Use the segment AI assistant in Amazon Connect - Amazon Connect
Inspiration Cards for Segment CreationGenerate a segment by promptProvide feedback on generated segmentsError handlingKnown limitations

Use the segment AI assistant in Amazon Connect

Amazon Connect Customer Profiles supports generative AI-powered segmentation, enabling non-technical business users to build audiences using natural language queries (segment AI assistant), and to receive recommendations based on trends in the customer data (inspiration cards for segment creation). These capabilities leverage advanced AI algorithms from Amazon Bedrock that help you improve customer satisfaction and drive revenue through proactive and personalized outreach. For example, you can create a segment of customers who reached out to customer support frequently last week with personalized service offers. You can also identify customers whose total spending increased and offer personalized discounts, fostering loyalty and also driving growth.

The following benefits are added by incorporating generative AI into the segmentation workflow:

  • Simplified segment creation: Build complex customer segments using conversational language, making the process accessible to non-technical users and driving the efficiency.

  • Data-driven segment creation inspirations: Receive AI-empowered segment inspirations based on trends in the customer data.

  • Enhanced personalization: Easily identify and target specific customer groups for tailored communications and offers.

The following sections explain each feature, how to use them, and the benefits they provide to help you improve your customer segmentation efforts.

Note

  • To use the segment AI assistant, users will need the permission for segment creation CustomerProfiles.Segments.Create.

  • While these AI-powered tools offer valuable suggestions, it's important to review and adjust the recommended segments to ensure they align with the organization's specific business objectives and comply with its data usage policies.

Inspiration Cards for Segment Creation

Inspiration Cards are an AI-powered feature on the Customer segments page. They simplify and enhance the segment creation process. The following image shows an example of three inspiration cards.

An example of inspiration cards on the Customer segments page.

These cards generate up to three categories of segment ideas each time, based on the Amazon Connect Customer Profile data to inspire and streamline your segment creation process. 

Note

The trend data is based on event ingestion dates of default calculated attributes.

Key features

  • Data-driven inspirations: Each inspiration card presents a segment idea tailored to specific customer data and trends.

  • Inspiration cards offer ideas across three business-focused themes:

    • Promotion: Ideas for targeting customers with specific promotional strategies.

    • Retention: Identify segments for customer retention efforts.

    • Support: Highlight customer groups that may need specialized attention for customer service.

  • Insight-based recommendations: Leverage historical trends, data insights and generative AI to create meaningful, actionable insights.

How to use inspiration cards

  1. Navigate to the Customer segments page.

  2. Locate the inspiration cards section. It displays three segment suggestions.

  3. Review each card to understand the proposed segment and its potential applications.

  4. When you find a card you want to use, choose Get started on that card.

  5. Choose Explore more to generate additional inspiration cards. These can offer fresh segment ideas based on your Amazon Connect Customer Profiles data.

  6. When you choose Get started, you are automatically directed to the Create Segment page.

  7. Your selected segment idea is populated in the segment builder, ready for your review and refinement.

Generate a segment by using natural language prompts

The segment AI assistant offers a guided approach to creating segments using natural language prompts, which simplifies the process of creating complex segments, allowing you to describe your target audience in natural language and receive a structured, actionable segment definition.

The following image shows an example of a segment AI assistant prompt.

An example of segment AI assistant prompts.

To access this feature:

  1. Navigate to the Customer segment page, and then choose Create segment.

  2. Locate the segment AI assistant panel on the right side of the page, as shown in the following image.

An example of the segment AI assistant panel on the right side of the page.

Use the segment AI assistant

  1. The assistant guides users through a series of questions to understand segmentation needs, all interaction paths with the assistant lead to generating a prompt.

  2. Users can provide a textual description of the desired segment.

  3. The prompt action step offers sample prompts as references for writing detailed descriptions.

  4. Based on your input, Amazon Connect generates a structured segment definition.

  5. The generated segment definition is automatically applied to the segment builder.

  6. You can further refine the generated segment using the standard segment builder tools. Modifying the filters on the segment builder overwrites the existing conditions previously generated. 

  7. After reviewing the generated segment and making any necessary adjustments, you can finalize the process by choosing Create segment. This action saves your segment and makes it available for use in your campaigns.

Best practices

As you use the segment AI assistant, keep the following best practices in mind:

  • Write specific descriptions. Segment AI assistant generates more accurate conditions when you use the names of existing attributes.

  • Ensure that all attributes you reference exist in your domain.

  • Start with simple prompts and try different prompts. If you don't receive what you want on the first try, rewrite your prompt. Submitting a new prompt replaces existing conditions, or by choosing New conversation.

