Last updated: September 12, 2026
What a call center audit checks
A call center audit covers two levels. One is a sample of calls, checked against recording, disclosure, verification, and documentation requirements. The other is the QA program itself: rubric versioning, calibration cadence, and whether a deduction can be traced back to the second of the call it came from. Both checklists below are pass or fail.
Almost nobody audits their own QA program out of curiosity. The checklist comes out because somebody outside the QA team wants assurance. A client runs a governance review before renewal. A carrier asks an agency how it monitors the people selling its product. Risk or legal asks after a complaint, or a procurement questionnaire arrives with a quality section nobody on the floor has read.
That trigger decides what the checklist has to prove. Self assessment is for getting better, and a rough answer there still moves you forward. Somebody from outside wants to know whether to believe you, and your own read on your program carries little weight in that room. The artifact does.
Which is why an item answered yes is worth nothing on its own. The program checklist below carries an evidence column for that reason, and the honest way to use it is to run the whole thing with a folder open, marking a fail anywhere the document does not exist yet.
Audit gets used loosely as a word, and half the checklists published under it are a weighted scorecard with the points stripped out. A scorecard grades a call and hands an agent a number they can argue with. An audit asks whether the machinery producing that number holds up, which tends to happen once a year and usually because somebody asked.
The call sample audit checklist
Fifteen checks, applied to a sample of calls you pulled by a method you can describe out loud. One check per line, so nothing ends up with half a pass, and a fail here is a finding rather than a deduction against the agent.
| # | Area | Check | What a pass requires |
|---|---|---|---|
| 1 | Recording | The call was recorded from greeting to disconnect | Audio that starts before the first spoken word and runs to the release, with no gap in the middle |
| 2 | Recording | The recording can be retrieved by call identifier inside the audit window | The file pulled from the archive during the audit using the identifier held in the system of record |
| 3 | Disclosure | The required disclosure was delivered | The disclosure heard on the recording, in the wording the current script sets |
| 4 | Disclosure | The disclosure landed before the part of the call it has to precede | A timestamp for the disclosure earlier than the timestamp for the first substantive exchange |
| 5 | Disclosure | The disclosure used matched the jurisdiction the caller was in | The caller's state on the record, checked against the routing table the script library holds |
| 6 | Verification | The caller was verified before any account detail was released | The verification steps heard on the recording, in the order the account or program requires |
| 7 | Handling | A transfer or escalation followed the routing the program defines | The routing rule in writing, against where the call actually went |
| 8 | Documentation | The outcome was written into the system of record | The record itself, timestamped inside the shift the call belongs to |
| 9 | Documentation | The written record matches what the recording contains | Both side by side with the reason for contact compared line for line |
| 10 | Documentation | The callback number in the record matches the number read back on the call | The digits on the recording against the digits in the record |
| 11 | Documentation | A promise made on the call has a follow up logged against it | The follow up task or ticket, with an owner and a due date |
| 12 | Scoring | The call was scored under the rubric version in force on the call date | The version number on the scored card against the change log entry covering that date |
| 13 | Scoring | Every deduction on the scored card carries a timestamp into the recording | The timestamp on each deduction, played back and confirmed |
| 14 | Scoring | Any critical fail on the card matches an event you can hear | The moment on the recording where the failure happened |
| 15 | Scoring | The agent named on the card is the agent on the recording | Speaker attribution confirmed against the login record or the agent's own introduction |
Items 3 through 5 are the ones that turn into real findings, since a script that was current two jurisdictions ago passes an internal read and fails the moment somebody checks which state the caller was actually in. Wording and timing rules live on the disclosure page rather than here.
