[Music] [Akin Ajayi] Good morning, everyone, and welcome to S522, Supercharge Marketing Productivity and Creativity with GenAI. My name is Akin Ajayi. I'm a Senior Product Marketing Manager for Adobe Experience Platform. And this morning, I have the privilege to be joined by my esteemed colleague. Do you want to introduce yourself? [Rachel Hanessian] Yeah. My name is Rachel Hanessian. I'm a Product Manager working on the AI Assistant for AEP. Akin and I have been working closely together on AI Assistant for about a year now, and we're super excited to share with you more about it.

Awesome. Thanks, Rachel. If you've been to the pavilion, you might have seen Rachel and I at the AI Assistant booth. If you haven't been to the booth, please stop by on your way out. We've got-- Let me just say we've got the best swag at Summit. It's the shirt that I'm wearing, and I think it's pretty cool.

We designed it, so please pick it up on your way up. I hope you've had a great summit experience. I'm sure over the last three days, we've pumped and inundated you with a lot of information. But I hope you're leaving inspired by all the announcements, all the innovation that we're bringing into the products, the Adobe products that you've already bought or are about to buy. I, for one, am particularly very excited about what we're going to talk about this morning. And to get us through it, we've broken it down into three themes. The first is what I call "In the Era of GenAI." And since we're in Vegas, if I was a betting man, I would bet that when you jumped on the plane to get to summit, the one topic that it was sure you were going to hear over and over and over again would have been GenAI. And I hope we didn't disappoint you.

And so this morning, we want to dive a little bit into what we mean by In the Era of GenAI, especially when you think about creating digital experiences and personalizing at scale to your end customers. And we're going to do this within the context of one of the big announcements you heard which is the AI Assistant. So we're going to talk a little bit more about, what it is within the context of where we are in this new era, in this new journey.

The second theme is around the application of GenAI. And I think this is where you start to see the rubber really meet the road. And what we want to show you is want to peel the layers a little bit and show you what the building blocks of AI Assistant in Adobe Experience Platform is. And we're going to show you that within the context of very realistic and tangible use cases. And you can start to see how that's going to basically supercharge productivity for your end users as they use all the innovative Adobe Experience Platform based applications. And then finally, I know I mentioned that we already pumped you with a lot of innovation and announcements. We want to give you a sneak peek of what's coming next, what you hopefully can expect when you come next year that we will be either announcing or you already have been using those products. So I want to give you a sneak peek into the future of what's coming next. Now before we dive into the first thing around in the era of GenAI, I want to take your mind back to Tuesday, main stage, where our SVP of Engineering, Anjul Bhambhri, first made the announcement around AI assistance in Adobe experience platform. And she talks about some of the key value points such as, it being able to answer questions, etcetera. That was at a very high altitude, right? And I hope you got excited when you heard all that great stuff. What we want to do in this session is we want to take you deeper. We want to show you what's under the hood essentially of the AI Assistant and Adobe Experience Platform. Where we currently are, like I said, and where we're going to.

It's no longer a science project. You know how these things start? It starts with universities trying to come up with the next advancement. And then some of those never make it out. This has made it out. We are, it's no longer a toy. These are life changing tools that, like I said, are going to just transform business, transform how we live our lives. And it's made possible by three things actually converging together around this time. The first is the explosion of data. Never before have we had this amount of data available at our fingertips to be able to process. The second thing is the technology, the technology to actually have the server infrastructure to be able to sieve and provide the processing power for computers to be able to sieve through vast amounts of data pretty quickly. And the third, which I think is the game changing piece of GenAI is the models. We've seen sophisticated models that have now been developed like large language models, large multi model models. These models are helping to allow you to have a conversation in natural language with the customer and with the computer, and the computer is able to interpret your intent and it's able to give you an answer that's contextual, relevant and fact based. And that's the beauty of GenAI.

So I'm really curious to see how many of you are already embracing the new. So I have these three questions and I would just like you to indicate with a show of hands, if you are already using ChatGPT or GenAI or any chatbots to answer questions today. Almost everyone in the room. Wow, that's amazing.

How many of you think GenAI could be more productive, could help you be more productive and help you get things done faster and at a higher quality? Wow. We've got believers in this room. That's awesome. And now this is more of a cheeky one. Who of you think that by 2030, a GenAI copilot will be running the universe? This is some Star Trek stuff right there.

Love it. Love it. I'm sure you're like a big sci-fi.

There we go. Awesome. It's great to see that but I want to ground this in some statistics, especially when it comes to how this is likely to affect business, organization, and the workplace. We're already seeing a massive expectation of rapid growth. To put it in context, I'm going to call out some stats that are not even on the screen. You wouldn't believe that it took ChatGPT just five days to get to one million users.

