Overview
Every LuMay agent ships with a measurement plane. Containment rate, escalation rate, connector success rate, sentiment, cost per outcome, and the specific business KPIs you define - these are wired in from day one. They don't require a separate analytics project or a reporting sprint at quarter-end.
The measurement plane is powered by OpenTelemetry instrumentation embedded in the platform. Every API call, database query, RabbitMQ message, and connector dispatch emits a trace span. Every call session produces a complete trace with timing at every hop. If an agent is slow, you can see exactly which step is the bottleneck - the STT transcription, the LLM call, the CRM lookup, or something else entirely.
This also means the platform can show you the cost case both ways. If an agent is resolving 70% of calls without human transfer, that shows as containment. If the agent is regularly producing low-confidence responses and escalating to a human, that shows as escalation rate. If it doesn't save money, you'll see that in the data - not at the end of a quarter, but in real time.
Built-in dashboard signals
Containment
Resolved without transferConnector Success
System actions completedCost per Outcome
Unit economics over timeSentiment
Experience quality signalDesign principle
Show the dashboard as a feedback loop - LuMay as a learning operating system, not just a reporting tool.
Built-In KPIs
| KPI | What it measures | Why it matters |
|---|---|---|
| Containment Rate | % of calls resolved without human transfer | The core automation metric - directly measures whether the agent is replacing human effort |
| Escalation Rate | % of interactions transferred to a person | Highlights gaps in the conversation flow and agent knowledge base |
| Total Calls Handled | Volume of call sessions processed by the agent | Volume baseline for capacity planning and cost modelling |
| Avg Call Duration | Mean length of agent-handled calls | Long calls often signal flow inefficiency or missing connector integrations |
| Connector Success Rate | % of tool calls (CRM, ticketing, etc.) that completed successfully | Tracks integration health - connector failures silently degrade agent quality |
| Outbound Answer Rate | % of outbound calls answered by a live person | Key efficiency metric for outbound campaign agents |
| Tickets / CRM Updates Created | Count of completed work items created by agents | Direct measure of business value: work that previously required a human operator |
| Sentiment Distribution | Positive / neutral / negative split across call sessions | Quality signal - deteriorating sentiment often precedes escalation rate increases |
Observability Stack
The platform ships with a full observability stack that runs identically in every deployment model:
- OpenTelemetry instrumentation - embedded in both the Management API and Voice Agent Engine. Every request, database query, queue message, and connector call emits a trace span with timing, status, and context.
- Grafana Tempo - distributed tracing backend. Per-call traces are viewable end-to-end: from the inbound Twilio webhook through STT, LLM reasoning, every connector call, and TTS synthesis.
- Prometheus - metrics collection at the
/metricsendpoint. Includesactive_calls_total(used for Voice Engine auto-scaling), request latency histograms, error rates, and connector call counters. - Loki - structured log aggregation. Every log line includes the X-Correlation-ID so logs, traces, and metrics can be joined for a single interaction.
- Grafana dashboards - pre-built dashboards for containment, escalation, connector health, and cost per call. Customisable per tenant.
What Is And Isn't Measured Today
The KPIs above are available from day one of a deployment, using the platform's built-in instrumentation. The following metrics require post-call survey integrations that are on the product roadmap and not yet available as built-in features:
- CSAT (Customer Satisfaction Score) - requires a post-call survey integration (IVR or SMS). Sentiment scoring is the closest available signal today.
- NPS (Net Promoter Score) - requires a survey tool integration. Currently tracked as a custom KPI when the client configures the relevant connector.
- Appointment conversion rate - available when the agent is configured with a calendar or scheduling connector and the outcome is recorded.
Related
- Architecture - the Analytics & ROI layer (06) in the six-layer stack
- Deployment - the same observability stack runs in every deployment model
- Voice Engine - the source of per-call traces, sentiment, and containment data
- Platform overview - all four engines and four trust pillars