The QA program audit checklist
Nineteen checks across ten areas, each with the artifact that satisfies it. An item where the practice is real but the artifact does not exist is a fail, because an auditor has no way to tell an undocumented habit from a story.
| # | Area | Check | Evidence that satisfies it |
|---|---|---|---|
| 1 | Program documentation | A written QA policy names who scores calls and how often and against what | The document itself, dated, with the date of its last review on it |
| 2 | Rubric versioning | Every rubric carries a version number and an effective date | The rubric file with both on it |
| 3 | Rubric versioning | Rubric changes are logged with the reason for the change | A change log with dated entries, each naming the criterion and why the weight moved |
| 4 | Rubric versioning | A scored call resolves to the rubric version in force on the day of the call | Three scored cards from three different quarters, each matched to its version |
| 5 | Sampling | Calls are selected by a written rule rather than by reviewer choice | The selection rule in writing, plus the list of calls it produced last month |
| 6 | Sampling | Coverage is reported next to every average | Calls handled and calls scored, per agent, per period |
| 7 | Sampling | The sample spreads across agents and queues and across the hours the floor runs | Last quarter's sample broken out by agent, by queue, and by time of day |
| 8 | Calibration | Calibration runs on a stated cadence | Dated session records naming who attended and which calls were used |
| 9 | Calibration | Variance between reviewers is measured rather than assumed | The same calls scored blind by each reviewer, with the spread written down |
| 10 | Calibration | A criterion that produces repeat disagreement gets rewritten | A change log entry that traces back to the session where the disagreement showed up |
| 11 | Dispute process | An agent can dispute a score inside a stated window | The written policy naming the window and who hears the dispute |
| 12 | Dispute process | Disputes are logged with their outcome | The log, with dates, resolutions, and any score that changed |
| 13 | Coaching | A card below the program's threshold carries a coaching action with an owner and a date | The coaching record, tied to the call identifier it came from |
| 14 | Coaching | The coaching action gets checked at the next review | The next period's record referencing the prior action by name |
| 15 | Compliance coverage | Criteria that carry regulatory weight are marked on the card and mapped to the rule | The rubric with the rule cited beside each of those lines |
| 16 | Compliance coverage | The disclosure script is current for every jurisdiction the operation takes calls from | The dated script version and the routing table that picks between them |
| 17 | Recording and retention | A retention schedule is written against the longest obligation that applies and deletion happens on it | The policy, one recording retrieved from the oldest month still inside it, and the deletion log |
| 18 | Access control | Playback of recordings is restricted by role and access is current | The user list with roles attached, an access log showing who played what, and two recent leavers checked against it |
| 19 | Client reporting | Client reports state how many of that client's calls the score was built on and which rubric version produced it | Last quarter's report as it was sent |
Items 17 and 18 get checked here and decided elsewhere. How long you have to keep the audio depends on the industry and the contract, which the retention page works through rule by rule. The audit only asks whether a schedule exists, whether it was written against the longest obligation on your list, and whether anybody has actually deleted anything on it.
Coaching follow through nearly came off this list, since an auditor testing whether your scores are defensible has no obvious business asking what you did about them. Both items stayed because a program that scores calls and changes nothing is a reporting exercise wearing a QA name tag, and a client reviewing a vendor asks about it every time.
One honest caveat about the list as a whole: passing it certifies nothing. No standard sits behind these items, and marking all nineteen green means your program can show its work, which is a lower bar than any certification and a higher one than most floors clear.
Download both checklists
Two files, named so you can tell them apart before you click. The call sample audit covers a folder of calls, and the QA program audit covers the program that scored them. Columns match the tables above, with nothing in front of either download.
Choosing the sample
A sample survives a challenge when the selection rule was written before the calls were pulled. Random inside a defined window, spread across agents and queues, sized against volume rather than against whatever fits in an afternoon.
The failure mode worth naming is the supervisor chosen sample. Calls somebody remembers well enough to suggest are usually calls somebody already complained about, so a sample built that way measures your complaint log.
Sizing is arithmetic, and it lives on the agent scorecard page next to the coverage column that belongs beside every average. Call Centre Helper puts a statistically defensible read at roughly 65 evaluations per agent per month for an agent taking 1,400 calls, which is a long way from the handful most floors score.