Isn't that amazing? Like, even Instagram, Facebook, all the innovation that came before that, it took them way longer to get there. And so here we're seeing about 40% growth rate as the expected growth rate for the adoption of GenAI. And one of the most profound things is a lot of businesses have identified that the key value prop for GenAI would be employee productivity, helping with knowledge management, so democratizing access to knowledge, and self-service data and analytics. So getting you faster access to insights, without necessarily being an engineer or a techy user.

And when you think about it from the context of the marketing tech, there's a lot of expectation here. In fact, a lot of organizations, we conducted a survey that shows that 89% of the folks we surveyed, customer experience leaders, believe that this technology would help them better personalized customer experiences. And that's just beautiful because now we can accelerate what we've all been trying to do. One to one hyper personalization. And I'm sure coming into Summit, you have very high expectations of Adobe in terms of what you expect to see in terms of GenAI being embedded in our products. And so we're already seeing that in all the surveys, from all the industry analysts.

But I want to stop before I go to the side. I actually want to go back, and I want to address the elephant in the room.

While there's all this excitement, there's also anxiety. Whether we want to admit it or not, there is anxiety. Is this going to take our jobs? Is this going to replace humans? Are we going to have AI making all the decisions in the world for us? Those are valid concerns but here's how I like to think about it.

We've gone through several eras in the past where we've had to shift how we do things or shift the business model. And I don't think this is going to be any different. There were folks who were worried about what would happen when computers were created. Would typists go away? What happens? But what happens is we continue to shift the levers to more strategic stuff. And so now we don't have to worry about some of the routine things. We can give that to our computers to do, right, and we can focus on the more strategic things. And with that comes a shift also in productivity, in the boundaries of innovation that we're able to explore and that's exactly what this technology is going to do. So I do understand the concerns and they're totally valid. But if we can continue to do this in a very brand safe manner and where it's governed and regulated, I believe that it's the good definitely outweighs the fear.

And that's exactly what we've always done at Adobe. We're all about reinventing ourselves. Over six-years-ago, we made the decision to start to invest in what we believe is the technology of the next generation, of the next decade. The technology that's going to power personalization for our customers like never before. And what did we have to do? We really had to re-architect our technology to be fundamentally based on what's called Adobe Experience Platform, which is basically the connected foundation that helps you to bring data together very quickly, both batch and real-time data. And you're able to basically create a very comprehensive, unified, actionable view of the customer, which is called the real-time customer profile. And you're able to use that profile to basically power all the insights, measurements, activation and engagement, real-time one-to-one engagement to your customers. The use cases that you're powering to be able to ensure that you're showing up where your customers are with the right context, the real-time context around what they're doing. And that's exactly what we did six years ago, which has birthed the applications that are built on top of platform that you see on the screen. These applications are best in class and they're really changing the narrative and the landscape of how we do personalization and customer experience management and how we show up to our customers. So we've always been at the forefront of innovation, and you see that through all the eras. If you've always been-- If you're very familiar with all our products, you would be able to see how we've always come up with innovation at every single step of the journey. And it's no different with AI. In the AI era, we're doing a lot. You saw a lot of announcements. You might have seen announcements around GenStudio. You might have seen announcement around how we're bringing Firefly into the workflows, into the marketer's workflows, so that you can quickly iterate on content that you're delivering to customers and you're able to collaborate in the case of GenStudio, have better collaboration between your marketers and your content creators. And AI Assistant is the focus of this morning. I think it's the one that I'm most excited about because I can see the value it's going to have to every practitioner, every user within your business. And once again, I'm not just going to tell you, we're going to show you, but before that, I'm going to pass it on to Rachel. So Rachel can talk us a little bit more about what exactly AI Assistant is based on the current state of the product. Over to you, Rachel. Thanks so much, Akin. Thanks for that great intro on Generative AI. So as you may have heard earlier in Summit, we've been building an AI Assistant, or some of you I know have visited our booth and the booth will actually be open until 3 PM today. So if you want to actually get your hands on the product and try it for yourself after this session, please come on down.

So what the AI Assistant is, is a conversational interface that is natural and easy to use. You can ask questions of the assistant, and it will get you back answers in natural language, with full explanations of how the answers were generated. We believe, as Akin mentioned, that this is really going to reimagine software by increasing your productivity, just making it faster for you to accomplish tasks that you already do, also democratize access to the tools. So what we mean by that is tasks that, before we're held up just for very technical users. Now less technical users are able to accomplish those as well. So kind of breaking down the silos between roles. And the last thing that we believe this will do is unleash new ideas and more of them. So later, we'll talk a bit about some of the improvements we've seen in idea generation and how we've seen it go from, it takes an hour to generate one idea to it takes minutes to generate 10 ideas.

