In most orgs, the CDR endpoints do not expose a single queue-level “abandoned” metric as a dedicated field. You usually have to derive it from detailed call history and call-leg records.
Recommended approach:
* Poll GET /v1/cdr_stream with a rolling startTime/endTime window (max 2 hours). Those timestamps filter by record insertion time into the cloud, not strictly call start/end.
* Stitch records by a correlation or session identifier so you operate at the interaction level, not per-leg.
* Identify interactions that touched a Customer Assist queue (use queue or routing indicators in your payload).
* Determine whether a human agent actually connected. Be careful here because platform/queue processing legs can look “answered” even when no agent spoke to the caller.
* If the interaction touched a queue, no agent-connected leg exists, and the caller disconnects after a few seconds, classify it as abandoned.
Caveats:
* Callback requested, overflow/transfer, and short abandons can look like “no agent answered” unless you classify them separately.
* Records can be updated after first insertion, so dedupe and upsert by a stable key.
* Handle 429 rate limits using Retry-After and backoff.
Two questions that shape the exact logic:
1. How do you define “abandoned” (hang up before agent only, or include overflow/timeout)?
2. Do you need per-call detail, queue aggregates, or both?
