What is average speed of answer (ASA)?
Average speed of answer (ASA) is the mean time an inbound caller spends in queue before a live agent connects, measured from the moment the call enters the queue to the moment an agent picks up. It is reported in seconds and calculated across answered calls only.
A team posting an ASA of 45 seconds holds the average caller for 45 seconds, even when some calls connect instantly and others sit for five minutes. The average conceals that spread, which is why experienced operators read it beside a distribution of wait times.
How average speed of answer is calculated
The formula is a single division: total queue time on answered calls divided by the number of answered calls. If 1,000 answered calls waited a combined 30,000 seconds, ASA is 30 seconds.
Both inputs come from the telephony layer. The automatic call distributor timestamps the moment a call enters a queue and the moment an agent connects, and the interval between those two stamps is the only span ASA counts. Time spent in a menu before queue entry usually sits outside it.
The denominator does the quiet damage. Abandoned calls normally drop out of it, so the callers who waited longest and gave up leave no mark on the average. If 200 callers abandon after 120 seconds each while 800 connect after a combined 32,000 seconds, ASA reads 40 seconds; counted across all 1,000 callers who dialled, the average wait was 56 seconds. Reading the metric beside call abandon rate closes that gap.
Upstream, call routing decides which queue a call joins, and a misrouted call that re-queues starts a second clock the report will count separately.
What counts and what does not
ASA is narrower than most dashboards imply. Four boundaries decide the number.
Queue time only: The clock starts at queue entry and stops at connection, so talk, hold, and wrap time belong to average handling time.
Answered calls only: Abandoned calls leave the denominator in most configurations, which lets the average fall while service is degrading.
Menu time excluded: Self-service navigation before queue entry is usually outside the measurement, so a long menu can hide a long total wait.
Callbacks and transfers: A queued callback often stops the clock at the offer, and a transferred call may open a fresh queue interval with its own timestamps.
Average speed of answer vs average handle time vs service level vs call abandon rate
Contact center dashboards show these four side by side, and teams routinely move the target on one while the damage lands on another. Average handle time counts the minutes an agent spends on a contact after the connection is made. Service level counts the share of calls answered inside a threshold the operator chooses. Call abandon rate counts the callers who left the queue before any agent reached them. Average speed of answer counts one thing: the mean wait of the callers who did get through, which makes it a staffing metric before it is a service metric.
What it counts | What it misses | Typical benchmark | |
|---|---|---|---|
Average speed of answer | Mean queue wait across answered calls | Abandoned callers, talk time, whether the issue was solved | No published cross-industry figure; targets are set per queue |
Average handle time | Talk, hold, and after-call work per contact | Everything that happened before the connection | Varies by contact type; read against your own baseline |
Service level | Share of calls answered inside a stated threshold | How long the callers outside the threshold waited | Threshold is chosen by the operator and stated in seconds |
Call abandon rate | Callers who hang up while still in queue | Why they left, and whether they dialled again later | Compared against the center's own seasonal history |
If you need one number to feed a staffing forecast, ASA is the input the Erlang models expect. If you need to know whether the worst-served callers are being sacrificed to protect a good average, read service level and abandon rate together before touching the schedule.
Why average speed of answer matters for customer experience
Without a measured queue wait, nobody owns the hold. Staffing arguments run on anecdote, the worst hour of the week never surfaces in a report, and the cost appears later as abandoned calls, repeat dials, and callers who open the conversation already irritated. ASA gives the wait an owner and a number a schedule can be built against.
The failure mode is subtler once the number exists and is trusted alone. A center can hit an aggressive ASA by answering fast and resolving slowly, pushing the delay past the greeting into transfers, holds, and promised callbacks. Total time to resolution grows while the headline improves.
The tradeoff is worth stating plainly: every second removed from the queue is bought with capacity, and capacity spent on speed is capacity unavailable for depth. Teams that force the average down without touching demand end up with fast greetings and thin answers.
How is average speed of answer benchmarked?
No standards body publishes an ASA target, and the seconds quoted in vendor collateral each describe one installed base. The defensible external anchor sits downstream of the queue: the American Customer Satisfaction Index scores industries on a 100-point scale, and sector results sit broadly between the low 70s and the low 80s, a spread that no single wait-time target explains.
Build the target from your own distribution. Pull queue waits for a full month, split them by queue and by daypart, and read the median beside the ninetieth-percentile wait: the median describes the typical caller, and the tail shows who is being sacrificed to hold the average down.
Then set the target per queue. A billing line with predictable demand and an outage line that spikes without warning cannot share a number, and averaging them produces a figure that describes no real caller.
How AI agents change average speed of answer
An AI voice agent answers on arrival, so for every contact it handles end to end, queue time collapses toward zero and the interval ASA measures barely exists. An AI IVR front end does a weaker version of the same thing, holding the caller in conversation while intent and identity are captured, which moves work off the queue clock entirely.