So the AI Assistant makes it easy for you to find your data. You don't need to know how to write a sequel query to get really deep insights. You just ask AI Assistant, and it will explain back to you, and we'll show you that in a few minutes. It also makes it much easier to learn new concepts and to troubleshoot. So giving you information right at your fingertips, instead of reading through long documentation manuals, you can just ask AI Assistant a question, get an answer back, and it's grounded in your specific organization's data. The way I like to think of it as, is in the past, we used to need to kind of memorize facts, right? And then we got Googled. Now we just Google it. In the past, you had to really memorize and learn how to use these UIs. You needed to learn how to write SQL queries. Now you just use AI Assistant for that.

So, also, you may have seen this slide in some of the keynote talks, but what is powering AI Assistant? It's a collection of generative experience models and these are broken down into three parts. There's decision services that sits on top, and then we have a set of base models. These include public LLMs, linguistic models, and these are fine-tuned on Adobe data and our Adobe documentation. There's also a set of custom models, like propensity models, recommendation models, and these are built on each specific customer's data. So you can kind of think of this right-hand column as it's individualized per customer. This enables there to be no data leakage between customers, and also, as you can see by this model, no GenAI is trained on customer data.

So why did we build it in this way? Why is there this kind of decomposition? And also, why am I taking the time to show this to you? It allows us levers to control for data security, for us to be able to be transparent about how the answers are generated, and it allows us to ensure a level of correctness. So the elephant in the room for me is kind of hallucination, right? We know in LLMs, we sometimes get answers back, we don't know if they're true or not. This decomposition allows us to understand exactly where each answer is coming from and correct it if there's any issue. It also allows us to surface back exactly how we got the answer, so that you yourself have the ability to verify the correctness of the answer.

So AI Assistant. So you may have seen some of these screens in the demos. We've been in an alpha for this feature since September of last year, and we're moving into the beta phase in just two weeks. During alpha, we analyzed all about 5,000 interactions from our customers to try to understand what were they using it for, and what were they finding it most useful for. We have an in product, thumbs up, thumbs down, and comments section, so we've carefully looked through all of those and analyzed what are the major use cases. So Akin and I went through those, and we picked out a few examples to show you live what people have been finding the most value in. So I'll go through a series of demos, which are product recordings of what the product looks like today. So if any of you are current customers and are interested in joining the beta that's starting in just a few weeks, we'll actually pop a QR code up on the screen and love you to scan that and sign up.

And Akin and I want to make this a bit of fun to go through these examples, so bear with us as we introduce you to BugWarts.

Well, Rachel. So as Rachel mentioned, we're going to try to make this as fun as possible and, we'll take you through a day in the life of the marketing department at BugWarts Enterprises. You're going to see first-hand how a data engineer, a marketing ops analyst, and a marketing manager are able to go through their day and perform tasks and increase their productivity using AI Assistant.

So in the moment, we'll start by shadowing the data engineer. It's 8 AM in the morning and they're about to start their day.

But before we go into this demo, you might be curious about what we've got on stage and you're like, what exactly is happening? I'll start with the Jenga Power I've got over here. I'm a big Jenga player. I think I'm one of the best.

And so I just had to show you my skills. I'm going to remove, try to remove one of these pieces, and let's see what happens.

No.

Gosh. I'm so sorry I just embarrassed myself.

I'll admit it. The pile fell because I took out one piece. But what piece was it? And why did it bring out the whole structure? Like what exactly happened? Rachel, I know that you are a data engineer and you're all about data structure. So I'm wondering if your skills are applicable with Jenga here. Can you help me figure this out? What piece did I take out that's caused this? Honestly, I can't. My job as a data engineer does feel like I'm playing a constant game of Jenga, balancing updates to the data setup. It's really important for me to keep tabs on the usage of upstream and downstream usage of the data in my job. I do organizes structures, kind of similar to how a librarian organizes books, categorizing everything and labeling, ensuring that everything's where it should be, and is also in the place where it's safe for everyone to use. My challenge is really remembering, where each piece of data is and who used it. What I usually need to do today is ask team members where things are being used, which sometimes they don't appreciate and it's a very manual process that can sometimes feel like a dead end. Because of this complexity, I sometimes just shy away from making any changes and then things get a little messy. I don't want to send the whole structure toppling down, like Akin just did. It can cause quite a headache. So to make it more concrete, I'll show you how in AEP, I sometimes need to find where things are downstream.

So here I'm logging into Adobe Experience Platform.

And in this example, I'm trying to figure out which segments contain a specific attribute. So I'll pop out AI Assistant, and I'll ask, which segments contain the person.birth year attribute? So I have created this attribute in one of my schemas and I want to see where it's being leveraged. Now how I would have to do this today? So first we'll wait for a response here. So what's happening behind the scenes is, it's creating a SQL query, sending that back, and giving me an answer. What I would have to do today to solve for this is potentially go to the segment screen as you're seeing now.

