Queue time: Definition, formula, and how to measure it correctly
Green queue time and average hold time tiles can conceal the longest waits your callers experience. Consider a quarterly dashboard that shows average speed of answer (ASA) inside target and hold time under a minute as abandonment rises because the reporting layer excludes several forms of caller waiting. Reconcile the dashboard by reporting abandoned-call waits, callback delay, and Interactive Voice Response (IVR) time beside ASA.
The CFO can then see why the headline tiles diverge from caller experience. The missing waits include callers who left the queue before a human agent answered or accepted a callback and waited hours. Menu navigation also occurs before the queue clock starts.
What queue time is and how to calculate it
Queue time is the interval between a caller entering the queue and a human agent answering. Under the most common convention, calculate average queue time by dividing the total queue time for answered calls by the number of answered calls. For example, 360,000 seconds of queue time across 6,000 answered calls gives an average queue time of 60 seconds.
Vendors may label queue time as wait time or average waiting time; the terms describe the same pre-answer interval. Queue time never includes hold time, because a caller on hold has already reached a human agent who chose to park the call. Document the formula, the clock boundaries, and the included calls beside every queue-time figure so the number stays comparable across sites, vendors, and years.
How queue time relates to hold time, AHT, and ACW
Queue time is one clock in a family of contact center metrics that measure caller waiting and human agent effort at different points in the call. Each metric answers a different operational question, and confusing them produces reports that look consistent but describe different behavior. The table below sets the four most common metrics side by side.
Metric | What it measures | When the clock runs | Typical formula |
Queue time | Wait before a human agent answers | From queue entry to answer | Total queue time ÷ answered calls |
Average hold time | Time a human agent parks an already-answered caller | During the call, after answer | Total hold time ÷ answered calls |
Total post-answer effort per call | From answer through after-call work | (Talk time + hold time + ACW) ÷ handled calls | |
Work a human agent completes after the caller hangs up | From disconnect to case closure | Total ACW ÷ handled calls |
Average hold time and average handle time live inside the post-answer window, while queue time lives entirely before it. Reducing hold time lowers AHT even when talk time is unchanged, and reducing queue time doesn't affect AHT at all. Report each metric with its own denominator and clock boundaries so the four numbers describe four distinct parts of the call.
Why the queue time formula depends on your denominator
Two contact centers with identical staffing and identical call arrivals can report different queue times because the reporting layer decides what the clock counts. Some platforms divide queue time by total calls handled and others by answered calls, and the two are not the same set: when handled includes calls the IVR systems closed or calls that took a callback offer, the same illustrative total produces a smaller average.
A large operation running several platforms across regions can end up comparing a queue time in one region against a lower figure in another that describes identical behavior. Pick one denominator, document it in the metric definition, and apply it to every site so the trend stays comparable.
The four boundary decisions that shape queue time
Four boundary decisions do most of the damage to queue-time comparability. Each decision moves seconds into or out of the numerator, moves calls into or out of the denominator, or does both. Two vendors can quote the same headline figure while making opposite choices on all four, so the numbers describe different populations. Before adopting any queue-time target, document each of the four decisions and apply the same resolution across every site and platform.
1. Start point of the queue clock
The clock can start when the caller enters the queue or when routing completes and assigns a skill group. Skill-based routing adds seconds that one convention counts and the other does not.
On a long call flow, the gap between the two start points can reach several seconds per call, which compounds across a shift. Fix a single start point in the metric definition and apply it consistently, or year-over-year queue-time trends will move for reasons unrelated to caller experience.
2. Treatment of abandoned calls
A caller who waits several minutes and hangs up contributes that waiting time, which never enters a formula divided by answered calls. Including abandoned calls adds their waiting time to the numerator and adds those calls to the denominator, which usually raises the reported average.
The answered-only convention describes the experience of callers who reached a human agent; the inclusive convention describes the experience of everyone who queued. Choose the convention that matches the question the report is meant to answer.
3. Treatment of callbacks
A caller who accepts a callback typically leaves the queue as a handled or deflected call at that moment. The time until the callback arrives is real waiting and usually doesn't show up in the no queue-time report at all.
Callers who accepted a callback instead of holding waited an average of one hour and 51 minutes for the agency to call them back, a wait the SSA callback wait audit puts outside the ASA figure entirely. Report callback delay as its own metric so it stays visible.
4. Treatment of IVR time
ASA typically excludes IVR navigation, so total caller wait exceeds reported wait by the length of the menu path. A long menu in front of a short queue reports only the short queue, and callers who spend two minutes navigating options before joining a 30-second queue experience two and a half minutes of waiting.
Report total caller wait, from connection to a human agent answering, as a separate metric so the full experienced wait stays visible on the same dashboard as queue time.
What good queue time looks like in practice
Well-run operations treat queue time as a capacity signal, not a coaching target, and they publish denominator-qualified figures that hold up across sites and years. The strongest evidence comes from named customer deployments where a documented change in staffing, routing, or automation moved queue time in a way executives could verify. The three examples below show queue-time improvements alongside the operational change that produced them.
