AI Assistant in Adobe Experience Platform product documentation

Transcript

Introducing AI assistant, a natural language interface built into Experience Platform designed to enhance productivity, expand product mastery, and help users efficiently navigate enterprise data objects. But what exactly can I assistant help with? It can quickly share product knowledge, helping users learn concepts and troubleshoot, and it can provide operational insights to assist with lifecycle management, impact and value analysis. All product knowledge answers are verifiable and cited, linking to product documentation.

The suggested prompts make it easy for me to continue the conversation to get an overview of capabilities. I can head over to the discoverability panel where they are clearly outlined. Now that I know what I assistant can help with, I’ll use it to help me clean up my environment. Let’s figure out which of my audiences have never been used in journeys. Wow, that was so easy. Before, I would have had to look through all of my journey definitions to identify used audiences, and then find the remaining audiences that were not used. It would have taken me hours and this took seconds to get a list. I can see how my question was interpreted to ensure no misunderstandings. Navigate to the audience page. Review the step by step process that generated the answer, and even see the SQL that ran behind the scenes. This really helps me trust the results. Before I consider getting rid of some of the unused audiences I just uncovered. I want to see if they’re being used by other audiences. Oh well, it looks like three of them are. I’ll be sure not to delete those.

I am interested to find out what attributes are used in the in segment audience. Isn’t it nice to have all the information at your fingertips? Next, I’ll identify unused journeys. It makes it easy that I can navigate directly to journey. Optimize from the response. Last, on my hygiene journey, I’ll ask AI assistant. Best practices for deleting a schema AI assistant gives me all the information I need, including contextualizing the information for my specific sandbox.

I have learned so much from AI assistant, and I have a rich chat history that I can go back to and visit. AI assistant truly is a shortcut for getting value out of experience. Platform.

Transcript

AI Assistant in Adobe Experience Platform is a conversational experience that you can use to accelerate your workflows in Adobe applications. AI Assistant responds to your submitted questions by querying a database and then translating data from the database into human readable answers. It’s important to note that the product context you are in will determine which products the AI Assistant will consider when returning information. For example, while you are in Adobe Experience Platform, you will get the best result asking questions specific to Adobe Experience Platform. If you want to ask questions about customer journey analytics, your results will be better if you first navigate to CGA within the UI before using AI Assistant. In this video, we are going to explore what AI Assistant can provide in the areas of product knowledge. Product knowledge refers to concepts and topics grounded in experience-led documentation. Product knowledge questions can be further specified into the following subgroups, pointed learning, open discovery, and troubleshooting. With that in mind, let’s get started. If you are not sure what the AI Assistant can do, why not ask it that? Let’s ask, what can the AI Assistant do? As you can see in the response, AI Assistant can guide you with learning concepts and continuing workflows to help ramp up your skills and knowledge, troubleshooting, learn how to debug basic errors that you might encounter, sandbox hygiene, leverage those capabilities to keep your data optimized, value analysis helps you find your most used data objects. Oh yeah, you can search too. You can find information on your audiences, datasets, destinations, schemas, and sources. Be sure to check out the related suggestions at the end of the responses as well. These often can help guide you to great follow-up questions or help inspire you to think through what to ask next. If you are still stuck and not sure what to ask, click on the lightbulb icon to enter the discoverability view. On this view, you will find even more suggested questions that are broken out into categories like operational insights for audiences, datasets, and more. Also there are categories for more product knowledge questions specific to pointed learning and troubleshooting. At the end of the responses, make sure you are checking out the sources of the response. You can show the sources, see the inline citations, and even click on the links to read even more on the topic of interest. But wait, there’s still more to see. Let’s take a look at some of my favorite bonus tips or features that I rely on when using AI Assistant. What’s new? I like to ask what’s new in AEP to see a highlight of the most recent updates and features. This can help you stay current with the evolution of the Adobe Experience platform. Ask for steps. Help AI Assistant help you by being as specific as possible when crafting your questions. Instead of just asking how do I create an audience, ask what are the steps to create an audience. This will provide a set of instructions for you to follow and accomplish the task. Ask anyway. Sometimes we ask AI Assistant questions that it flags as out of scope and returns a message that informs you of this and offers some suggestions on adjusting your question. But you can leverage the ask anyway suggested prompt to bypass the out of scope message and try your question anyway. Sometimes you might land on answers that you are looking for, but be sure to double check the sources used in the answers. Your mileage may vary, but it’s worth a shot. Hopefully you now have a better understanding of how to use AI Assistant within Adobe Experience platform to expand your product knowledge. What are you going to ask Adobe AI Assistant?