And in this org, you see there's 21 segments, but some of the organizations, I'm sure you have up to thousands of segments, go through and manually check where each attribute is used. But with AI Assistant, I'm just able to ask a question, get the nice printout of the audiences and I get verification of how I got the results. It explains how I joined a table, a segment table, an attribute table, and figured out exactly where this attribute was used downstream. Since I'm a more technical user, I can also look at this source query. It opens up our query service tool, which shows me how the query was created on my behalf. This is super cool because I can edit the query if I want, change the attribute, and I didn't have to do all the hard work of creating a SQL query. It was done for me by the AI Assistant.

The next example that I'll show is, if I want to understand more about one of these specific audiences, instead of navigating to the right screen where that audience exists or bothering my marketer, I can just ask AI Assistant, show me all of the attributes in using auto-complete, so I don't need to worry about typing something wrong or if I don't remember how something is spelled. Select that audience and I can just send this question to AI Assistant. Similarly, what's going on in the backend is it's taking this natural language question, converting it to SQL, pulling it from the database that we've constructed and giving me the answer in seconds. So I can see right here the city attribute, the birth year attribute, and again it's giving me a very detailed explanation of how the answer came to be. So this way, if this doesn't match with what I was looking for. I know right away. It's very transparent. I can see exactly the steps that were taken to get me to this answer. As I mentioned earlier, we have this feedback mechanism that we've been using in our alpha program, so that our customers can let us know, is this something that's useful? Is this something that's not hitting the mark? We're taking every single interaction. We're basically looking under a microscope and seeing how can we make this better, how can we increase the number of thumbs-ups and decrease the number of thumbs-downs.

So the next example I'll show as a data engineer trying to figure out where all of my fields are being used is trying to understand, has this audience actually been used? Has it been activated to any destinations? So, again, this would be a very big game of kind of cat and mouse trying to find and track down this audience. Instead, I can just ask the AI Assistant. It gives me the activation count as one, explains exactly how it got to that answer.

Maybe I want to know which destination, so I want to call out here. All that the query says is which destination. I'm not giving any context of which destination is that specific audience in. I'm able to do something we call this multi-turn, where I just ask a quick follow-up phrase and it understands all the context of my previous questions. It also gives me the link to the destination. So again I don't need to remember how the UI works. I can just click the link, brings me right to that destination, loads it behind the scenes and I can dig into it more. Again, just, would like to highlight the explainability here, explaining exactly how it got the results, so that I know if it is actually answering my question or not. All right. Let's get a little bit more complex. Now I'd like to see a table of attributes based on the highest activated segment count and include the total segment count. So I want to see which attributes have been used most in audiences. This will help me understand if there's a policy attached to one of these attributes, which ones are having the highest impact here.

So same thing. It's taking this natural language prompt, converting it to a SQL query, so I don't need to spend the time doing that and I get a beautiful table that lists all of the attributes, which segments they are or how many segments have they've been activated in and the total segment count that they participate in.

So you're seeing how these questions, each one is probably taking just a few seconds, right? And if I weren't to have AI Assistant, I would have to reach out to another human or spend time writing the SQL query, iterate on that. I'll now ask a tricky one. So which fields occur in segments tend to occur together? So what are the most popular field combinations? I'm not even sure how I would do this without AI Assistant. This is kind of unlocking a completely new possibility here. So we'll see how AI Assistant deals with this one.

And how it responds here. It's providing a visualization for me. So it's giving some attributes on the y-axis, some attributes on the x-axis, and giving me a sort of heat map. So as I highlight over each particular intersection between the attributes, it's telling me how many times each attribute has been used together in an audience and the number of segments that it's been used in as well. So this is something that if I'm trying to create a presentation on this, instead of manually creating this chart, I can have it be created by AI Assistant, export it out, and use it.

Again, we have a very detailed explanation there. Now I'd like to show bookmarking. So we have a bookmarking feature where we can save questions for future use. So this is one that I as a data engineer, like to kind of rerun. Show me all fields containing the string name, what schema they're in, and when that schema was created. So this is something that I can rerun over and over, just go back to my bookmarks, and run it. I'll pop out the table a bit, so we can see it a bit better there.

And so here I can see the field name. You can see that all those field names have name in them, so able to do, like, a very powerful search here. We have the schema name, the created time. I'd like to point out that there's basically an infinite number of questions that I could ask the AI Assistant. These are just some examples of them, right? But the beauty in this is just how flexible it is. We can't think ahead of time all the questions that our customers would like to ask, but that's why this open-ended inter phase.

All right. Well, that's all for the data engineers morning at BugWarts. I'll hand it back to Akin.

