence has become a contentious topic worldwide. The applications are still blurry, as are the ethics involved with the technology. What the consumer goods industry has come to understand, however, is that there is no ignoring it. Generative AI platforms like ChatGPT, Open AI, and others are here to stay, will likely transform the industry, and must be embraced to proactively affect change in a responsible manner. In this edited webinar transcript, Albert Guffanti, VP of the Retail Technology Group at Ensemble IQ, leads a discussion on the topic of generative AI with AWS experts Michael Connor, global CPG leader, and Justin Honaman, head of worldwide retail and consumer goods, go-to-market. Connor and Honaman shed light on the real-life applications and use cases we can expect moving forward, and what today’s CPG executives should do to stay ahead of quickly shifting trends in the ChatGPT space. Albert Guffanti: Hello everyone, welcome to today’s webinar. My name is Albert Guffanti and I am the VP of the Retail Technology Group here at EnsembleIQ and publisher of Consumer Goods Technology. We're thrilled for you to join this important discussion entitled, “Cut Through the Clutter: Decoding Conversational AI and ChatGPT.” As a publisher and a media brand, our job is to help consumer goods executives navigate the micro-trends happening in the industry and help understand what that means for both the business and the future. ChatGPT is one of those phenomena that became a micro-trend, then immediately jumped into the everyday lives of people and businesses, becoming a source of question, mystery, and opportunity for many of us. This webinar will uncover what ChatGPT specifically means for us as consumer goods executives and how to leverage it in business and get through all the mystery. We have some amazing speakers here from AWS, including Michael Connor, global CPG leader and Justin Honaman, head of worldwide retail and consumer goods, go-to-market. We've been talking about ChatGPT, most recently at our League of Leaders meeting where one of the sessions was focused on responsible AI. Justin and I were talking about bringing CG executives a timely webinar that takes the topic of AI, uncovers it, and peels back the layers. ChatGPT, as we'll talk about, is one of those game changing moments in the industry, so much so that some experts are saying it will be the equivalent of what the internet was when it was first launched, and how that has affected all of our personal lives and our businesses. With that, I’ll turn it over to Justin. Justin Honaman: Thanks, Albert. I’m excited for the conversation today and appreciate the partnership with CGT and RIS. Don’t be afraid if you’re not in IT or not technical; do not be afraid of today's discussion. We're going to simplify what has been a couple of months of excitement, energy around AI, and the topics of AI. Today, we’re going to give a little bit of AI and ML 101, so when talking about artificial intelligence and machine learning, you'll have the basics of it. We'll also jump into this idea of generative AI, large language models, and what that means. Then, what are the applications in CPG? At the end of this, you'll have these three things and will be able to take this back to your organization and be a bit more informed as far as each of these topics. For those that know Amazon, you probably know us from all different mechanisms, whether it's something delivered to your home or working with AWS in the technology space. Our vision is to be the Earth's most customer-centric company. Within consumer products – the industry we're focused on today – we're focused on helping customers. We've worked with the vast majority of CPG brands, helping them grow brands that consumers love and innovate. That's one of the big topics for today. Then, the last piece is around operational efficiency. We're good at supply chain, which is a big focus of ours in the industry. For those that have been around data and analytics for a while, you might think this is a model. In fact, the vast majority of the major CPG brands we work with worldwide are still trying to integrate data across multiple ERPs, make sense of POS data coming from multiple retailers, and make sense of syndicated data that might be linked on top. That's a huge data issue. Most of the brands we work with are focused on that, let alone AI and machine learning. If you've been around retail and CPG analytics the last 4-6 years, the model has been to become more data-driven. All of a sudden, in months, AI has taken center screen. On LinkedIn, everybody is referencing AI; at the upcoming ShopTalk and Analytics Unite, everything now has AI as part of it. This is important, it's real. This is not just a concept – a buzzword, flashy and interesting just to talk about – it’s real. Here's the definition of AI: A technique that enables computers to mimic human intelligence using logic, if/then statements and machine learning. Machine learning is actually a component of artificial intelligence. But guess what? It's not new. This has been around since the 50s and 70s. This has been an evolution over time. What’s happened in the last couple of months is a rapid acceleration in AI: the applications of it and the use cases of it. Last year, there were many different focused discussions on responsible AI. Then, here we are with the topic of the hour, and this idea of generative AI. Michael Connor: This has been such an interesting space. I've been involved with AI for about 20 years and have