Average speed of answer (ASA) is the average time answered calls waited in queue before an agent picked up. To calculate it, add up the queue time of every call answered in an interval and divide by the number of calls answered. Callers who hung up before anyone answered are left out, and on most platforms so is the time spent in the phone menu before the call reached the queue.
The same figure can be read another way. ASA is the share of callers who waited at all, multiplied by the average wait of the ones who did. On a queue taking 180 calls a half hour, with 5 minute calls and 35 agents on the phones, ASA comes out at 17 seconds. Behind that number, 71.5 percent of callers got straight through, the other 28.5 percent waited 60 seconds on average, and 3.9 percent were still waiting after two minutes. The tables below show that breakdown at other staffing levels and queue sizes, along with what each platform puts into its ASA and the spreadsheet formulas to work it out yourself.
What is ASA in a call center?
ASA is the standard measure of how long the callers who reached a person had to wait for one. Most automatic call distributors report it for each interval, usually a half hour, next to calls offered, calls answered and the abandonment rate. It covers the wait in the queue only, so it says nothing about how long the call itself took.
In the half-hour report Gans, Koole and Mandelbaum walk through in Telephone Call Centers: Tutorial, Review, and Research Prospects (Manufacturing and Service Operations Management, 2003), ASA is the time answered calls spent on hold waiting to speak to an agent. Because it leaves out the time abandoned callers spent waiting, they write, "a reasonably full picture of congestion requires, at a minimum, both ASA and Abn % statistics," Abn % being the abandonment rate. Our call center metrics guide works through a half hour where a good-looking ASA and heavy abandonment come from the same queue.
ASA is also a different number from average handle time, which runs from the moment an agent answers to the end of wrap-up. A center can have a 10 second ASA and 8 minute calls, or 8 minute waits and 3 minute calls.
How do you calculate average speed of answer?
Divide the total queue time of answered calls by the number of calls answered in the same interval. If 400 calls were answered in a half hour and together they spent 6,800 seconds in queue, ASA is 6,800 ÷ 400 = 17 seconds. Abandoned calls stay out of both the top and the bottom of the fraction.
In a spreadsheet with one row per call, the outcome in column C and seconds in queue in column D, the interval's ASA is =AVERAGEIFS(D2:D2000, C2:C2000, "Answered"). With one row per half hour instead, calls answered in column B and that interval's ASA in column C, the day's ASA is =SUMPRODUCT(B2:B49, C2:C49)/SUM(B2:B49).
Weight by calls answered whenever you combine intervals. Say a 9:00 half hour answered 400 calls at a 17 second ASA and a 2:00 half hour answered 100 at 90 seconds. Weighted, the pair comes to 31.6 seconds. A plain =AVERAGE(C2:C3) says 53.5, because it gives the quiet half hour the same weight as the busy one. Neither figure shows that the 2:00 callers waited more than five times as long as the 9:00 callers, so report ASA per interval and keep the daily number for trends.
What each platform counts as speed of answer
Platforms agree on the basic fraction and differ at the edges: whether ring time and the agent's whisper count as waiting, whether a call moved between queues brings its earlier wait along, and which calls go in the denominator. Read the definition behind the number before you compare two centers or two vendors' reports.
| Platform or source | What the wait includes | Calls it divides by |
|---|---|---|
| Amazon Connect, Average queue answer time | Time in queue until an agent answers, including the agent and customer whisper | Contacts connected to an agent |
| Genesys Cloud, Average Speed of Answer | Wait time of answered conversations (tAnswered) | Answered conversations |
| Cisco Unified CCE, skill group ASA | Queue time, ring time and network time | Calls answered |
| Five9 | Hold time and queue time, without IVR or call routing time | Not stated in its FAQ |
| HDI service desk benchmark | Time callers spent in queue | Calls handled, including calls the IVR handled |
Putting calls the IVR resolved in the denominator, as the HDI definition does, moves the number most: it adds calls with no queue wait and pulls ASA down. Requeued calls can push it the other way. Cisco's documentation gives the example of a call queued at one skill group, moved to a second after 30 seconds and answered 10 seconds later: it reports 40 seconds of queue time against the skill group that answered. Amazon Connect also publishes separate answer-time metrics for callback contacts, so check which one a dashboard is showing before you read it.
What does an ASA hide?
An ASA hides how unevenly the wait was shared. Most callers wait less than the average, many of them not at all, and a few wait several times longer. Gans, Koole and Mandelbaum write ASA as the probability of waiting times the expected wait of those who wait, and under the Erlang C model the waits of callers who queue follow an exponential distribution, so a long tail comes with every ASA.
