ÃÛ¶¹ÊÓƵ

Transcript

Welcome to the AEM Edge Delivery interview series. I’m here today with Karthik Murli-Duran, and he’s a Principal Product Marketing Manager at ÃÛ¶¹ÊÓƵ, and he’s going to talk to us about some of the exciting new AI capabilities of Edge Delivery Services. Karthik, do you mind introducing yourself? Yeah, thanks, James. Happy to be here. So my name is Karthik. I’m on the AEM Product Marketing and Strategy team. Been on the team for a little over six years now, and focusing on all aspects of our product strategy and roadmap, including our Gen AI initiatives. Fantastic. Well, let’s get right into it. Can you talk to us a little bit about the different AI variations that are available and why this is useful to marketers? Sure. So we have a feature now called Generate Variations. An AEM sites that’s really useful for marketers because it uses the power of Gen AI to accelerate content creation, and specifically personalized content creation that is on-brand, and specifically really tailored to those audience segments that you’re targeting. So using Gen AI, you can spin up different variations of content to personalize against your different segments and then deploy them quickly all within AEM. The other benefit we have within AEM is the ability to take these variations, test them, run experiments, and quickly identify what performance gains they’re driving, what conversion uplift they’re potentially driving. So it’s a one-stop solution for you to really create content, deliver it, and make sure it is performing well. Fantastic. You talked a little bit about the audiences and about the segments. Can you expand on that a little bit and talk about where these come from? Sure. So one of the challenges we see with personalization is, you need to create a lot of different content variations, depending on the number of segments you’re trying to target. So for many customers, you have those segments either defined in a solution like ÃÛ¶¹ÊÓƵ Target, or you may have them stored in somewhere else. What this tool allows you to do is we can import segments directly via native integration with ÃÛ¶¹ÊÓƵ Target, import that segment information from Target and use to create the audience-specific variations. If you don’t have Target, you can also define those segments in a CSV file and upload that CSV file into the tool and use that to create the audience-specific content variations as well. Very cool. That’s great that it integrates with Target. Does it also integrate with any of the image generation capabilities that we have on the creative side, like the really cool Firefly stuff? If so, how does that work? What models are you actually using to power this feature? This feature allows for both copy and image generation. For copy generation, we’re using Microsoft Azure OpenAI GPT-4. GPT-4 is the underlying model used for copy. For image generation, we are using ÃÛ¶¹ÊÓƵ Firefly, which is our proprietary image generation technology. The way Firefly is actually incorporated is we have an integration into ÃÛ¶¹ÊÓƵ Express in the tool. You can generate copy directly in the tool, if you want to generate an associated image, you can click a button, it opens up a window for ÃÛ¶¹ÊÓƵ Express where you can then use text-to-image to create the rich images as well. Very cool. Related to that, of course, how are we ensuring that the trust in privacy for the customer data? Is data being used to train the model here? Great question. One big objective of our customers is to use their brand-specific guidelines, their inputs to really get the right output that they need. But they also don’t want to use their data or give their data to the models let any of the proprietary data go somewhere else. We actually store all customer data in isolation. Everything’s done at the prompt level and nothing is actually going and getting stored back into the model. Customer data is isolated, protected. We have an agreement with Microsoft Azure to make sure that trust and privacy is kept at the forefront. Customers should have no concern using this feature, using their data to get the best output that they need. That’s great to hear. Related to that, how much flexibility does a marketer actually have to edit these prompts? They have a lot of flexibility. With the prompting itself, we’ve done a couple of things. We’ve built an out-of-the-box prompt library. We know that prompting is one of the biggest challenges with Gen AI because your output is only as good as your input. Oftentimes, it takes several iterations to get to a very good output. We’ve built enterprise-specific, use case-specific prompts based on our experience working with customers for specific parts of your website, so like a banner or a headline. The prompts give you just the right number of input fields and minimize the amount of information a marketer needs to enter in order to get the best output possible. Now, at the same time, a marketing team can also go and create their own prompts or customize one of the out-of-the-box ones we provide by adding additional input fields. Maybe they want their teams to provide more context. They’re highly customizable but also built to be out-of-the-box and they can start using right away without much engineering required. Fantastic. That’s great. I know that this is done from AM sites Edge Delivery, but can you talk a little bit about how to access some of this content from traditional AM as a Cloud service? Sure. There’s a couple of different ways you can access this feature. One is in Edge Delivery in document-based authoring. When you create your Word document that you’re going to use to power the site, in the AM Sidekick extension, there is a button now that says generate variations, which opens up the tool and allows you to start generating that content with Gen AI. One second. We’re going to be introducing this feature for access directly in the main Cloud service window, where you can click into sites, assets, forms. There’ll be a button called generate variations that allows you to open that up and use those variations directly in traditional AM authoring modes as well. Really, our strategy is to support both the document-based authoring workflow with Edge, but also AM authoring so customers can access this generative capability wherever they are. Fantastic. Well, thank you. This was a great overview. We very much appreciate you taking your time for this. Thanks again, Karthik. Of course. Thank you, James.