Last updated: September 12, 2026
The metrics answering services track
Answering services track answer time, abandon rate, average work time, message accuracy, script adherence per account, and critical fail count. The first three fall out of the phone platform on their own. Accuracy and adherence need somebody to read the call against the account instructions, which is why the metrics clients complain about are the least measured.
Two different machines produce those numbers. Your platform counts seconds and calls whether or not anybody is watching. The rest come from a person with the recording open beside the account script, which costs supervisor hours and covers the thing a client picks up the phone about.
Metrics your platform already reports
Four numbers arrive at no cost to you. Read them for staffing, and be careful about what you promise on them.
Answer time
Average seconds between a call landing in the queue and an operator picking it up: total wait seconds on answered calls, divided by answered calls. Some boards report a service level instead, the share answered inside a threshold, which is the version most client agreements use.
The standard lives in the service agreement rather than in an industry rule. One account signs for 90 percent inside 20 seconds and another never raises it, so a floor target measures you against a promise you never made.
Weight it too heavily and the board picks up fast and writes down the wrong thing.
Abandon rate
Callers who hung up before an operator answered, divided by total inbound calls.
We almost left this one off the list, since it describes your schedule rather than your quality. It stays because clients read it, and because a rising abandon rate shows up well before anybody files a complaint.
Average work time
Talk time plus whatever the operator does after the caller hangs up, divided by calls handled. Dispatching to the on-call tech and typing the message both land inside it.
Push it down and message accuracy tends to follow, because the seconds being cut are the ones spent reading a callback number back digit by digit.
Calls per operator hour
Calls handled divided by operator hours staffed, which answers a scheduling question and nothing else.
QA has nothing useful to say here, since an operator at 40 calls an hour is either excellent or pushing callers off the line.
Metrics somebody has to read a call to produce
Each of these needs somebody to review the call against the account's own instructions, which is why most services measure one of the four and skip the rest.
Message accuracy
The share of reviewed messages with no field error in them: messages scored clean, divided by messages scored.
Producing it needs the recording sitting beside the message, because a message on its own always looks complete. The long version, including the three ways one goes wrong, is in message accuracy in detail.
Script adherence per account
Script points earned divided by script points available, scored against the account the call came in under. Greeting wording and the required triage question both sit here.
A floor average for adherence tells you almost nothing, since every account has its own script and an operator works a different one every 45 seconds. A 94 across 140 accounts says most of them are fine and nothing about which ones are not.
Critical fail count
A count of calls carrying at least one auto fail, per account, per period. The wrong client's business name in the greeting is one, and an emergency that never reached the on-call schedule is another.
Count these rather than rating them. Two auto fails out of 400 calls is a 0.4 percent rate that vanishes into a monthly average, and two conversations you need to have today. A scorecard built for 45 second calls carries six of them.
Client complaint volume
Complaints logged against an account over a period, which is the metric most services already run on. Calling it your quality number is fair up to a point.
A client only complains when the failure was expensive enough to be worth the phone call, so what you are counting is the tail of your error rate.
The whole set on one page
The column worth reading twice is the third one, since it says what has to happen before the number exists.
| Metric | Formula | Where the number comes from | Who reads it, and when |
|---|---|---|---|
| Answer time | Total wait seconds on answered calls / answered calls | Phone platform report | Supervisor daily, client at the review |
| Abandon rate | Calls dropped before answer / total inbound calls | Phone platform report | Whoever builds the schedule, weekly |
| Average work time | (Talk seconds + after call seconds) / calls handled | Phone platform report | Supervisor weekly |
| Calls per operator hour | Calls handled / operator hours staffed | Platform report against the schedule | Owner monthly, for capacity |
| Message accuracy | Messages scored with no field error / messages scored | Review of the recording against the message | Supervisor weekly, client monthly |
| Script adherence per account | Script points earned / script points available, per account | Scorecard review against the account script | Supervisor weekly, client monthly |
| Critical fail count | Count of calls with at least one auto fail | Scorecard review, flagged the same day | Supervisor same day, client monthly |
| Client complaint volume | Complaints logged against the account / period | Your own ticket log or account notes | Owner monthly |
Calls per operator hour and complaint volume are in the table for completeness, and neither one belongs in a quality target.
Reporting per account instead of per floor
Say your board runs 18 operators and 31,000 calls a month across 140 accounts. Score all of it, average it, and you get a floor number around 92.7 that looks fine on a slide.
One of those accounts is a plumbing contractor with 400 calls that month, scoring 68 because the operators keep missing a triage question added to the script in March. The other 30,600 calls average 93. Drop the bad account out of the math and your floor average moves three tenths of a point, so your reporting cannot see the client who is about to cancel.