never seen anything evolve so quickly. I talk to customers about this every day, working with customers in CPG and retail. Everyone can tell something powerful is happening; they know this is important to business, but aren't sure exactly how. The focus of today's conversation – we spent a lot of time looking at news and evaluating new products and models that are coming out – we’ll bring you up-to-speed on the most important things happening in terms of the news, provide a basic understanding of how the technology works (it’s fairly straightforward), then go into key use cases. People see this stuff coming out, it looks powerful, their kids are using it for school, but no one knows how it'll affect the industry. We're all focused on CPG and retailers, and have put a lot of thought into and worked with a lot of customers that are in production on this. We want to provide a good awareness of where this is happening and how it can help the business. The music that just played was Drake and The Weeknd, it was an AI-generated song called “Heart on My Sleeve,” which came out the last few days, and generated millions of hits. It was purely AI-generated, so I'm sure it gave the record label a heart attack. A lot of people have seen some of the deep fakes out, with the Trumps, the Pope with the puffy jacket, and we’re all wondering how to trust any images or videos anymore. Sony had a photo competition recently, and the submission that won was AI-generated. The person who submitted it did it to prove a point, then rescinded the submission, which made headline news. Nvidia announced video generation capabilities, which isn’t there yet, but we're getting really good at photos, and video is close behind. It took years for Netflix to get to 10 million users, Instagram was faster, ChatGPT hit 10 million users in a single day. This is moving really fast. Sundar Pichai, Google's CEO, issued a code red internally. He saw this as a fundamental threat to their business model. Of course, Google makes a lot of their money with advertising, YouTube, ads, and also search ads. The other day I went to Google to look for a recipe for a smokey old fashioned, and I had to wade through 22 ads before I found it. Of course, with these large language models, I type in “smokey old fashioned recipe,” I can even give it some guidance, and it comes back in milliseconds without any ads. It's a threat to a lot of the business models out there that are based on advertising. Elon Musk, and a lot of other leaders, signed a paper a few weeks ago calling for a pause on AI. They're concerned about profound risks and are asking for a six month delay to think about the societal implications. Of course, OpenAI's CEO shot back and thought that was irresponsible; they're leading a lot of this charge. Italy came out and banned ChatGPT. China quickly followed, saying, "All AI workloads have to now have approval from the government before they go into production." France and Germany are also debating bans. We're starting to see governments react to this. We get into discussions with customers about human bias and harm that models can create, etc. Ethics and societal implications are always the backdrop of these conversations that we're having. Midjourney is where you can type a paragraph of text and it generates photographs based on your direction. These photographs are now indiscernible from real photographs. Of course, Amazon, I feel like we were late to the game. We've been using this internally, but Google, Microsoft, and Open AI got in quickly with their APIs to use this technology. A few days ago we announced our generative AI models with Amazon Bedrock – we're finally coming with a product that's comparable. I follow this gentleman on Twitter, and this was, "A Tuesday in AI." On that day, Google opened up their large language model, which is a competitor to Open AI. NVIDIA announced their tools. Adobe announced Firefly, an AI image creator, and their claim to fame is that all the images are licensed, where some of them are not with other tools. Microsoft announced the Bing Image Creator, and that was all before 11:00 a.m. We're seeing this Cambrian explosion. The Cambrian explosion with biology is when all of these fossils appeared in history in a very short period of time when there was a massive explosion of life that evolved. We're seeing the same thing happening with OpenAI, large language models, and generative AI in general. Auto-GPT and Baby AGI are two projects that are available on GitHub. Both are open source. Auto-GPT I've been following, they got 40,000 stars on GitHub within a few days. Normally, when I use OpenAI, ChatGPT, or large language models in AWS, we're submitting a question and asking it to respond or complete something, but the human's in control. Auto-GPT turned that on its head – you give it a directive like finding pricing for competitors’ products, and it submits it to the large language model. The large language model creates an overview of how it approached the problem, then generates a plan: Step 1, 2, 3, 4, 5. It also generates some concerns and risks. Given that goal, it will go out and start hitting the internet. It uses Google search APIs, downloads information, pulls images, details, and scrapes the net. It's the idea of this inversion of control. Instead of the human being in control and asking questions, you're giving it a goal and the bots are in control. It goes out and works tirelessly to try to meet that goal. Then, when it meets the goal, it's done. This is transformative. It's