Last updated: September 12, 2026
What average speed of answer measures
Average speed of answer is total queue time on answered calls divided by the number of calls answered in the same window. One survey of 225 UK operations printed a mean of 116 seconds against a median of 48 from the same respondents, which is what an average does to a queue with a long tail.
Both figures come out of ContactBabel's 2024 survey, same question, same respondents, same year. A mean sitting at more than double the median means the distribution has a tail, so most callers waited well under a minute and a smaller group waited long enough to carry the average up by more than a minute on their own.
That tail is the part worth your attention, and an average is the one reading that cannot show it to you. Callers who sat four minutes arrive in your report as a few seconds added to a figure that still looks fine. Both readings sit on the benchmark page with the population and the year attached, next to the abandon rate the same respondents reported.
The formula, and what goes on each side
Add up the queue time for every call an operator answered, then divide by the count of those calls. Say a month produced 6,400 answered calls and 179,200 seconds of queue time across them. The month reports 28 seconds.
Both sides of the fraction describe the same population, which is where most attempts to improve the number go wrong. Adding abandoned calls to the denominator without adding their wait to the numerator produces a smaller figure that describes nothing at all.
Where the clock starts
Two calls both answered in 30 seconds of queue can be half a minute apart in what the caller lived through.
Four moments are in use as the start, and your platform picked one of them for you. A clock can open when the carrier delivers the call, when the recorded greeting prompt ends, when a caller picks an option and enters a queue, or when an operator's handset rings after the platform offers them the call. Queue time and ring time sit in separate columns on some reports and get rolled into one on others, while the caller counts from the first ring regardless.
Pick one, write it down beside the number, and hold it through a platform change, because an average that improved by nine seconds during a migration usually improved by a definition. Rebuilding a year of history after a switch like that sucks, and it is the only way the line means anything across it.
Abandoned calls, and the number they quietly improve
The standard formula counts answered calls only, on the plain logic that a caller who hung up was never answered. That logic has a side effect worth naming, since the caller who waited longest and gave up leaves the numerator and the denominator at the same moment.
Take the same 6,400 answered calls at 28 seconds, and say 420 more callers abandoned after an average of 2 minutes 15 in queue. Put their waits back in and the figure across everybody who dialed you runs about 35 seconds. Neither number is wrong, and only one of them counts the people who never got through.
Our first instinct was to publish the honest version, the one counting everybody who waited. Then you get to explain a metric nobody else calculates to a client who wants the number in their agreement, every quarter, forever. Keep the standard formula for the report and put abandon rate on the same page, so the pair reads together.
Against a service level, the two answer different questions
Your service level is a headcount question, since it asks how many callers cleared a threshold. Speed of answer is a stopwatch question asked of the callers who got picked up at all.
The percentage says how many callers you missed and nothing at all about how badly. Twenty percent of calls outside a 20 second threshold could be a floor where those callers waited 24 seconds, or one where they waited eleven minutes, and the percentage reads the same either way. Carrying the size of the miss is the one job the average does better. Where that threshold sits in a client agreement, the contractual side of service level is a separate problem with its own definitions.
Percentiles, because a month is not one number
Sort the month's answered calls by queue time and read the value nine tenths of the way up the list. The 90th percentile is the wait that 90 of every 100 callers came in under, and it climbs as soon as a single interval goes bad, which is what you wanted from the average in the first place.
That same hypothetical month of 6,400 answered calls, read five ways.
| Reading | Value | What it describes |
|---|---|---|
| Median | 14 seconds | The middle caller of the month |
| Mean, the figure you report | 28 seconds | Every answered call, tail included |
| 80th percentile | 34 seconds | Where the slow fifth of your callers begins |
| 90th percentile | 1 minute 35 | The worst tenth, and the ones who mention it |
| 95th percentile | 3 minutes 10 | The calls your client hears about directly |
Nothing in that column is false. The 28 seconds goes in the report and the 3 minutes 10 is the call that prompted an email to your account manager.
Split the month into half hour intervals and the tail usually turns out to have a schedule, with Monday openings running long and the overnight on-call window running longer. A month averaging 28 seconds can hold a Tuesday at 9:00 above four minutes, and the monthly figure will never show it to you.
Percentiles need volume, which is the limit on all of this. An account taking 90 calls a month has a 95th percentile built from four or five calls, so read percentiles at the queue level and keep the account conversation on the calls themselves.
What the number cannot tell you
Everything above stops at the second somebody said hello.
Answering in 20 seconds and then writing the callback number down wrong is worse for that caller than answering in 40 and getting the message right, and your report files the first one under good. Speed comes off the phone platform, quality comes out of somebody applying a written standard to the recording, and the QA metrics that describe your program are built on that split.
Two readings put them back together. Pull the fastest answered tenth of the month and score those calls against the account script, since a QA average on that tenth sitting below your floor average means the speed is getting bought with something. Then look at critical fails inside the intervals where queue time was worst, because an operator working a backed up board is the one who skips reading the callback number back digit by digit. A handle time target pushes on those same seconds from the other end.
A criterion like "sounded rushed" stays a supervisor's judgment, since automated scoring is reliable on the rule based items and weaker on tone, which Aircall and StackAI both note. What CallSniper adds to the rule based half is evidence: every flag against your scorecard carries a transcript timestamp and a confidence score, so the fast call that went wrong is findable at the second it did. Take the intervals where your queue time ran worst last month and put those recordings through a call review, which is where your two numbers either agree with each other or quietly stop.
Frequently asked questions
How do you calculate average speed of answer?
Divide total queue time on answered calls by the number of calls answered in the same window, counting answered calls on both sides. Decide where the clock starts, at queue entry or at the end of the greeting prompt or at the ring on a handset, and keep that definition written beside the number.
What is a good average speed of answer?
No industry standard survived our checking. ContactBabel's 2024 survey of 225 UK contact centre managers reported a mean of 116 seconds and a median of 48, and choosing between those two decides whether your floor looks quick or slow. The benchmark page carries the population and date for both.
Does average speed of answer include abandoned calls?
The standard formula leaves them out, since a caller who hung up was never answered. That means it improves when your slowest callers give up waiting, so publish abandon rate beside it.
Should we report average speed of answer or service level?
Report both, because they answer different questions about the same month. The percentage says how many callers waited past your threshold, and the average carries how long the wait ran. When the two move in opposite directions, a single bad interval is usually what moved them.
Does average speed of answer include IVR time?
That depends on where your platform starts the clock, and the menu seconds are real to the caller either way. A threshold that starts counting after the greeting prompt ends buys you seconds your caller still sat through.
Sources
- ContactBabel, UK Contact Centre Decision Makers Guide 2024, 21st edition, on speed to answer at a mean of 116 seconds and a median of 48. Survey of 225 UK contact centre managers and directors answered October and November 2023. Third party, read from the full report PDF on 2026-09-12.
- Aircall and StackAI on rule based criteria scoring more reliably than tone. Third party, unverified.
- Trinsic feature sheet and call notes, 2026-08-27 and 2026-09-03. Verified internally.
- The 6,400 call month, the percentile table, the four clock start positions, and the position that the definition belongs in writing beside the number. Editorial judgment as of 2026-09-12.