The table takes one queue at six staffing levels: 180 calls in a half hour at a 5 minute average handle time, which is 30 Erlangs of offered load. The figures come from Erlang C, the model behind our service level guide, which uses the same queue.
| Agents on the phones | ASA | Callers who waited | Average wait of those who waited | Waited over 1 minute | Waited over 2 minutes |
|---|---|---|---|---|---|
| 33 | 49 seconds | 49.0 percent | 100 seconds | 26.9 percent | 14.8 percent |
| 34 | 28 seconds | 37.6 percent | 75 seconds | 16.9 percent | 7.6 percent |
| 35 | 17 seconds | 28.5 percent | 60 seconds | 10.5 percent | 3.9 percent |
| 36 | 11 seconds | 21.2 percent | 50 seconds | 6.4 percent | 1.9 percent |
| 37 | 7 seconds | 15.5 percent | 43 seconds | 3.8 percent | 0.9 percent |
| 38 | 4 seconds | 11.2 percent | 38 seconds | 2.3 percent | 0.5 percent |
At 35 agents, a 17 second ASA sits beside one caller in ten who held for more than a minute, more than three times the ASA. At 33 agents the report shows 49 seconds while about one caller in seven waits more than two minutes.
Brad Cleveland, a founding partner of ICMI, makes the same point: "ASA is not a bell-shaped curve. Longest waits are far beyond what ASA, at face value, suggests" (ICMI, 2022). He recommends service level over ASA and using ASA only as a supplement to it. Whichever you lead with, put two numbers next to ASA: the share of callers who waited at all, and the share who waited longer than a limit you pick, such as two minutes. Amazon Connect's Maximum queued time is a quick check as well, and it includes contacts that abandoned.
Erlang C assumes nobody hangs up. Real queues lose some of their longest waiters to abandonment instead of answering them late, which makes the tail in your own report look shorter than callers experienced it. Read abandonment rate beside it for that reason too.
What is a good average speed of answer?
There is no industry standard ASA. A sensible target comes from the service level you staff to and the size of the queue, because the same 80/20 target produces a different ASA, and a different spread of waits, on a queue of 4 agents than on one of 108. Set ASA after service level, from your own queue's numbers, rather than borrowing another center's.
Each row below is a half hour with a 5 minute handle time, staffed with the fewest agents that answer 80 percent of calls within 20 seconds under Erlang C. The agent counts match the table in our occupancy guide.
| Calls in the half hour | Agents for 80/20 | Service level reached | ASA | Callers who waited | Average wait of those who waited | Waited over 2 minutes |
|---|---|---|---|---|---|---|
| 12 | 4 | 84.8 percent | 26.1 seconds | 17.4 percent | 150 seconds | 7.8 percent |
| 30 | 8 | 86.3 percent | 16.7 seconds | 16.7 percent | 100 seconds | 5.0 percent |
| 60 | 14 | 86.7 percent | 13.1 seconds | 17.4 percent | 75 seconds | 3.5 percent |
| 120 | 25 | 85.0 percent | 12.5 seconds | 20.9 percent | 60 seconds | 2.8 percent |
| 300 | 57 | 84.5 percent | 10.6 seconds | 24.6 percent | 43 seconds | 1.5 percent |
| 600 | 108 | 80.7 percent | 12.3 seconds | 32.8 percent | 38 seconds | 1.3 percent |
The 12-call queue meets 80/20 with the highest ASA, and the callers who queue there wait two and a half minutes on average. On the 600-call queue a third of callers wait, for 38 seconds on average. The difference is spare capacity: the wait of a caller who queues depends on agents minus offered load, which is two on the smallest queue and eight on the largest. An ASA ceiling of 20 seconds sounds looser than 80/20, yet it would fail the smallest queue at the staffing that meets 80/20.
The 12 call centers of a large United States health insurer in the Gans, Koole and Mandelbaum paper ran a daily average ASA of 31 seconds, with 318 second handle time, 91 percent agent utilization and 2.8 percent of callers abandoning. A large catalog retailer in the same paper had an ASA of about 1 second in its 10:00 to 11:00am peak at about 65 percent utilization. In efficiency-driven operations such as many help desks, the authors write, "essentially all customers are delayed in queue, ASA is on the order of an expected service time."
IT service desk benchmarks run longer. In benchmark data Jeff Rumburg published through HDI in 2020, service desk ASA averaged 98 seconds worldwide and ranged from 10 to 526. He suggests 25 percent of average handle time as a rule-of-thumb target, and his data showed no customer satisfaction gain from pushing ASA below about 30 percent of handle time.