Wow, Rachel, that's awesome to see the Data Engineers. No longer caught up in a game of-- I'm able to get things a little more organized instead of having to do it manually, and go figure out things in a lot of places and what's-- In particular, how often downstream AI system has been all around that platform. I bet it was in collaboration between Data Engineer-- I got. All right. I'm trying to make collaboration easier between data engineer and marketing personas, isn't it? Absolutely. Yeah. We have time to focus on more strategic tasks and less of the kind of mundane. They're like, she's back with another one of these questions. That's awesome. And I think what I love most about that demo was how natural it felt. Like the conversation was so natural. You could type the questions in, and it was just beautiful to see that. The other thing was how there was one place where you asked a follow-up question, and the AI system just picked it right back up and was able to give you an answer based on the previous context of the previous question. - Yeah. - That's amazing. A multi-turn. We love it. Multi-turn, that's what it was. Awesome. All right. Since you're not able to help me out with my Jenga, Rachel, let's crossover to the marketing department. It's about to be launch time just like it is about to be launch time over here. And I know this session is standing your way, but we're going to get to it very shortly. Let's just quickly go see what's happening with the marketing ops analyst and how the AI Assistant is able to help support some of her tasks. But before that, I've been doing some house cleaning recently and I came upon this pile of socks. Now the challenge I'm having, Rachel, is I have different pairs with different colors. Some are old, some are not even clean. And I'm thinking, what do I do with this pile? There's something I need to throw away. Rachel, I need some help to sorting through these socks. Can you help me? Can you take some time to help me, please? Akin, it's really funny you asked because as a marketing ops analyst, sometimes as I look at my inventory of data and objects, I feel like I'm staring down a messy heap of socks. I've always trying to figure out what data needs to get cleaned, what data needs to stay. And as much as my colleagues and I try to clean things, folder things, avoid duplicates, organize, messes happen. And it's a true pain towards all, especially when I'm the one that has to resolve it. But no, to answer your question, even though this is becoming quite a mess up here, I don't have time. I just have a few minutes before my lunch break to show all these lovely people how AI Assistant makes my time as a marketing ops analyst, easier to organize things in AEP.

So as a marketing ops analyst in AEP-- All right. So I'll show how, one of the tasks that I need to do is, I need to figure out duplicate segments, right? I'd want to make sure that the marketer upstream is using the right things and doesn't get confused about which audiences to use. So, normally, what I would have to do is look through this whole list of 21. We even heard from our early adopter customers that they export Excel sheets of these, look manually, sort through the definitions, try to see what are the duplicates. With AI Assistant, you'll see it's much simpler. I can simply just ask AI Assistant, show me audiences with duplicate definitions, takes this natural language query, flips it into a SQL query, especially for a user such as myself who is a little bit less technical. Normally, I would have to ask my data engineer, but that woman is always busy playing Jenga or something. So this prints out just the audiences that I need, tells me exactly which ones have duplicates.

It tells me how it got the results very clearly. So even for a person less technical such as myself, I'm able to see how this works. It's even helping me become a little bit more technical because I can see how the SQL query was generated based on my natural language, and I am getting a little bit more comfortable now tailoring the SQL query based on some really simple things, like you can see limit 25. I can understand that. I can change the limit. So I'm even almost learning some SQL from AI Assistant.

I'll go back to show another example. So that's a really good one. I'll bookmark that for later.

I want to figure out, of these duplicates, which ones were created in the last six months? So it looks like there's some duplicates. I want to see, also who are the culprits of creating these duplicates but when were he the recent ones created? So I'll ask this question.

An AI Assistant prints out a table for me, so it gives me the segment ID and the name, and all of these were created in the last six months. So I can dig into these a bit more to see who created them, why are we getting these duplicates, what are they being used for. I might want to figure out the specific fields in one of the audiences, so I'll use auto-complete. Again, this is much simpler than typing out the entire audience name. I can just get a friendly list and select there. Ask AI Assistant. What are the fields that are used in this? So this helps me understand the definition instead of, in AEP we have this profile query language. Again, a little bit too technical for someone such as myself, so I like to just see it playing with AI Assistant. Okay. Let's get a little bit more complex. Show me audiences that were created in the last 30 days. I want to make sure that they're just batch segments, so we have batch, streaming, edge segments, I'm limiting just to batch, their associated schema, and the profile count in each. So pretty complex question. Again, we have to write up a SQL query to get this kind of information. I don't even know how I would do this one manually. Let's see how AI Assistant handles it. Perfect. So it prints out a nice table for me. I'm able to blow it up a little bit bigger here.

And so you can see the different columns that we show our segment ID, name. You can notice total profiles, it's not ordered. A follow-up question I could ask is, hey, rank this by profiles or order it by profiles, so that type of iteration is possible as well. If I'm using this for, maybe, a presentation to try to show folks to keep things a little cleaner, I could use that and then eventually we'll have functionality to export it. All right, let's see the top segments by profile count and then also the created date. So digging a little bit deeper here, trying to figure out just in general what are my most populous segments.

AI Assistant provides a beautiful chart here. So, again, this is gold if I want to create a presentation on something like this, I'd have to manually create it. There's a lot of detail here, and instead, AI Assistant, in seconds, was able to just generate this for me. As I hover over, I'm able to see the created month, the name, and the total profiles. So this is super helpful and just is really increasing my productivity a ton.