  • Allocate time for segment refinement and validation on the segment builder to ensure segments accurately reflect your actual data values.

Note

The segment AI assistant is designed to work with general descriptors and criteria. Always adhere to data protection regulations and company policies when describing segments. Ensure that your prompts and descriptions do not contain any sensitive or personal information. 

Provide feedback on generated segments

After a segment is generated, users are encouraged to evaluate the feature's performance and provide feedback. This feedback mechanism helps improve the segment generation process and ensures it meets business needs effectively. The following image shows a feedback page.

An example of a message that your feedback has been recorded.

The feedback process consists of two stages:

  1. Initial reaction: In the bottom right corner of the alert section, you'll find thumbs up and thumbs down icons. Click on either of these to indicate your general satisfaction with the generated segment.

  2. Additional feedback: After selecting either the thumbs up or thumbs down icon, you are presented with an option to provide more detailed feedback. This takes the form of a text input field where you can leave free-form comments.

We encourage you to use both the quick reaction (thumbs up/down) and the text input for a comprehensive evaluation, provide specific examples or use cases when applicable, focus on how the generated segment aligns with business objectives, and suggest improvements or additional features that would enhance the segment generation process.

Suggest improvements or additional features that would enhance the segment generation process.

By actively participating in the feedback process, users contribute to the continuous improvement of the segment generation feature, ultimately leading to more effective customer segmentation and targeted marketing strategies.

Error handling

When using the segment AI assistant to generate customer segments, you may occasionally encounter an error message stating: We can't process your request right now. This error can occur even after providing a valid prompt for segment creation. Use the following steps help you understand and troubleshoot this error.

Possible causes:

  • High system load: The segment AI assistant may be experiencing high demand or processing multiple requests simultaneously.

  • Temporary service disruption: There could be a brief interruption in the service's ability to process new segment requests.

  • Complex query: The system might need more time to process particularly complex or resource-intensive segment queries.

What to do:

  • Wait and retry: The error message suggests waiting a few minutes before trying again. This allows time for the system to resolve any temporary issues.

  • Create a segment manually: If you need the segment immediately, you can opt to create it manually using the segment builder

How to retry:

  1. Wait for a few minutes.

  2. Choose New conversation at the bottom of the chat interface.

  3. Start a new conversation and re-enter your segment creation prompt.

Best practices

  • If the error persists after multiple attempts, consider simplifying your segment criteria or breaking it down into smaller, more manageable requests.

  • Try to avoid making repeated requests in quick succession, as this may contribute to system overload.

  • If the issue continues, reach out to customer support for further assistance.

Remember, this error is typically temporary, and following the provided instructions should allow you to successfully create your desired segment.

Known limitations

Understanding the data processing lifecycle is crucial for effective use of the segment AI assistant. This section outlines what business users can expect during different phases of data integration and how it affects segment suggestions.

Data processing and quality impact: Segment AI assistant evolves through two main phases: initial data ingestion and post-processing. During initial ingestion, the system may not fully utilize actual attribute values, relying more on prompt interpretation. For example, a prompt for VIP customers might suggest a VIP segment instead of using the existing Gold tier from your data. After complete processing, the system leverages actual attribute values, resulting in more accurate segment creation, reduces reliance on prompt interpretation and improves overall segmentation quality.

Note

Allow sufficient time for complete data processing before relying on advanced features. Regularly update customer profile data. Segment accuracy depends on the completeness and recency of customer data in Amazon Connect Customer Profiles. The system flags any missing attributes in its responses.

System performance: During high-volume periods, expect potential delays in segment generation. The system is optimized for typical workloads, but businesses with extensive segmentation needs may need to adjust their processes accordingly.

Attribute availability: The quality of the generated segments is dependent on the customer data available in Amazon Connect Customer Profiles. The more comprehensive and up-to-date the customer profile data, the more accurate the system can be in interpreting prompts and defining relevant segments. If there is an attribute that does not exist, we will return a message with the missing attribute. 

Prompt complexity: For very complex or nuanced segment definitions, the natural language processing may have limitations. Customers should start with relatively straightforward prompts and gradually increase the complexity as they gain experience with the feature.

Segment refinement: Although the system-generated segments are a great starting point, customers may still want to review and refine the details to ensure the segment aligns perfectly with their business objectives. The segmentation interface allows for full customization after the initial generation.

Performance and scaling: Under high concurrency, there may be some latency in the segment generation process as the language model needs to process each prompt. The system is designed to handle typical segmentation workloads, but customers with extremely high segmentation demands may need to adjust their workflows accordingly