early days. Sometimes it works well, sometimes it doesn't, but we see how quickly this is evolving. It's a fascinating space. Last week, I did a talk about 2030 and what life looks like for CPG and retailers in 2030. William Gibson said, "The future's already here. It's just not evenly distributed." A lot of the tech that will be prevalent in business in 2030 is already here, there's just some people that are already starting to use it. The way to identify this technology, bring it into the company, apply it to workloads and business problems – that's more about culture and creating mechanisms and leadership in the company to identify innovation and move on it. That's separate from the technology, but that's a different discussion. A lot of brands and people we talk to are funny about imagery. I worked at Coke for 10 years. Think about how much care we take – the right models, photographers, lighting, and everything – it's got to be on brand. When I saw some of the original image generation, I said, "I don't think this is going to ever be ready for brands," and I'm eating my words. Looking at v4 and v5 of Midjourney, v4 is the last version and v5 is the current one. It's subtle, but v4 looks a lot more like a video game. In v5, we start to see these specular highlights, the shadows are more realistic, it looks like a photograph. Lifestyle imagery, product shots, this is ready for primetime, or really close. You can see how quickly this evolved. About a year ago, if you went into the tool and said, "Doctor on a plane using wifi," this is what you got. It's creepy looking. Fast-forward to v3 about six months ago, "Doctor on a plane using wifi," here you see this doctor sitting in Comfort Plus with access to the flight controls, which is disconcerting. You think it will never be ready for primetime. Now, v5, you have to wonder what version 10 will look like. When we anticipate where the technology's going, we have to think about how we're operating internally and start to plan for this. When I was at Coke, we had something called a Design Machine. All the POS materials and imagery in restaurants, stadiums, etc. was in there. It’s a huge archive of stock photos in imagery that showed pairings and things like that, pricing, trade promotion. Massive amounts of money went into that. This will be disruptive in that sense too, and allow brands to move quicker. You're generating these in seconds. Let's focus on a large language model. ChatGPT is a large language model. These have been out for a long time – Facebook uses them, we use them, Google. What ChatGPT and OpenAI did was democratized it and got it out on the internet for everyone to use, which was super cool. They work by giving it text input and direction on what you want, then it goes through an AI model (the brain), which is trained on what we call the pile, and generates text to output. The pile is basically every document we can scrape from the internet. It's every Reddit post, Facebook post, Tweet, everything that we can hoover up that's been out on the internet, and it trains this large language model. It has really good insights, knows 38 languages, and has a huge archive of all the information on the internet. When you give it a bunch of text, it takes those words (called tokens) and starts generating the most likely next token. That's a bit of an oversimplification, but not by much. It keeps on doing that. What's the next token that's most likely? This is one of the fundamental leaps in technology of our time. A lot of people are laughing that this is just like bitcoin, NFTs, and the metaverse – it’s just another fad that's going to pass. I beg to differ. I have seen very few worthwhile use cases for those technologies and environmental implications. There’s already a lot of people using this in production. Each one of these technologies builds on itself. The internet gave way to cloud computing, which created this compute power and capabilities for startups to do innovative things, along with AI. Mobile and edge computers. We have these devices in our hands, which give us the ability to put every photo, every text, all this information online. General intelligence depended on all those other technologies in order for it to be here. Most companies, the way Amazon uses AI, have very specific products to help with coding or for language transcription. When I use the term “general intelligence,” we're saying this technology has evolved where it can solve a wide range of problems. Generally, what are these things good at? Talking to humans, understanding intent, feelings, taking information and creating a narrative, summarizing text, translating from one language to another and even one language to code or to SQL. Writing and debugging code, converting speech to SQL, sentiment analysis. Sentiment analysis used to think if this tweet was negative or positive. I have a passionate project where it helps people with emotional intelligence, and generates nuanced assessments of feelings, 100 different human feelings. It's much more nuanced and subtle. It understands that someone is frustrated, feeling hopeful, or they're excited – those types of things. It's gotten a lot better. Answering questions. A lot of brands have compliance where they have to answer nutrition questions in call centers. It's really good. Let's get into some use cases that are emerging. All of these are sanitized, a lot of them came from my brain while talking to different customers, but I want you to see how these are being used. These