Scores roll up per account, and the report goes out per account. Dashboards by agent and by client account exist for that reason, and that structure is what makes the quality report you send a client worth sending.
Setting targets when nobody publishes a benchmark
No answering service benchmark set held up when we went looking on 2026-09-08. What exists is general contact center data, collected on floors where a call runs three minutes and the agent works one employer's script all day.
SQM Group puts the industry average QA score at 85 percent, with 90 and above counted as good. Read that as the shape of a contact center rather than a target for a board answering in 140 business names.
SQM Group also reports that 60 percent of call centers evaluate five or more calls per agent per month, while Call Centre Helper puts a defensible sample at roughly 65 evaluations per agent per month, at 90 percent confidence, for an agent taking 1,400 calls. Both are true and they sit thirteen times apart.
Sixty five hand scored calls per operator per month does not happen on a floor of 18 with one supervisor. Build a baseline instead: score a month, take the median per account, and hold each account to movement against it.
The three numbers that belong in a client agreement
Answer time goes in because the client will ask for it regardless, and you would rather set the threshold than inherit one from whoever they used before. Message accuracy goes in because it is the thing they believe they are buying. A critical fail should reach them from you, the same day, before it reaches them from their own customer.
Average work time and calls per operator hour stay on the supervisor's wall, since a client who hears you are pushing work time down will read it as a promise about their calls.
What a recording proves, and what stays a judgment call
The split at the top of this page has a second half inside the metrics a person produces. Some of those criteria are decidable from a transcript, and the rest are somebody's opinion wearing a point value.
A recording settles the mechanical ones: whether the greeting matched the string stored for that account, whether the callback number matches the digits the caller said, and whether a call the account defines as an emergency reached the on-call schedule. Those are rules, each carries a timestamp, and an operator disputing a flag can be taken straight to the second it happened.
The others do not resolve that way. Whether the urgency call was right when the caller was vague about the pain, or whether the message reason matches what the caller meant instead of the words they picked. Aircall and StackAI both report that automated scoring holds up on binary rule based criteria and gets weaker on tone and empathy, which matches what we see. Every flag carries a confidence score and a link into the transcript rather than a verdict.
Two supervisors scoring the same ten calls will disagree on the subjective criteria, and the calibration variance between them is the honest measure of what those criteria are worth. Keep them on the scorecard, weight them under the rules, and stop reporting them to a client as facts.
A scored sample of your own calls is one supervisor, one recording, and the account script open beside it, deciding which flags survive.
So start with the two accounts that generate the most complaints, which is where QA for telephone answering services earns its first hour. Run a month of those calls against the rule based criteria, then spend the supervisor time only on the flags that came back under the confidence line.
Frequently asked questions
What KPIs should an answering service track?
Six carry the load: answer time, abandon rate, and average work time out of the phone platform, then message accuracy, script adherence per account, and critical fail count out of a review of the call. Calls per operator hour matters for scheduling and says nothing about quality.
What is a good answer time for an answering service?
The standard lives in the client's service agreement, and no industry rule overrides it. Score each account against what it signed for, since a 20 second threshold on a medical account and no stated threshold on a contractor are two different promises.
How do you measure answering service quality?
Score calls against a rubric written for the account the call came in under, then roll the scores up per account rather than per floor. The criteria carrying the weight are the account greeting, caller capture, urgency triage, and message accuracy, and the recording is what makes any of them provable.
What is average work time in an answering service?
Talk seconds plus after call seconds, divided by calls handled. Supervisors push on it when the board is busy, and pushing it down usually costs message accuracy, because the time being cut is the read back and the confirmation.
How often should an answering service report to clients?
Match the client's own reporting period, monthly for most accounts and quarterly for the ones that run formal reviews. A critical fail waits for neither, so send that one the day it happens.
Sources
- Trinsic feature sheet and call notes, 2026-08-27 and 2026-09-03. Verified internally.
- Answer time standards set per account in the client service agreement. Verified internally from account script practice.
- SQM Group on the 85 percent industry average QA score and 90 percent as good. Third party, unverified.
- SQM Group on 60 percent of call centers evaluating five or more calls per agent per month. Third party, unverified.
- Call Centre Helper on roughly 65 evaluations per agent per month at 90 percent confidence. Third party, unverified.
- Aircall and StackAI on automated scoring holding up better on rule based criteria than on tone. Third party, unverified.
- No published answering service benchmark set found. Verified as an absence on 2026-09-08.