Summary
Astra AI Agents can help automate customer conversations across different channels while providing responses based on your business knowledge.
This guide explains recommended practices for configuring knowledge sources, managing multiple business domains, using agents across channels, and reducing the risk of out-of-scope responses.
Instructions
Organize knowledge by business domain
If your business operates across multiple products, services, departments, or knowledge areas, you do not need to create a separate Astra AI Agent for each one.
Astra supports knowledge segmentation through its Focused Search capability. This allows a single AI Agent to access multiple knowledge domains and automatically retrieve information from the most relevant source based on the customer's question.
Examples of knowledge domains include:
Product information
Billing and payments
Customer support policies
Service-specific documentation
Industry-specific information
This approach allows you to manage multiple knowledge areas from a single AI Agent while maintaining accurate responses.
Use separate agents for WhatsApp and web chat
Currently, Astra requires separate AI Agents for:
WhatsApp
Web chat widgets
Each agent has its own:
Knowledge base
Configuration settings
Instructions
To maintain a consistent customer experience, use the same instructions, knowledge sources, and response guidelines across both agents.
While responses are generally very similar, minor variations may occur because AI-generated responses are not deterministic.
Restrict responses to your knowledge base
You can configure Astra to answer questions using only information from your uploaded knowledge sources.
To achieve this, add clear instructions such as:
Answer only using information available in the knowledge base.
Do not generate information that is not present in the knowledge base.
Respond with a predefined message when the requested information is unavailable.
When combined with Retrieval-Augmented Generation (RAG), Astra retrieves relevant information from your uploaded content before generating a response.
This helps ensure that responses remain grounded in your organization's approved information.
Understand the limitations of AI-generated responses
Although Astra can be strongly guided to use only approved knowledge, no Large Language Model (LLM)-based system can guarantee the complete elimination of inaccurate or unsupported responses in every scenario.
This is a limitation of generative AI technology and is not specific to Astra.
With proper configuration, out-of-scope responses are uncommon, but additional safeguards are recommended.
Recommended safeguards
To improve response quality and reduce risk, consider implementing the following practices:
Add clear instructions that prohibit answering questions outside the available knowledge base.
Configure human handoff workflows for situations where the AI Agent is uncertain or unable to find relevant information.
Include appropriate disclaimers, terms, or informational notices when required by your business or industry.
Regularly review and update your knowledge base to ensure information remains accurate and current.
Conclusion
A well-configured Astra AI Agent can help deliver accurate, consistent, and scalable customer support experiences. By organizing your knowledge effectively, configuring clear instructions, and implementing safeguards such as human handoff, you can improve response quality while ensuring customers receive information that aligns with your business requirements.
For best results, keep your knowledge base up to date, regularly review agent responses, and use clear guidance to help the AI stay focused on approved information sources.