are illustrative. CVS. I'm saying, "you're an empathetic health assistant that helps users find a product from CVS. It can solve health issues." Then, you give it to the consumer. Imagine on CVS – cvs.com or the mobile app – the consumer can ask a question and the AI model will give a response. The interesting thing about this response is, originally, I did not have the empathetic word in there and I was just saying, "Help customers with product," and it quickly just rattled off products. Then, I added empathetic, and it comes in and it addresses what the person's going through, which is interesting. When I added empathetic, it started recommending lifestyle changes rather than product recommendations. If you look at what this is doing, it sees that they're looking for a low cost product. It's combining a few things – the fact that it's CVS, the products that are sold there – it's looking at low cost and acid reflux. It starts by recommending Tums or Rolaids because those are low cost and arguably start with the least powerful drug available. Then, it goes into CVS Health’s acid reducer tablets, which is more of a generic, lower cost, and then it recommends a brand. The interesting thing happening here is, if you can compare this to Google, it's doing some very interesting things: Understanding context and intent. It knows what I'm asking it to do and understands the intent of what I'm asking it to do, but also the consumer. Reasoning. If I go into Google and type in a keyword, it's like Go Fish, playing a match across all these documents on the internet. This is different. It's pulling information and knowledge from disparate sources, and bringing them together in a powerful way. Personalizing the result. How do you want that information? You can have it generate images, be it data in the form of JSONs or CSVs, it could be lists. It’s very specific to what you want. Traditionally, search engines give you a bunch of ads and a bunch of results, which leaves you digging through sites trying to find the text. It's very powerful in terms of providing back what you want. When I was at Coke, we had 750 brand products in the United States alone, 207 countries, across 30-plus languages. A lot of times when we were generating product descriptions, campaign management, sending emails or text messages, we wanted that message to be intimate to let the consumer know that we cared about them, that we were communicating with them in a nuanced way that showed we knew them and valued the relationship. We're always talking about brand voices. Every brand has a different voice. For instance, Coca-Cola is much different than Monster Energy. A lot of customers start to generate product descriptions. I asked it to generate this fabricated product description for males 13-24 that live in the U.S. Southeast that likes NASCAR. It talks about how this product would be great at the track, cruising with friends, and it solved a problem that we needed to solve at Coke. But here's what I got wrong. We talked about people in terms of segments, consumer segments, because we didn't have the ability to not understand them intimately. All these brands are talking about consumer data platforms, they have more data than they had before, and we're still looking at people as segments. Now, as a brand, I have the ability to generate a product description for this specific person given what I know about them. When you're giving people merchandising, product descriptions, product recommendations, content recommendations, what you're going to see over the next few years is the opportunity to make that very specific to the person you're talking to. When we're talking about product descriptions, this is a real customer. I’m using Nestle here, but this is not the customer, I wanted to protect their privacy. They had 300,000 products in their database and looked a lot like this, needing to clean that up and they were overwhelmed. The large language model was given some information to clean up the product information, make it easier to read, remove all these crazy HTML characters and tags, and special characters. There was weird spacing, things were duplicated – it cleaned it up for us. There's a couple interesting things to note here that blew me away. One is the word “ispirazione.” I don't think that the intent of the brand owner would be to remove that. It's in Italian, it means inspiration, but because the model saw this as in English and it wants to make it easier, it converted that into English. The other thing it did was recognize there's a 10-count and a pack of five, and actually did some math and said five packs of two. I don't know if that's right, but there's interesting things going on there. When we presented this to the enterprise architecture team, they asked why they wouldn't write a regular expression. First of all, to do this in the translation is sophisticated. It requires software developers. They're busy. You've got to get on their roadmap. You've got to go through Jira board, QA, etc. This is something an intern can do in hours. What we're seeing is this democratization of technology, where things that required computer scientists can now be done with someone who's in an agency or the brand. It’s super powerful. Give it some code and it can find if there's a bug in it. Not only does it tell you what's wrong with it, but it'll rewrite the code for you. We have something called Code Whisperer, which will fundamentally change coding forever. Looking at how this technology will change data and