When inbound demand exceeds the number of people available to answer, ringing every phone at once stops working. A queue holds the caller in a waiting state (music, periodic announcements, optional callback) and an ACD policy picks which waiting call goes to which free agent. That is the difference from a hunt group, which offers ring cycles and then fails or overflows.
What ACD actually distributes
Automatic Call Distribution is the policy layer on top of the queue. Classic policies:
| ACD policy | Next agent is | Typical use |
|---|---|---|
| Longest idle / most idle | The ready agent who has waited longest since last call | Fairness across a single skill |
| Round-robin | The next agent in a rotating list | Simple teams without occupancy data |
| Least recent / fewest calls | The agent with the fewest handled calls in the window | Balance volume, not occupancy |
| Skills-based | The ready agent whose skill profile best matches the queue or caller data | Multi-queue contact centers |
| Preferred / sticky | A last-agent or account-owner if ready, else fallback skill | Continuity for an open ticket |
Wait time and position
- Wait time (speed of answer for this caller)
- Elapsed time from queue entry until answer or abandon. Estimated wait time (EWT) is a forecast from recent average speed of answer and queue length. EWT is an estimate, not a promise.
- Position
- How many callers are ahead in that queue (sometimes per skill). Position announcements reduce uncertainty. They can increase abandon if the number is large and no callback is offered.
- ASA / average speed of answer
- Mean wait of answered calls in a period. Ignores abandons unless you also track abandon time.
- Abandon
- Caller hangs up while waiting. Short abandons (a few seconds) are often filtered so a misdial is not treated as a staffing failure.
Overflow
Overflow is what happens when the queue cannot meet its own limits. Limits are usually max wait, max queue length, no agents logged in, or after-hours. Overflow targets are another queue, a hunt group, voicemail, a callback request, an IVR retry, or forwarding to an answering service.
- Max wait: after N seconds, leave this queue even if agents exist (they may be on long calls).
- Max length: reject or divert when too many callers are already waiting (protects EWT for those already in line).
- No agents: do not play 20 minutes of music to an empty staff group. Divert immediately.
- Priority / VIP: a second queue or a weight that jumps position. This is still ACD, not a hunt.
Service level (SLA)
Contact-center SLA is usually stated as "X percent of calls answered in Y seconds" (for example 80/20: 80% answered within 20 seconds). It is a staffing and routing metric, not a carrier uptime SLA and not a MOS target.
| Metric | What it measures | What it hides |
|---|---|---|
| Service level (e.g. 80/20) | Share of offered calls answered before the Y-second threshold | If only the binary threshold is tracked, a call answered one second past Y counts the same as a 10-minute wait. |
| Average speed of answer | Mean wait of answered calls | A few very long waits can hide in a good mean, or a good mean can hide high abandon. |
| Abandon rate | Share of queued calls that hang up before answer | Does not say whether those callers redialed. |
| Occupancy | Share of logged-in time agents spend on calls vs idle | High occupancy plus a tight SLA means the queue will grow at any spike. |
Where the queue sits
- Inbound call to a DID or toll-free number, often after an IVR choice.
- Caller enters queue treatment: MOH, position / EWT prompts, optional DTMF to voicemail or callback.
- ACD assigns an agent. The agent's client (SIP phone or WebRTC app) is offered the call.
- Talk, then wrap-up (ACW: after-call work) before the agent is ready again. Wrap-up is part of capacity.
- Recording, transcripts, and disposition codes attach to the queue and agent, not to the DID alone.
Queues vs hunts vs forwarding
Use a queue when callers should wait for a skilled person. Use a hunt group when a small team should just ring. Use forwarding when the target is one person or one external number. Mixing them is normal: an IVR can feed a queue whose agents also have CFNR to a supervisor. The mistake is using only simultaneous ring and calling it a contact center.