AI response and resolution metrics

AI performance metrics tell you how well your automated communication layer is working. Key measurements: response time (how fast the AI responds to inbound messages), resolution rate (what percentage of conversations are fully handled by AI without requiring a human), and escalation rate (how often the AI sends conversations to your team). Tracking these over time reveals whether your AI configuration is working well or needs adjustment.

Response time

Rezi measures AI response time as the elapsed time between when an inbound message is received and when the AI sends its first reply. This metric is captured for every conversation. The Reports > AI Performance page shows average response time by day, week, and month, and breaks it down by property. For most accounts using Rezi's AI, average response time is under 30 seconds.

High average response times can indicate a configuration issue (AI model timeout, slow network), or they may reflect that a high proportion of your conversations require the AI to search through a large knowledge base before responding. Review your knowledge base content if response times are consistently above 60 seconds.

Resolution rate

A conversation is considered AI-resolved when it reaches a final state (guest question answered, check-in handled) without a team member sending a manual reply. Resolution rate is the percentage of conversations that end in AI resolution. A well-configured Rezi account should achieve 70-85% AI resolution rate for common questions. Lower rates suggest the AI knowledge base needs more content.

Escalation rate

The escalation rate is the inverse of the resolution rate, the percentage of conversations where the AI determined human involvement was needed and escalated. Not all escalations are bad: complex situations (refund requests, emergencies, ambiguous requests) should escalate. A high escalation rate for simple topics (wifi, parking, checkout time) suggests the knowledge base is missing information the AI needs to answer those questions.

Guest satisfaction signals

Rezi tracks two satisfaction signals: follow-up rate (guests who send a second message soon after an AI reply, suggesting the first response was not sufficient) and re-escalation rate (conversations that escalated, had human involvement, then re-escalated, suggesting the first human response did not resolve the issue). These metrics help identify patterns in where the communication system is breaking down.

Using metrics to improve AI performance

Low resolution rates and high escalation rates for specific topics point directly to knowledge base gaps. If 40% of escalations are about parking, add a comprehensive parking knowledge base entry. If response times spike on weekends, check whether there is a system bottleneck during those periods. Metrics translate directly into actionable improvements.