I think I showed about, right, like, eight questions in four minutes, so each one, otherwise, probably would have taken me a week. Maybe I would have had to rely on another person to help me get it done, so this is really just supercharging my workflow. I'll hand it back to Akin.

Thanks a lot, Rachel. I'm sad you weren't able to help me sort out my socks, but that looked pretty neat. It looks like you're going to be relying less and less on your data engineer for those things that you just want to get very quick at your fingertips. That's amazing. I think it's also amazing to see how you're basically getting more independence, but not just only that. I know that from my work as a product marketing manager, I think AI gets a very bad rap as being a black box of magic, right? But I notice how there's a lot of explainability in the results that AI Assistant is displaying to you, so you can understand how it got there and the granular steps that we're taking to get to the answer. I think that's super-duper cool. But now that we're done with you marketing ops analyst, it's time to go to your boss, the marketing manager. And just like we're about to wrap up the session pretty soon, it's also getting to the end of day at BugWarts Enterprises. And so now we want to see what's in store for the marketer at BugWarts. But before then, I'm trying to complete the perfect toy.

So I've got this toolbox that I want to use to build a toy house.

But once again, how do I figure this? How does this work? There are so many tools that I need to figure out. Let me see if I can check my manual and get some instructions.

My God. Who's going to-- Who reads instructions anyways? Rachel, I'm struggling. How do I use these tools? Where do I even start? And this is not just about the fact that this is new. Even for some of the toys that I have built in the past, when I get back to trying to reassemble them, I have exactly the same problem. I'm trying to figure out what exactly, what tool is meant for. How can you help me, Rachel? Yeah, so we can work through all of this after the presentation, okay? It's getting to be the end of the day at BugWarts and I'm sure folks want to have lunch as well. But seriously, all of that, I feel your pain completely. As a marketing manager, right, it's very hard to keep up with all of the new things and best practices, and it can be very overwhelming. It can almost feel like a full-time job on top of my full-time job. But it's critical to stay up-to-date with all these things to be successful. In the past, how I would get around this is reading instruction manuals, honestly not that dissimilar to your scroll, ask consultants for help, ask my colleagues for help, but even as hard as I tried to do that, new things come up, there's new best practices, there's new tools to stay on top of, and I constantly find myself kind of caught up in this hamster wheel of trying to stay on top of the newest and best things to do. That's why I love how AI Assistant acts as basically an in-pocket consultant to keep me up-to-date with all the new stuff that's going on in Adobe Experience Platform. I'll show you. So in this example, as a marketing manager, I just found out that there's a new feature called Experimentation in Adobe Journey Optimizer, and I want to create an experiment in Adobe Journey Optimizer, but I want to learn a bit about the functionality first, right? This is brand new. I'll start by asking AI Assistant, why should I build an experiment in Adobe Journey Optimizer? It lets me know that it helps with data decision marketing, optimization of customer experiences, improved campaign performances. This all sounds great. I'm pretty bought in. I see that, also, this information is coming from some specific sources, so that I know that I can trust it.

Now that I'm bought in on doing this, how do I do it? How do I create this experiment in Adobe Journey Optimizer? So I'll ask AI Assistant, prints out some perfect steps, gives me just, like, the one, two, three, four, five, six. Instead, I would have had to go through a long manual. Now I'm just able to easily see the steps of how to build an experiment.

I also notice here that there are some sources that I can look at. So when I click sources, you can see that there's a long list of different articles. So this didn't just pull one article and plop the exact text. It actually took information from across many different articles, combined them all into this answer, and you'll see these little numbers. Those are inline citations. So each piece is cited, and I can click into each of those to learn what document it came from and to learn more if I want to dig deeper. So I know that this answer is really grounded. I'm now interested to find out what types of experiments can I run in Adobe Journey Optimizer? Gives me some examples, so I can do content experiments, I can do channel experiments, personalization experiments. I'm really interested in content experiments, so I'll dig a bit deeper there. And I'll specifically ask, what types of content experiments can I run-in AJO? Tells me subject line experiments, image experiments, copy experiments. So it's giving me all sorts of ideas of what I can do, grounded in the product truth and what I'm actually able to accomplish in Adobe Journey Optimizer.

I see that there are some suggested prompts here. So if I don't know what to ask next, I can just click on one of these suggested prompts and ask what metrics can be used to measure the success of a content experiment. I see email opens, click through rates, delivery rates.

And these suggested prompts are really helpful here. I'm actually going to show clicking on a source and how it shows up. So it highlights the exact portion of the document where the answer came from. So this helps me know exactly what piece of the document was referenced in the answer.

The suggested prompts are super helpful because if I'm just learning something, I might not exactly know what question to ask next, and AI Assistant makes it really easy for me to just continue on a learning journey.

