AI Support
Ingestion and Evaluation
Recommended metadata and evaluation checklist for the Lexul support AI agent.
Appendix B: Recommended ingestion metadata
| Metadata field | Recommended value |
|---|---|
| product | Lexul Field Service |
| content_type | customer_support_chatbot_training_record |
| audience | Lexul customer support chatbot, support team, onboarding team |
| chunking | One sub-feature training card per chunk; preserve record_id and source_urls. |
| retrieval_priority | Prefer exact feature/sub_feature match; then alias/sample utterance match; then source URL match. |
| answer_style | Concise, workflow-specific, ask one clarifying question when needed, include guardrail if relevant. |
| escalation_behavior | Escalate sync/data integrity, accounting advice, permission/security changes, custom development, unsupported roadmap requests. |
Appendix C: Agent evaluation checklist
- Can the agent answer a setup question using the correct feature record?
- Can the agent choose between related features such as contracts, recurring work orders, and summary invoicing?
- Can the agent ask only one focused clarification question when QuickBooks version, user role, customer/asset structure, or billing workflow matters?
- Can the agent avoid unsupported claims and escalate custom-development, accounting, sync, and data-integrity issues?
- Can the agent cite or surface the relevant Lexul source URL when responding?