first pass

Author: rchinn1

src/engage/audiences/generative-audiences.md

@@ -0,0 +1,126 @@
+---
+title: Generative Audiences
+
+---
+
+Generative Audiences, part of Segment's CustomerAI, uses generative AI to create Engage Audiences with natural language prompts. 
+
+Build audiences by describing your desired audience based on events performed, profile traits, or referencing existing audiences in your workspace. Based on your prompt, Segment builds the audience with Generative AI.
+
+In this article, you'll learn how to use Generative Audiences along with some best practices.
+
+<!-- CUT
+## Build an audience with Generative AI
+-->
+
+## Getting started with CustomerAI
+Use these steps to get started with CustomerAI:
+<!-- ask for a demo of this process -->
+1. From your message composer, select **Launch AI Copywriter**.
+2. In the drag-and-drop editor for in-app messages, select a text block and select in the block's toolbar.
+3. Enter a product name or description in the input field.
+4. Select the output language. This can be different from your input language.
+5. Select an appoximate output length.
+6. Click **Generate Copy**.
+
+## Create an audience with Generative Audiences 
+
+To use Generative Audiences to create your audience:
+
+1. Navigate to **Engage > Audiences**, and click **+ New audience**.
+2. From the dropdown menu, select **Audience**.
+3. Select **Generative Audiences** as your audience type. Generative Audiences is available in all audience types except Linked Audiences.
+4. From the Build screen, click **Help me build**.
+5. Enter your prompt in the audience description box. 
+- Use a minimum of 20 characters and up to 300 characters maximum.
+6. Click **Build**. Based on your prompt, CustomerAI generates audience conditions for you to review. 
+- Segment displays a progress bar until the audience conditions are generated.
+
+> success ""
+> View some [example prompts](#use-cases-and-examples) and [best practices](#best-practices).
+
+### Modify an audience description 
+
+Once Segment generates the audience conditions, the prompt box will remain open for reference. You can close this box, or modify your audience description and click **Build** again. 
+
+Modifying an audience overwrites the existing audience conditions Segment previously generated. 
+
+> success ""
+> Once Segment generates the audience, you can adjust or remove any conditions straight from the audience builder. 
+
+
+## Use cases and examples
+
+View the table below for a few example prompts:
+
+
+| Audience description   | Prompt           |
+|------------------------|--------------------------------|
+| Build an audience with customers who haven't made a purchase in the last 30 days | "Customers who haven't purchased in the last 30 days."  |
+| Build an audience with customers who spend over $50 on an order | "Customers who have orders greater than $50."   | 
+| Find all profiles that have recently opened an email. | "Profiles that recently opened an email."   |
+
+
+## Best practices
+
+As you use generative audiences, keep the following best practices in mind:
+
+- Write specific descriptions. CustomerAI generates more accurate conditions when you use the names of existing events and traits. <!-- should I say "customerAI generates" or "Generative Audiences generates"? -->
+- Ensure that all events and traits you reference exist in your workspace.
+- Avoid using sensitive data, or customer Personal Identifiable Information (PII).
+- Experiment with your prompts. If you don't receive what you want on the first try, rewrite your prompt. Submitting a new prompt replaces existing conditions.
+
+> success ""
+> Segment recommends previewing your audience to ensure you're matching with the correct profiles prior to moving on to the next step.
+
+
+## Known limitations
+
+### Limited Space Schema 
+
+Segment's generative AI service is handled by a third party that needs to be fed context about your Engage workspace and has limitations to how many contextual parameters Segment can send it. Because of this, Segment limits the amount of data it sends to the AI service. Segment solves this limitation by including up to 100,000 of the most recently used events and traits in your Engage space.
+
+Event properties may not be populated due to this.
+
+### Foreign language support
+
+At this time, Segment only supports audience description prompts in the English language. Foreign language support is currently unavailable and might provide undesired results. 
+
+### Negative conditions
+
+Audience descriptions around not having profile traits or users who have not performed certain events are expected to have their conditions configured in the following way:
+
+- "Customers who have not purchased in the last 30 days."
+
+![Creating an audience where customers haven't made a purchase in the last 30 days.](/docs/engage/images/No-purchases.png)
+The expected output is for Segment to generate a condition where the *event is performed at most 0 times*.
+
+- "Customers who don't have a phone number."
+
+![Creating an audience where customers don't have a phone number.](/docs/engage/images/phone-doesn't-exist.png)
+
+The expected output is to generate a condition where the *trait doesn't exist*.
+
+
+## Error handling
+
+Engage uses the following error messages with Generative Audiences:
+
+| Error message        | Cause                  |
+|---------------------------|---------------------------------------|
+| Something went wrong      | This error displays when one of the following occur: <br>  - Not found exception <br> - Invalid LLM output <br> - LLM error terminal <br> - Uknown exception                  |
+| Something went wrong. Try again later. | The AI service is down or the LLM returned an error. |
+| Segment had trouble creating an audience from this description. Try rewording it using these [best practices](#best-practices). | Referenced an invalid or non-existing trait, audience, or event within the workspace. You may also see this when an objective is impossible or misunderstood. |
+
+
+## More about Segment's Generative AI service
+
+#### How is my data used?
+
+In order to generate the audience based on your description, Segment sends your query to OpenAI, Segment's 3rd party AI service. All queries sent to OpenAI from Segment are anonymized, meaning that OpenAI won't be able to identify from whom the query was sent unless you include uniquely identifiable information in the input.
+
+Behind the scenes, Segment instructs GPT to generate an audience based on required parameters from Segment's system and contextual information such as traits and events performed in your workspace.
+
+GPT is OpenAI's state of the art natural language generation tool powered by artificial intelligence. It can perform a variety of natural language tasks like text generation, completion, and classification. CustomerAI uses the service to help generate audiences and inspire segmentation.
+
+Per OpenAI's policy, data sent to OpenAI's API with Segment isn't used to train or improve their models and will be deleted after 30 days. Any content generated using GPT is your intellectual property. Segment won't assert any claims of copyright ownership on such content and makes no warranty of any kind with respect to any AI generated content.