So scrolling down and looking at the sources here, I want to learn a little bit more about the statistical calculations that are used. I know that I took a couple of statistics classes. I learned about this stuff and reading through, even reading all of this, it looks a little bit complex to me. I see that Adobe's statistical method is any time valid confidence sequences, and I'd like to learn more about that but not from reading through all this dense information. I'll just ask AI Assistant, what are these anytime valid confidence sequences in Adobe Journey optimizer? Maybe they can give me some key points here. Perfect. It lists out the definition and the key points. Now this is kind of jogging my memory that these anytime valid confidence sequences allow me to peek at the results early and it won't even mess up the experiment. This is a pretty complex topic, and I'd like to bookmark this definition just so, if I need to ask it again or if I forget again, I can always come back to it.

And speaking of, maybe not remembering everything that I've done in the past, let's say one week later I come back. I can go back and see my chat history. So I can go back seven days and see what I asked and kind of review my history. I can go back 30 days, review my history. This is something that our alpha or early adopter customers really asked for. All right. One last thing to show on these more informational use cases. If I ask, let's say, what is the difference between a batch and a streaming segment? I get a definition. It also really helps me navigate, so you can see that segments is hyperlinked.

And I'd also like to call out that it grounds it in your specific sandbox. So I can see that in this sandbox that I'm in, there's 17 batch segments and 24 streaming segments. So it doesn't just give me the information from the documentation resources, it also grounds it in my specific data, so it makes it a little bit more contextual for me. It also helps with navigation, so I'm able to navigate to the specific examples that are called out. I can navigate to the segments inventory page. I think I mentioned this at the beginning. We really, imagine that this AI Assistant will revolutionize the way that you interact with software. You don't need to learn the UI as much anymore. This can kind of be your navigation plane. Datasets go to datasets. So it brings in all the relevant pages and you can link to them right from there.

All right. Akin, I'll hand it back to you.

Wow, thanks, Rachel, for sharing how marketer can-- My goodness.

You can see how it helps the marketer increase their productivity, even unleash new ideas, find out about new features that they didn't know about, and they can go on this deep learning journey of how to use the technology.

I use a lot of ChatGPT and one of the features I like most is how I can go back to previous questions because I bookmarked them, and I can just get to use that same prompt over and over again. And in this case, I like the fact that, you can rerun a prompt that you run, I think it was a week before and then you can see how the answer changes over time based on new information. I think that's really powerful. All right. I think we've gone through an entire day at BugWarts Enterprises, has no relationship with BugWarts.

You've seen AI Assistant working. Looks like this is not working, actually.

Yeah, you've seen AI Assistant working through the lens of three practitioners at BugWarts Enterprises. And you can see how it's offering up tremendous value to these personas. But, Rachel, I'm curious to see what else you heard from our customers during the alpha program. What other features, what other use cases were surfaced during that process? So, yeah. So we just showed three kind of main sets of use cases, but wanted to call out the main themes that we saw during our alpha program. So the first one was around object counts and lookups. So other questions included here are things like rank my audience sizes, or show datasets with their corresponding IDs, list inactive campaigns, show trend of audiences over time. The second piece was around learning new concepts, so that was that last example with the marketing manager. Very easy to learn things, get just little snippets of information. Lineage and impact was another big one. So which fields are used most? Has a certain field ever been used? These are kind of unlocking questions that we were previously unable to answer, took a lot of manual work to do. Hygiene. So we saw the marketing ops analyst finding duplicate segments, find, are there any, in the booth actually this has been coming up a lot. Folks have come up and said, is this, can this help me find, for my 6,000 segments, something that has test in the name? We showed it, and yes, it can do that. Graph of overlapping profiles as well. And the last piece is troubleshooting. So everyone has probably experienced one time or another, you get hit with an error and you don't know how to get past it. So with AI Assistant, you can say, hey, I'm getting this error, what's going on? And it'll give you some steps to move on. We've seen really-- If this click-- Perfect. That, as I mentioned earlier, right, it's taking down the time to do these tasks from hours to minutes. And even more so than hours to minutes, the hours is the amount of time that it takes to do one of these little light bulbs. Get one idea, past idea, into action. And now you're seeing, as you saw in those three demos that I displayed, there were at least eight or ten questions that I asked, and it was in fewer than ten minutes each time.