analytics, there are a lot of brands that have a tremendous amount of data. They've got their data warehouse, but that doesn't translate into actions. The reason for that is you have people, brand owners, or people in R&D or finance that don't know how to get at that data to understand it. I led a data science team at Coke. A lot of the people that I was trying to support couldn't go into Tableau and make sense of it. Their brains didn't work that way. To give a salesperson, brand manager, or category manager the ability to query that data in natural language using English by asking: "Tell me how my sparkling beverages are doing in the southeast in the e-commerce channel over the last month. What are the top cohorts that are purchasing Coca-Cola?" The ability for the large language model to understand what you're saying, go and find that data by translating it into SQL, but then tell a story on the way back. We always failed to tell a story with the data to realize there was a big surge late at night at the drive-through, etc. That's really powerful. What does it mean? What's the narrative? People need that story, they need that narrative. Otherwise, the data is just millions of rows and it's hard to parse out. Even if the model can tell you what you should do about it, or that customers are at risk and here's some recommended actions. This is going to really transform that. Honaman: In the CVS example, the FDA requested that the product be pulled from the market. Why would the AI include that as part of its recommendation? This speaks to the fact that it's not perfect in terms of how the AI is working. Connor: That's right. I don't know if we'll have time to get to some of the pitfalls and things, but with AI, there are some problems. One is, it will lie, it gets things wrong. And that's a good example. What happens is, it has all this information on the internet, but it may not always be timely. What we see brands doing is putting some really cool technologies on top of the AI – Lang Chain for example – where you give it a list of the products that are available. What you can do is put some guardrails and controls on top of the model in order to give yourself that control. You can also train it. With training, you can tell what products are available. You're seeing the power of what's available out of the box. That's a great point. Almost all brands are going to want to do something where they take those results and programmatically look at them, or do some fine-tuning and training, in order to make sure the things that are coming back are accurate. It's something that every brand is asking right now. One of my customers asked about compliance. When we make claims about a product and give that to an agency, we typically have compliance departments. Can these models go and do some of that compliance for us? The legal teams are going to have some heartburn over this, we've got to figure out how to work with these models to address that. I'm going to blow through a couple of these – help with merchandising. A lot of times you post a product, consumers come back and ask a lot of questions. Given this product description, it's telling me what all the questions people will likely ask are. Is it organic? Is it non-GMO? It can generate ratings, etc. Marketing teams are using this for social calendars. Someone was using this to generate avatars for her cereal brand, where each avatar had its own brand voice. This is one of those pitfalls. These large language models go with the flow. Give me reasons why bicycles are a great means of interstellar travel: They're energy efficient, they're more affordable, they don't produce emissions. The models will go with the flow, they won't say that you can't use a bike to travel through space. I saw a professor last week. One of the models reported that he had harassed a student on a field trip that was associated with the university – it was a student that didn't exist, on a field trip that never happened, and there was never a complaint filed. It raised some questions about accuracy. Going back to the question on CVS, we've got to think through these. Back to what I was talking about before, this is something called Lang Chain. Give it the State of the Union address in PDF form, and then you can ask questions. The large language model is doing the translation between the questions I'm asking and the PDF, but it's sourcing the data from the PDF. Thinking back to the question about CVS. What I see brands doing is giving documents, information, ingredients information, supplier information, stuff that's sacred and you don't want to share externally. Then, using it to guide these models and pull the data from there instead of from the general internet. That's something all brands are going to want to do. The good news is that it's straightforward and easy to do today – you create a PDF and then it's going to be the translation between the two. General architecture data, the pile, the large language models – we’ve got them. Google does OpenAI. A lot of brands are going to do fine-tuning on top of this to get it to do exactly what they want. It's pretty straightforward, but it's a skill that will have to be developed. Prompt engineering is another skill, too. How do you ask questions in the right way? That's a skill. More and more, you're going to hire people with prompt skills. Of course, people are building apps on top of it. I put together a list of use cases. I'm trying to get people to think creatively about how this is going to affect