The Challenge: Many teams struggle with efficiently training their AI agents without manually extracting and reviewing vast amounts of historical support data.

The Solution: Freddy AI Agent learns through its knowledge base rather than traditional retraining methods. Since large language models power these agents, direct retraining isn't possible. However, every new knowledge base article and Q&A pair you add significantly improves response accuracy and coverage.

Currently, Freddy AI leverages Knowledge Base Articles as its primary learning source. The roadmap includes exciting developments where Freddy will surface insights from incoming tickets, helping account owners create new content that continuously trains the AI on expanding knowledge sets.

Preventing Bot Loops and Ensuring Seamless Handovers

Smart Fallback Mechanisms: When Freddy AI encounters questions it cannot answer, it automatically provides fallback messages. You can configure these moments to seamlessly transfer conversations to human agents, preventing frustrating endless loops.

Proactive Intelligence: The Instructions module enables you to enforce intelligent behaviors, including probing questions and contextual responses. These don't require feeding during setup – they're built into the system's natural learning process.

The Instructions feature includes built-in suggestions for optimal phrasing. Focus on providing business context, such as:

• Industry-specific terminology and meanings

• Brand-specific language preferences

• Common customer scenarios and appropriate responses

Property Collection Capabilities: Freddy can collect essential customer properties at conversation initiation, ensuring security requirements are met before proceeding. Advanced property collection at handover is currently in development, expected to launch within the next few months.

Form Integration: Currently, forms compatible with Freddy AI must meet specific criteria:

• Maximum 8 fields

• Under 100 dropdown choices total

• No dependent fields or multi-select options

• No attachment capabilities

While pre-chat forms are available today, native form integration within AI Agents isn't on the immediate roadmap.

Brand-Specific Deployment: For businesses managing multiple brand websites, we recommend creating unique AI Agents for each brand. This approach offers better scalability and easier future maintenance.

Issue Type Considerations: Unlike brand-specific agents, creating separate AI Agents per issue type typically isn't necessary. The system handles various issue types effectively within a single agent framework.

Freddy AI Agent automatically falls back to human agents when technical issues arise or when answers aren't available in the knowledge base. This ensures continuous customer support without manual intervention.

Advanced Automation Capabilities

Ticket Creation: Through advanced automations, AI Agents can create support tickets at handover points instead of routing to live agents, streamlining your workflow processes.

Email Integration: Email AI Agents for support portal tickets are expected to launch between January and February of next year.

Managing Response Consistency

Understanding LLM Behavior: Since large language models power Freddy AI, occasional response variations are expected. When you notice inaccuracies, modify Instructions or knowledge sources to improve consistency and accuracy.

Evaluation Tools: Regular testing helps identify areas for improvement and ensures your AI Agent maintains high-quality responses across various scenarios.

While traditional chatbots excel at structured, flow-based interactions, Freddy AI Agents deliver human-like conversational experiences that adapt to customer needs dynamically. This flexibility makes them ideal for complex customer service scenarios requiring nuanced understanding.

Getting Started: Your Next Steps

1. Audit Your Knowledge Base: Ensure comprehensive, up-to-date articles covering common customer queries

2. Define Clear Instructions: Establish business context and industry-specific terminology

3. Plan Your Brand Strategy: Consider separate agents for different brands or customer segments

4. Test Thoroughly: Use evaluation tools to refine responses and identify improvement opportunities

5. Monitor and Iterate: Continuously add new knowledge and adjust instructions based on performance