Transcript

Sometimes the hardest part of using AI Assistant is just knowing where to get started. That is where the discoverability panel within Adobe’s AI Assistant comes in. Just click on the light bulb at the top of the AI Assistant to start using the discoverability panel. On this view, you will find suggested questions that are separated into categories like operational insights for audiences, data sets, destinations, and more. Also, there are categories for product knowledge questions specific to pointed learning and troubleshooting. Let’s try one of the suggested questions out. Which audiences have zero profiles? Click on the suggested question you want to try. This will copy the question to the dialog box where you could make any modifications if you wanted to. Then submit the prompt. A few moments later, and now we have our list of all the audiences with zero profiles that we can now take action on. Take advantage of these suggested questions to get a better understanding of how AI Assistant with an Adobe Experience platform can help you.

Transcript

In this video, we will review the security details for the Adobe Experience Platform AI Assistant. Please note that this video is based on the current security fact sheet located at the link on the screen. If there is ever a discrepancy between this video and the current security fact sheet, then the fact sheet takes precedence. AI Assistant and Adobe Experience Platform is a generative AI tool integrated within native applications built on Adobe Experience Platform. Designed to enhance productivity and help users expand product mastery, efficiently navigate enterprise data objects, and simplify tasks while ensuring adherence to the customer’s organization data security standards. AI Assistant and Adobe Experience Platform can answer questions about product knowledge and operational insights. Adobe’s agnostic approach to large language model enables us to choose the best-in-class technology for the task at hand. AI Assistant and Adobe Experience Platform currently leverages Microsoft’s Azure Open AI service to answer both product knowledge and operational insights questions. There are three key components in AI Assistant. The Adobe Experience Platform user interface. Users interact with the AI Assistant by clicking the icon in the upper right-hand corner of the Adobe Experience Platform UI, which reveals a right rail screen with a text box where the users can enter prompts. Generative Experience Models, or GEMs, the primary brains behind AI Assistant and Adobe Experience Platform. The GEMs include foundation and custom models that power AI Assistant use cases. For the details on the specific models, please see the security fact sheet. Data Services. API services invoked by GEMs to query the data stores that contain relevant data. Data in the data stores is organized, pre-joined, and indexed into a knowledge base, which then enables the GEMs to interact with it in an open-ended fashion. To enable a user to access AI Assistant and AEP, the customer’s Adobe admin must grant specific permissions. For real-time customer data platform and Adobe Journey optimized users, the Adobe admin must grant permissions within the permissions UI of the Adobe Experience Platform. For customer Journey analytics users, the Adobe admin must grant permission for the users to access the AI Assistant within the Adobe admin console. For more information, please review the security fact sheet. Data Encryption. In transit, all data is encrypted in transit over HTTPS using TLS 1.2 or greater. At Rest. Any data stored by AI Assistant is encrypted at rest using AES 256-bit encryption. All data is encrypted in transit over HTTPS using TLS 1.2 or greater. Step 1. User opens the AI Assistant and Adobe Experience Platform user interface. Step 2. AI Assistant authenticates the user with Adobe Identity Management Services, IMS, and checks that the user is entitled to use the AI Assistant. Step 3. Users enter a product knowledge type question in the prompt text box. Step 4. AI Assistant UI sends the prompt text to the dialog management GEM, which classifies the prompt into the appropriate question type, product knowledge, operational insight, or out of scope. If the question is in scope for AI Assistant and AEP, the process moves to step 5. If the question is out of scope, the user receives an error message. Step 5. The dialog management GEM checks with the AEP access control service to confirm that the user is entitled to ask product knowledge questions. Questions outside the scope of AEP and its native applications, including questions about other Adobe products such as Adobe Target and the Creative Cloud Suite, cannot be answered by the AI Assistant in AEP. Step 6. If the user is entitled, the dialog management GEM applies a series of content filters to determine if the prompt adheres to Adobe’s generative AI user guidelines. If any part of the prompt violates these guidelines, the user receives an error message. Step 7. The dialog management GEM then sends the prompt text to the product knowledge GEM, which uses semantic search to retrieve relevant snippets of documentation from the product knowledge data service to answer the question. Step 8. The dialog management GEM combines the prompt text with the retrieved snippets of documentation from the product knowledge data service and sends them to the Azure OpenAI service. Step 9. Before sending the formulated answer back to the dialog management GEM, the Azure OpenAI content filtering service moderates generated responses that violate Azure OpenAI user guidelines. Step 10. The product knowledge GEM cross-checks the answers provided by the Azure OpenAI service against the documentation snippets, adds the appropriate citations, and sends the complete answer and citations to the dialog management GEM. Step 11. The dialog management GEM returns the answer and the relevant citations, along with