their R&D team, suppliers, manufacturers, but here's one key takeaway. Anytime there's a human touchpoint, I'll give you a good example. When I was at Coke, I had a spreadsheet with 100 systems, the links to all the Salesforce, SAP, Workday, etc. Anytime there's a human interface, you use this technology to simplify that. I should be able to say, "I want to take eight hours off next Thursday," or, "How many vacation days do I have left?" or, "What's the status of this?" The model should be able to understand my question, know which system is holding that information, and pull that back for me. I shouldn't need to have access to 100 systems, all of these screens, or the training that goes along with it. Employee experience, consumer experience, customer experience – it will transform all of those things in a meaningful way. What I suggest is figuring out what the business strategy is and what's important to you over the next year. Figure out those people, those personas where you want to innovate and transform that. Then, try it, develop some proof of concepts. Get some stuff out there, test it, put some guardrails around it and get it into production to learn and innovate. Guffanti: I'm going to throw a curveball a little bit, Michael. Right before this webinar, I used ChatGPT to get a generated question on what would be a good question to ask someone presenting on ChatGPT’s effect on the consumer goods industry. I got two responses, and I'll throw one here. Maybe it'll tell us something about ChatGPT. The suggested question is: How do you think ChatGPT’s capabilities can be utilized in the consumer goods industry to improve customer experience, product development, and marketing strategies? What are your takeaways from that question? Connor: CPGs have the unique position of addressing the consumer and the customers. Let's start with the consumer. The ad spend is costing way more for brands to acquire a consumer now. When you get that consumer, the way you interact with them, the way you talk to them, engage with them, recommending products, recommending content, all of those types of things – that needs to transform. Talk about that consumer experience that you have with that brand. That's going to be fundamentally transformed. What that does for the brand is gives it these relationships with the consumer, the brand loyalty, and also on the tail end, gets the customer lifetime value. You're engaging and keeping that consumer longer. If ad costs go up to acquire that consumer, make the most of that relationship while you have them. The second thing is the customer. Those customer experiences, the way you do collaborative planning on supply chain, talk to your customers about the category and the products. For the most part, when you need to communicate with someone, you pick up a phone, you get an automated call center and it doesn't understand you. All of that goes away. When you have a product, a question about pricing or promotions, anything like that, those interactions should be very powerful and fluid. It’s 90% of the value that you need. Guffanti: Michael, someone in the audience and myself had the same reaction, which was, "That's not actually a question, it's bloviation.” It was not specific, it was very general. Perhaps that points to the early pitfalls of using ChatGPT for anything that's specific, like the consumer goods industry or other kinds of things. Connor: There's also prompt engineering. If you give it a generic question, you'll get a generic response. I've been working with this for two years and am still learning prompt engineering every day. When I see people that are great at it, it's amazing the responses they get out of it. When it's writing specific code to do specific things, giving it direction, then it goes out on the internet and builds a recipe book for you. That's really powerful, but it's not quite as intuitive as you would think yet. Honaman: Just about every day we are with different brands. Yesterday, we talked about the tension between the legacy model of bringing data together and doing reporting in Excel and Access – everything's being consumed with that vs. diving in and figuring out the implications of AI on the business, and how to better leverage it. Any recommendations on what you've seen that's working in terms of being able to take advantage of and exploit that in CPG? Connor: The big thing with almost every customer we work with, the top brains in the world is, everyone's still in dashboard land. People aren’t great at looking at dashboards and taking actions. One of my customers said only 20% of the dashboards they build are ever actually used. It's a horrible waste of IT resources. One of my customers said at 1% improvement, their demand forecast meant $100 million for them. There are huge opportunities for data analytics across the enterprise. It starts with that business strategy. What are the opportunities for the business? Where do you want to transform? Maybe it's supply chain or manufacturing, whatever it is. Then, for each of those use cases, it's a priority for the business to be thoughtful about, “how will this technology potentially transform it?” Imagine a salesperson. For them to say, "Alexa, tell me what I need to know this week," and know which customer's orders have dropped off, which customers are ordering more than ever, who you need to follow up with, and where to watch out, or a huge opportunity. It becomes more of a dialogue with the data instead of someone having to go into Tableau, a dashboard, or