The consequence lands on the calls that remain. Deflection removes volume, and a lighter queue drains faster, so human-queue ASA usually falls first. The residual mix is heavier: the calls left over run longer, and a schedule trimmed in proportion to call count will rebuild the wait that deflection just removed. Staffing has to follow workload minutes, because contact count no longer describes the load.
Teams weighing the two paths generally model them together, and the economics are laid out in this comparison of AI voice agents and staffing.
How to reduce average speed of answer
Four levers move the number, and they cost very different amounts.
Forecast and schedule accuracy is the cheapest, since most queue spikes are predictable by daypart and adherence gaps of a few minutes per agent aggregate into whole positions at peak. Routing precision is second, because every misrouted call re-enters a queue and is counted twice. Demand reduction is third: automating the repeat reasons that fill the first tier shortens the queue itself, the mechanism behind these ways AI agents cut response time. Headcount is last and least efficient, because Erlang C staffing curves flatten as wait falls, so the final seconds of average wait cost far more agents than the first ones did.
Governance decides whether a gain holds: name one owner per queue who may change the target, and log every change with its date. Where automation answers before a human does and touches account records, regulated buyers ask how identity checks are evidenced, which is where SOC 2 Type II reports get requested.
Average speed of answer and queue management
ASA is the scoreboard for queue management, the practice it grades: how contacts are ordered, which ones jump, and what happens when a queue exceeds its planned depth. Priority rules that push a VIP segment forward buy those seconds from someone further back, and the average absorbs the transfer silently.
Inside a call center the same seconds are argued over by three functions: workforce management owns the forecast, operations owns adherence, and support leadership owns the target. ASA is where those three decisions meet.
What does average speed of answer mean in plain terms?
ASA stands for average speed of answer, and the full form is sometimes written as average speed to answer. Think of it as the counter at a pharmacy: you take a ticket, you wait, and ASA is the average time between taking the ticket and hearing your number called. It says nothing about whether the prescription was correct.
Picture a queue where nine callers connect in five seconds and one waits eight minutes. Total wait is 525 seconds across ten callers, so the average lands near 53 seconds, a figure that describes none of the ten people who dialled.
The tradeoff in plain terms: you can always answer faster by answering worse. A greeting inside ten seconds followed by a hold, a transfer, and a promised callback is a fast ASA and a slow morning for the customer, and only one of those two facts reaches the dashboard.
Common average speed of answer mistakes
Reading the mean without the tail is the first pattern. Averages compress, so a small share of very long waits disappears into a healthy-looking number, and the callers who churn are the ones the metric hid.
Celebrating a fall that abandonment caused is the second. Because abandoned calls leave the denominator, a queue that gets worse can post an improving ASA as impatient callers drop out and stop contributing long waits to the calculation.
Publishing one blended target across unlike queues is the third. Sales, billing, and outage lines have different arrival patterns and different tolerances, and a single company-wide figure lets a fast queue subsidise a failing one for months.
Pitting the metric against handle time is the fourth. When both are enforced as agent-level targets, agents close contacts early to free themselves, and the unresolved issues return as repeat calls that refill the queue they were meant to shorten.
What is a good average speed of answer?
A good average speed of answer is defined by the queue, not by a published industry figure. Emergency and outage lines are held to seconds; billing and account queues tolerate longer waits when callers are told what to expect. Set the target from your own wait distribution, per queue and per daypart, then review it seasonally.
How do you calculate average speed of answer?
Average speed of answer is calculated by dividing total queue time on answered calls by the number of answered calls in the same period. If 1,000 answered calls waited a combined 30,000 seconds, ASA is 30 seconds. The automatic call distributor supplies both figures from its queue-entry and connection timestamps.
What is the difference between average speed of answer and service level?
Average speed of answer reports one mean across every answered call, while service level reports the percentage answered inside a threshold the operator sets. The average can look healthy when a minority of callers wait a very long time; service level exposes that minority directly. Most contact centers publish both and staff to whichever is tighter.
Average speed of answer vs average handle time: which should you target?
Average speed of answer measures waiting before the connection; average handle time measures work after it. They pull in opposite directions, because agents who rush contacts to shorten handle time generate repeat calls that lengthen the queue. Target ASA at the queue level and handle time at the process level, and never enforce both on individual agents.
Does average speed of answer include abandoned calls?
Average speed of answer normally excludes abandoned calls, because most platforms compute it across answered calls only. That exclusion means the longest waits, the ones that ended in a hang-up, never reach the calculation. A worsening queue can therefore show a falling average, which is why abandon rate belongs on the same report.
Can AI voice agents lower average speed of answer?
AI voice agents lower average speed of answer by answering on arrival, so the contacts they resolve never enter a human queue at all. The remaining queue is shorter and drains faster. The caution is workload: the calls that still reach agents tend to be longer and more complex, so schedules must follow minutes of work rather than call counts.