Queue Metrics & CDR

Every queue call writes one row to a durable CDR timing layer in the PresenceDO, and the standard contact-center KPIs (ASA, service level, abandon rate, and AHT) are derived from those four timestamps. As with agent KPIs, there’s no separate aggregation job: the metrics are a pure read over the CDR within a time window.

The queue CDR

CREATE TABLE queue_cdr(
  call_id      TEXT PRIMARY KEY,
  queue_id     TEXT,
  agent_id     TEXT,        -- who answered (NULL until answered)
  enqueued_at  INTEGER,     -- entered the queue (epoch ms)
  connected_at INTEGER,     -- answered by an agent (NULL until answered)
  ended_at     INTEGER,     -- talk ended (NULL until the call ends)
  disposition  TEXT         -- offered | answered | abandoned | timeout
);

One row per offered call, written and updated across the call’s life by the same single-threaded brain that runs the ACD fence:

Lifecycle event	Effect on the row
Enqueue	Insert with `enqueued_at` and `disposition = offered`.
Answer (agent confirms)	Set `connected_at`, `agent_id`, `disposition = answered`.
Abandon / timeout (before answer)	Set `disposition = abandoned` or `timeout` and `ended_at`.
Hang-up (after answer)	Set `ended_at` (the talk-end).

The three timestamps enqueued_at → connected_at → ended_at are all you need to derive every KPI below.

The KPIs

All metrics report over [since, now] and share a service-level threshold parameter, slSec (default 20 seconds).

Volumes

offered   = rows for the queue in the window
answered  = rows where disposition = answered
abandoned = rows where disposition ∈ { abandoned, timeout }

Abandon rate

abandonRatePct = round( 100 × abandoned / offered )          // 0 if offered = 0

ASA: Average Speed of Answer

The mean wait before an agent answered, over answered calls:

wait  = connected_at − enqueued_at
asaMs = round( mean(wait over answered calls) )              // 0 if none answered

Service level

The share of all offered calls answered within slSec:

slMet           = answered calls where (connected_at − enqueued_at) ≤ slSec×1000
serviceLevelPct = round( 100 × slMet / offered )             // 0 if offered = 0

Note the denominator is offered, not answered: an abandoned call counts against service level, which is the stricter (and standard contact-center) definition.

AHT: Average Handle Time

The mean talk time over calls that have ended:

ahtMs = round( mean(ended_at − connected_at over answered calls with ended_at) )

AHT here is talk time. It does not yet fold in the agent’s post-call wrap_up; that’s tracked separately in agent state-time and is a planned addition to AHT.

Reading the report

Queue KPIs are read from the public API at GET /v1/queues (the queues scope):

{
  "queues": [
    {
      "queue": "q_support",
      "offered": 120,
      "answered": 108,
      "abandoned": 12,
      "abandonRatePct": 10,
      "asaMs": 8200,
      "serviceLevelPct": 84,
      "ahtMs": 254000
    }
  ]
}

Roadmap

Configurable service-level target per queue (today slSec defaults to 20s at the report layer).
Wrap-up in AHT: folding post-call work into handle time.
Interval reporting: half-hourly buckets for the classic SL/ASA-over-time grid.
Full call CDR + the edge SQL database/Iceberg export for long-term retention and cross-call analysis.