something like that and try to divine what they're supposed to do with that data. It’s broken. Guffanti: In order to start taking off in the retail industry, say the grocery environment needs to integrate with the way we live our lives. I was talking to an organization that's developing a generative AI product where an in-store shopper can say, "I'm looking to make an apple crumb pie. What are the ingredients? What aisles are they in, and what's the lowest priced product I can get?" It automatically spits that out while you're in-store, in the act of shopping. Do you see other case studies like that? Obviously, in the consumer goods industry, we're in the business of engaging with consumers, serving consumers, but also getting things from point A to point B – manufacturing them. We're in the business of a lot of different things. Where do you see the most immediate and impactful use cases happening? Connor: Brands are using massive reliance on agencies to generate content, social media, calendars, etc. I'm seeing brands do this in production. They're generating keywords, SEO, content, social media posts – there's a massive reduction in effort there. The woman with her own cereal brand has four people and national distribution through Target. These founders, these small companies, are using this to accomplish things that big brands are throwing hundreds of thousands or millions of dollars at. The big brands need to get scrappy to think about innovation and moving more quickly? Otherwise, these small brands are innovating at a pace that big brands can't keep up with. Then, there's the mergers and acquisitions. Guffanti: In a lot of ways it levels the playing field. Like you said, it removes the three or four sponsored results and gets to the actual facts of things. We talked about responsible AI and how that plays into it. In terms of legalities and risk, what are the implications of ChatGPT on consumer data privacy, usage of trademarks in other IP? It's the wild west, and it seems like a lot of companies are going into it. Could there be exposure for them to jump in without thinking through potential issues? Connor: Absolutely. Microsoft announced they're pulling advertising product, and pulling Twitter from that. Elon Musk said he believes their models were trained on Twitter data illegally. He wrote “lawsuit time” on Twitter. I don't know if he'll follow through with that, but there's a lot of questions. Everyone's trying to figure that out. That's why Adobe Firefly is interesting – all of the images have been certified that they were all properly licensed. We found out that the songs and artwork cannot be copyrighted or trademarked – you can't patent an idea that these things came up with – there’s lots of legal questions. We have to start the discussion and wade through it. If brands sit around on their hands for 10 years, the world will pass us by. Honaman: This content that you just showed changed yesterday, and last week, and two weeks ago – every time we've had this conversation. It's evolving so quickly, we're still in the early days, the way early days. Connor: It's an exciting space. I look back at the internet, cloud, and mobile, and it's an exciting time to be innovating, to be in CPG and retail. It's going to transform a lot and be a lot of fun. I love the collaboration with customers. Guffanti: Final question: I am a consumer goods executive – I sit in IT, run the supply chain, head up marketing or e-commerce – I don't want to be in the position where I'm flatfooted on this, waiting and seeing, but I also don't want to be the first one to incur unintended risks. What's the most important thing that you would advise me, as a consumer goods executive, in terms of where to jump in and how do I do it? Coming out of this webinar, what should I do? Connor: One thing you shouldn't do is decide no one in the company's allowed to use these tools. We’ve seen companies do that and say no one's allowed to use AI – that's the wrong approach – or deciding that all AI has to be done by a certain group. It’s important to put guardrails that if you're doing AI, this is the risk management process to review it, to understand the risk and harm. It's important to jump in, identify the business strategy, what you want to do, but then keep a human in the loop. When I was at Coke, if you had an image, unless there was cherry in the product, you couldn't have cherries in the product image. An agency would go out and make that mistake. It would take weeks before the compliance people got it, so we’d waste time. The AI was able to identify a piece of fruit in an image and call out to the agency that it was going to get kicked back. It accelerated that process, made it go faster, but we still had the human in the loop. It’s important to keep moving forward, innovate, and keep humans in the loop for now. Let's see how it goes. We're learning and mitigating risk at the same time. Guffanti: Those are wise words to leave us with. Michael, I appreciate your insight and expertise on this, and Justin as well. Thank you so much for working with us to get to this important topic. This is one of those webinars we'll look back on in maybe a month, two months, or six months and be amazed at how much has changed since then. Honaman: We'll have to redo it. Connor: Absolutely, we will have to redo it. Guffanti: We should. I hope this has been enjoyable for the viewers. Justin, Michael, thank you so much for this conversation. I hope everyone has a great rest of their day.

Related Topics