Analytics and reporting

Saaya ships the dashboards every voice / chat team rebuilds badly: answer rate, average handle time, conversion by intent, sentiment by agent version, latency percentiles, cost per session. Cohorts and exports are built in.

Out-of-the-box dashboards

• Volume, sessions per channel, per agent, per version, per hour.
• Quality, answer rate, completion rate, sentiment distribution.
• Performance, p50 / p95 latency for STT, LLM, TTS, end-to-end.
• Cost, per token, per minute, per agent, per session.
• Reliability, tool error rate, retry rate, escalation rate.

Cohorts

Define cohorts on any metadata you sent with the session: `plan="pro"`, `region="IN"`, `campaign="april-refund"`. Saaya keeps cohorts as first-class objects so you can compare them across dashboards without re-defining the filter every time.

Comparisons

Every chart supports a side-by-side comparison: this week vs. last week, version A vs. version B, voice vs. WhatsApp. Statistical significance is computed on conversion-style metrics so you do not chase noise.

Exports

await saaya.analytics.export({
  metrics: ["sessions", "completion_rate", "avg_sentiment", "cost_usd"],
  groupBy: ["agentId", "channel", "day"],
  since:   "2026-04-01",
  format:  "csv",
  destination: { type: "s3", uri: "s3://acme-warehouse/saaya/" },
});