suggested next prompts, to the user in the AI Assistant for Adobe Experience Platform user interface. Now let’s look at the Data Flow Narrative for Operational Insights. Step 1. The user opens the AI Assistant in the Adobe Experience Platform user interface. Step 2. AI Assistant authenticates the user with Adobe Identity Management Service and checks that the user is entitled to use AI Assistant in Adobe Experience Platform. Step 3. User enters an operational insights type question in the prompt text box. Step 4. AI Assistant sends the prompt text to the dialog management GEM, which classifies the prompt into the appropriate question type, product knowledge, operational insight, or out of scope. If the question is in scope for AI Assistant in Adobe Experience Platform, then the process moves to step 5. If the question is out of scope, the user receives an error message. Step 5. The dialog management GEM checks with the AEP Access Control Service to confirm that the user is entitled to ask operational insights questions. Step 6. If the user is entitled, the dialog management GEM applies a series of content filters to determine if the prompt adheres to Adobe’s generative AI user guidelines. If any part of the prompt violates these guidelines, the user receives an error message. Step 7. The dialog management GEM sends the prompt text to the operational insights GEM, which retrieves a customer agnostic schema and sample queries relevant to the current prompt. Step 8. The dialog management GEM combines the prompt text with the customer agnostic schema and sample queries and sends the data to the Azure OpenAI service, which uses the information to formulate an answer. Step 9. Before sending the formulated answer back to the operational insights GEM, the Azure OpenAI Content Filtering Service moderates generated responses that violate Azure OpenAI user guidelines. Step 10. The operational insights GEM applies the relevant permissions on the business objects present in the query using role-based access control and object attribute level access controls. Step 11. The operational insights GEM runs the query in the context of the customer’s operational insights data service and generates an intermediate response, which is typically a single or multiple row table. Step 12. The operational insights GEM sends the query and the intermediate response to the Azure OpenAI service, which generates the natural language description of the answer and provides the natural language explanation of the query. This step-by-step explanation helps the user to verify the query’s accuracy. Step 13. The dialog management GEM returns the answer to the user. AI Assistant in Adobe Experience Platform and Azure OpenAI. AI Assistant in Adobe Experience Platform currently leverages Azure’s OpenAI to answer customer questions. The following data may be passed to Azure OpenAI to facilitate answering product knowledge or operational insight questions. Experience League Documentation. Information related to the page that the user is on. User’s Conversation History. The prompts and answers. The following data may be passed to Azure’s OpenAI to facilitate entering operational insights questions only. The schema of the tables being queried. Example questions with ground truth queries. Attributes within application business objects such as the name, description, and counts. Adobe has disabled logging in Azure OpenAI, helping to ensure that Microsoft does not collect or review any data sent for processing to Azure OpenAI by the AI Assistant in Adobe Experience Platform. More information is available at the Azure OpenAI Data Privacy and Security link. Adobe does not use any customer data to train or fine-tune the Azure OpenAI service. Chat History. Users can access their AI Assistant in Adobe Experience Platform Chat History, including the prompt and answer for 30 days. Chat History is stored in the same data center as the customer’s Adobe Data Storage location. If a customer would like to delete a user’s chat history, they should contact their Adobe Customer Support representative. Data Usage. Adobe uses customer-agnostic annotated data to fine-tune Adobe internal models. For example, the linguistic models for documentation and the operational insights and classifier models for prompt classification or out-of-scope detection. The responses from these models are not shown directly to the users. Data Processing and Storage Locations. Adobe Identity Management Services. Regardless of the geographic location of the customer, all identity data is stored in multi-region, load-balanced, cloud infrastructure providers with data centers located in North America, Europe, and APAC. Identity data is replicated across all data centers for reliability reasons. All identity data is secured at rest using AES 256-bit encryption in compliance with the Adobe Common Controls framework and meets our internal policies for encryption and storage of sensitive data. AI Assistant and Adobe Experience Platform and Azure OpenAI Service. All server-side components of AI Assistant and Adobe Experience Platform and corresponding data storage are co-located in the same region as the customer’s Adobe Experience Platform service infrastructure, which is determined upon initial provisioning. Data sent to the Azure OpenAI Service may be processed in a different data center but located within the same geographical region, per the tables that are in the security fact sheet. Questions. If you have any additional questions about the security posture and capabilities of Adobe Experience Platform, native applications, or AI Assistant in Adobe Experience Platform, please contact your Adobe account manager. For all other questions about Adobe security programs and processes and compliance certifications, please visit the Adobe Trust Center. Also, be sure to bookmark the security fact sheet for Adobe AI Assistant so that you can refer to it in the future.