Württembergische Versicherung reduced call wait times by 33% within four weeks, with customers rating its AI agent 3.8/5 for customer satisfaction score (CSAT) and the deployment went live in four months.
BER Airport provides zero wait times with 24/7 availability in four languages and went live in six weeks.
orderbird reduced customer wait time by 60%, from 98 to 39 seconds, by putting an AI agent in front of the human queue.
Each of these outcomes ties a queue-time figure to a specific denominator and a specific operational change. That is the pattern to imitate: publish the number, publish the denominator, and publish the change that moved it. A queue-time figure without that context tells a second team nothing it can reproduce.
How to monitor the cost of wait times
Callers abandon based as much on how a wait feels as on how long it runs, and every abandoned call carries a cost the queue-time report rarely shows. A caller on silent hold sees no evidence that the queue is moving: no position announcement, no estimated wait, and no visible progress. Silence and a lack of queue feedback can make a wait feel longer, so teams should monitor abandonment alongside queue communication.
When queue time and call abandonment rate rise in the same half-hour intervals, compare arrivals with staffed capacity first: more callers may have arrived than the human agents on shift could answer, and the callers who left will dial again and arrive as new volume. That repeat volume inflates arrival counts on the next interval and pushes queue time higher again, so the cost of a single understaffed interval propagates across the shift.
How to measure AI agent queue time correctly
Many legacy reporting stacks support a single human queue and break quietly when an AI agent answers first. An AI agent answers callers concurrently rather than queueing them, so a surge does not form a line. Measuring the change requires segmenting AI-handled from human-handled interactions, reporting AI response latency as its own clock, and preserving the human-queue denominator so year-over-year trends stay intact.
Segment AI-handled from human-handled interactions. For every interaction an AI agent fully resolves without human escalation, queue time is zero, and no hold event occurs. Averaging AI-handled and human-handled interactions into one queue-time figure buries a removed queue inside a blended number that looks like a modest decline.
Report AI response latency separately from hold. AI response latency is the interval between a caller finishing a sentence and the AI agent beginning its reply; it is far shorter than any hold and belongs in its own column. Latency is a property of the AI agent's response, while hold is a human agent's decision, and combining them makes both unreadable.
Track escalation queue time as a distinct wait. A caller who reached an AI agent and escalates to a human agent enters a second queue. Report that interval as a distinct wait and never fold it into the AI segment's zero.
Keep the human-queue denominator unchanged year over year: If a team is divided by answered calls one year, it must be divided by answered calls the next year, so the human-queue trend survives the AI agent's arrival intact.
A dashboard that follows these four rules gives the board separate views of the removed AI queue and the human-queue trend. Both views are needed: the AI segment shows what concurrency removed, and the human segment shows what staffing and routing still owe the remaining callers.
Rebuild how you report queue time
Queue time is a capacity metric with a stated denominator, not a coaching target that stays green while customers leave. Treating it as the former makes staffing decisions defensible and year-over-year trends reproducible; treating it as the latter lets teams meet a target by changing the report instead of the operation. Every second removed from the queue gives another customer a reason to stay on the line instead of hanging up and calling back.
Parloa manages AI agents across Build, Optimize, and Observe as one lifecycle, so AI agents answer first and verify identity inline, with compliance coverage including ISO 27001:2022, ISO 17442:2020, SOC 2 Type 1 & 2, PCI DSS, HIPAA, and DORA. Teams can segment the interactions AI agents produce and count them as zero for year-over-year comparison alongside the human-queue trend.
Book a demo to see queue time reach zero for calls your AI agents answer, and get a segmented reporting model your board can trust.
Get in touch with our teamFAQs about queue time and hold time
Does average handle time include hold time?
Yes. AHT sums talk time, hold time, and ACW for each handled call, so every minute a human agent parks a caller directly lengthens AHT.
In formula form, calculate AHT by adding total talk time, total hold time, and total ACW, then dividing the sum by the number of handled calls. A hold-time reduction therefore shows up in AHT even when talk time is unchanged.
Should queue-time calculations include abandoned calls?
Two conventions exist. One divides total queue time for answered calls by answered calls only, excluding the waiting time abandoned callers did; the other divides total queue time across answered and abandoned queued calls by answered plus abandoned queued calls, capturing their waiting and potentially producing a higher figure. Pick one, document it in the metric definition, and apply it to every site and platform so the trend stays comparable.
Does IVR time count as queue time?
Typically not. ASA and most queue-time reports start the clock when the caller enters the queue, after IVR navigation ends, so menu time drops out of the headline figure. Report total caller wait, from connection to a human agent answering, as a separate metric so the full experienced wait stays visible.
How do you measure a queue when AI agents answer part of the volume?
Segment the report. Compare the human-handled queue against its own prior years on the same denominator, and report AI-handled interactions as their own segment with zero queue and zero hold. A blended figure across both segments doesn't support a valid comparison, including against your own history.