ASA vs service level, wait time and hold time
These metrics get confused because they all describe waiting, yet each counts a different wait or a different set of callers. ASA averages the queue wait of answered calls. Service level counts the share answered within a threshold. Hold time is time on hold after an agent answered, and it counts toward handle time.
| Metric | What it measures | Callers who hung up in queue |
|---|---|---|
| Average speed of answer | Average queue wait of answered calls | Left out |
| Service level | Share of calls answered within a threshold, such as 20 seconds | Depends on the formula; see four common versions |
| Average queue abandon time (Amazon Connect) | Average wait of callers who hung up in queue | The only callers counted |
| Maximum queued time (Amazon Connect) | The longest wait in queue | Included |
| Average hold time | Time on hold after an agent answered, part of handle time | Not counted, since they never reached an agent |
| Average handle time | Talk, hold and after-call work | Not counted |
"Average wait time" has no fixed definition. Call Centre Helper describes it as "the period of time before being connected to an advisor," the same span ASA measures, but a report can label any queue average that way. Ask whether yours includes callers who hung up before you compare it with ASA.
How do you reduce average speed of answer?
Put more agent time on the phones in the intervals where waits build, or take calls out of those intervals. In the first table, adding a 35th agent cuts ASA from 28 to 17 seconds, so the fix is usually a few people in the right half hours rather than a bigger roster all day.
- Find the intervals. Pull ASA per half hour for four weeks and mark the ones over target; the daily figure won't show them.
- Rebuild the forecast and schedule for those intervals. The workforce management guide covers staffing to arrival patterns.
- Close the gap between agents scheduled and agents on the phones. Schedule adherence and shrinkage are where that time goes.
- Shorten handle time where calls run long for a fixable reason, since shorter calls lower the offered load each interval has to clear.
- Set an overflow rule that sends calls to a second trained queue, or offers a callback, once the wait passes a limit you choose. Check whether your platform counts callback waits in ASA or in a metric of their own.
- Take whole call types out of the queue with self-service or an AI agent, for the calls that don't need a person. The call deflection guide covers which of those moves actually remove agent minutes.
Where an AI agent shows up in your ASA report
Telvana answers immediately at any hour, so the calls it takes never wait in the queue. Offered as a phone-menu option when the wait climbs, or set up as a forwarding rule for overflow, it moves calls out of the queue in the half hours that run hottest, and callers who would rather wait for a person keep their place in line (hold queue relief).
Report the AI agent's answered calls in a row of their own instead of folding them into your team's ASA; mixed in, a stream of near-zero answers lowers the average without any caller reaching a person sooner. And when the agent transfers a call, the call goes to the number you choose. If that number feeds a queue, the caller waits there like any other call, and that wait belongs in your team's ASA.
Calls outside business hours are a separate gap. On our platform, 37.6 percent of calls to business lines arrived outside Monday to Friday, 8am to 6pm, in the caller's local time (call center statistics). A center that sends those calls to voicemail never sees them in its ASA. Our team builds each agent from your scripts and transfer rules and tests it with you before it takes a live call. Pricing starts at $399 a month based on call volume, plus a one-time setup fee (pricing).
Frequently asked questions
Does average speed of answer include abandoned calls?
No. ASA averages the waits of answered calls only, so callers who hung up in queue drop out of both the total wait and the call count. That is why a queue with rising abandonment can report a falling ASA. Read it with abandonment rate, and with average abandon time if your platform reports one.
Does ASA include time in the IVR?
Usually not. The clock starts when the call enters the queue, after the menu. Five9 says its ASA leaves out IVR and call routing time, and Call Centre Helper notes that ASA is measured "starting from when the customer is placed in a queue after they have navigated IVR." A long menu adds to the caller's wait without showing up in ASA.
Is average speed of answer the same as average handle time?
No. ASA is the wait before an agent picks up. Average handle time is the call itself: talk, hold and after-call work. Both feed staffing, where handle time sets the workload and ASA is one result of how well that workload was staffed.
What is the difference between ASA and service level?
ASA is an average wait in seconds. Service level is the share of calls answered within a threshold, such as 80 percent within 20 seconds. Both come from the same waits and usually move together, but service level counts how many callers were answered promptly, while ASA says nothing about how the wait was spread.
Can you staff to an ASA target?
Yes. Gans, Koole and Mandelbaum describe centers that set an upper bound on ASA and staff the fewest agents that keep it there. The second table shows the catch: the same ceiling allows a different share of long waits on each queue size, so pair the target with a limit on the share of callers waiting more than two minutes. Our staffing calculator shows ASA next to service level for your busiest hour.
Is there an ASA for chat?
HDI's version for chat is average response time: the wait from when a customer starts a chat until an agent first replies, calculated the same way as ASA, with abandoned chats left out.
Find the ASA your queue can hold
Put your busiest hour into the staffing calculator to see the ASA, service level and occupancy each agent count gives you, then see how they change when AI agents take part of the calls. To hear Telvana take a call from your queue, book a demo and bring a week of interval reports.