Some of the features that we want to highlight as well. So during our early adopter program, we worked really closely with the customers to figure out what were the key features that would make this AI Assistant valuable to them. And these were some of the things that came up. So we had multi-turn, which was right where we just asked this quick follow-up, and it retains all the context. Who wants to retype everything that they already asked, right? We wanted to make it very natural, as if you're talking to someone just with voice standing next to you, who's able to remember what you said five seconds ago. Verifiability is huge for us. We want to make sure that every single answer is easily verifiable. You can see exactly where the answer is coming from. You can see that the assistant understood your question, right? Because if the assistant misunderstood your question and answers something different than what you asked, that's not going to do it. So having that breakdown of exactly how it, understood your question is important. Bookmarking and chat history, I'll talk about those two together. So being able to go back and see what you've asked, and then also being able to rerun a question that you've asked in the past that you found a lot of value from. Giving that to you at your fingertips is something really valuable that we've seen. Auto-complete, we showed that a few times. Who wants to type in very long list of a segment name? I'm sure some of you have segment names that are, like, 100 characters long, right? They get pretty lengthy there. So having auto-complete is a big win for our customers, so that they can just hit that plus sign, choose from a list and then be done. And then suggested prompts, especially for newer users or people who might not know what to ask, these suggested prompts really help move the learning journey along.

So where are we today and where are we going? We promised we'd give you a little bit of a preview into the future. So today, we enabled, really asking questions, and receiving answers. So the scope that you saw today was around data objects and product concepts. The value that we've seen this bring is around productivity and democratizing access. So those are some of the examples that you saw in the demos today. Phase 2 is all around automation and proactive tips. So automation, we don't mean just everything kind of going without your hand on the computer, but really keeping the human in the loop. And again, working very closely with our alpha customers and then in the coming weeks, our beta customers, to identify what are those key tasks that you would want automated. And what we've heard so far and we'd love to hear from you if you have time to stop by the booth later today to let us know what you would like automated. But what we've heard so far is, it's really those mundane, repetitive tasks, where you just feel kind of like a cog in the wheel and you don't want to do those over and over again, kind of toil reduction. So those are the types of things that we would look to automate for you. And also audience insights. So as you're building something, getting feedback back from the system. So that's Phase 2. Phase 3 is taking it even further, where you can start with a goal. So something such as, I want to sell all my inventory by Christmas. What would you recommend that I do? Help me create a journey. What journey would be successful based on my past journeys and my past customer data? So that one we're really excited to look into as well.

Awesome. Thanks so much, Rachel.

We've gone through a lot this morning from playing Jenga with me, to watching videos. I'm sure you must have been entertained. But I don't want you to forget these three key takeaways. The first is Christmas is coming early to your Adobe Experience Platform apps where we're going to be supercharging them with GenAI in the form of AI Assistant that will be naturally embedded in these products. And as you saw, when Rachel talked about it, it's powered by generative experience models, which is a combination of models, LLMs as well as custom models. All geared towards helping you to be able to get the right answers and to be able to take care of the sophisticated use cases we even have on the road map. And we're doing this in a responsible manner. We're all about responsible AI. We think it's a very important ingredient of scaling this innovation and it's always going to be based on honoring your customer trust. That's always going to be what comes in front of this innovation. The second key takeaway is this is not for a select few. It's for all practitioners. Everyone within your business that is a customer experience maker can leverage this tool to supercharge their productivity, increase the adoption of the tools that are available to them, and it's also about democratizing access. So no longer do you have to wait on a colleague to be able to get a quick insight. You can just do it yourself without even having any technical chops. And it's all about unleashing your creativity with new ideas. Because now as you get more information, you get more knowledge, it starts to help you ideate even faster and you're able to get work done and deliver the right experiences to your customers. And finally, we're just scratching the surface. There's so much more that this innovation is going to be able to bring. Like I said and I declared at the beginning, we are firmly in the era of AI, of GenAI, but we're only just getting started. And you've seen some of the innovation that we plan to deliver. As I said, hopefully, some of these will start to come on board. By the time you join us next time on Summit, at Summit, hopefully, we would have some of those already live in your products. So thank you so much once again. I want to leave you with one final message. If there's a tagline you can take from this, the idea is, I'm going to take off my jacket for this.

Shaq would love that, right? AI Assistant plus you equals marketing magic. So go on out there and be wizards. Please come ahead and if you'd like to pick up some wizards, we've got some upfront. And like Rachel mentioned at the beginning, we want you to be part of this co-creation, right? Come join us. Scan this QR code, indicate your interest and we'd like you to be part of the process of bringing this innovation to life.

We got maybe 45 seconds. Maybe we can take one question. We're just curious to get some feedback, so maybe if we go to the next slide.

Yeah. How could you see yourself-- If anyone can maybe just tell us how they will see themselves using the tool.

And you don't have to answer the questions but, just curious if you have any thoughts to share with us.

What's that? - Campaign development. - Campaign development. Okay. That's a good one.

Product discovery? Product discovery. That's a really, really good one. I'm very excited about that use case as well. Because we want you to adopt fully, right? We want you to get all the juices running.

Please stop by. We've got wizards stress relievers on your way out if you don't mind. But thanks once again for being part of the session. Safe travels back and I will see you next year back on this floor in 2025. Thank you very much. Thank you.

[Music]