An audit sample can run smaller, since you are testing whether the system works rather than grading a person. Thirty calls across a quarter, pulled by a written rule and spread across agents and dayparts, is a reasonable start, and that number is our judgment rather than a published standard.
Auditing a vendor rather than your own floor
Both checklists work from either side of the table. A client auditing an answering service, a BPO, or an outsourced sales floor can ask for four artifacts before asking anything else: last quarter's sample selection rule, the rubric version in force with its change log, a calibration record with the variance on it, and a client report that names how many of their own calls the average was built on.
An operator who produces those in a day has a program. Three weeks means a spreadsheet and good intentions, which is fixable, though not during somebody else's review window.
If you run the floor, work through it on yourself in a quiet month instead. QA for telephone answering services hits this earliest, since the whole business is answering calls in somebody else's name and that somebody eventually asks how it is going. Scoring per client account is what makes branded client quality reporting worth sending, because the report is only as good as the script those calls were scored against.
Frequently asked questions
What is a call center audit?
A call center audit checks a sample of calls against recording, disclosure, verification, and documentation requirements, and checks the QA program that scores those calls against its own documentation. Clients and internal risk functions usually ask for the program level version. Compliance teams and carriers more often start with the call sample.
How is a call audit different from QA scoring?
Scoring produces a number for one call against a weighted rubric, and an agent can dispute it. An audit produces a pass or a fail against a requirement, with no points and no averaging. Both connect through evidence, since an audit of a scored call asks whether the deduction has a timestamp behind it. The full QA scorecard is the instrument an audit checks up on.
What should be on a call center audit checklist?
Two lists. One covers the calls: recording completeness, disclosure delivery and timing, caller verification, documentation accuracy, and whether the scored card matches what the recording holds. Its companion covers the program, including rubric versioning, sampling method, calibration, disputes, coaching follow through, retention, access control, and client reporting. Both download above.
How many calls should an audit sample?
Pull them at random inside a defined window, spread across agents, queues, and times of day, with the rule written down before the pull. Size against call volume rather than habit. The arithmetic behind a defensible sample sits on the agent scorecard page, along with the coverage percentage that tells you whether an average means anything.
What evidence should a QA program keep?
Four artifacts answer most of what an auditor asks: the evidence behind each deduction, the rubric version in force on the date of any scored call, calibration records with the variance on them, and the dispute log. Keep them as they happen. All four are cheap on the day and painful to reconstruct a year later.
What an auditor will ask you to produce
Play the last hour of any governance review forward and it lands in the same place. Somebody picks a scored call from eight months ago and asks to hear the thirty seconds behind the deduction. Then they ask which version of the rubric was in force that week, and whether the reviewer who scored it agreed with the other reviewer that quarter.
Most programs can answer the first question and stall on the second. The rubric changed in March, nobody wrote down why, and the March scores now sit in the same trend line as the January ones as though the standard never moved. Versioning the rubric is the cheapest habit on this page and the only artifact you cannot reconstruct truthfully once the quarter has closed.
Five things are worth keeping continuously, so that an audit becomes a retrieval task instead of a project: the change log entry written the day a weight moves, the calibration record written during the session rather than from memory, the dispute entry written when the dispute closes, the monthly selection rule saved next to the list of calls it produced, and the coverage number stored beside every average. None of that takes more than a few minutes on the day.
Evidence behind a deduction is the one piece software handles better than a filing habit. Searchable transcripts with timestamps turn those thirty seconds into a click, and every flag carries a confidence score so a supervisor spends the time on the shaky ones. Rule based criteria hold up well under that treatment and anything resting on tone gets less reliable, which Aircall and StackAI both note about automated scoring. A machine flag on empathy is a prompt for a human to listen rather than a finding you hand an auditor.
A separate checklist covers the compliance side, organized by rule with a source on every line. Open that one when the question is whether a practice is legal rather than whether it is documented.
If you want to see what the evidence layer looks like against your own audio, the paid pilot runs six months of your historical calls through the rubric and puts the findings in front of your supervisors, at a fixed pilot price.
Start with the change log. Fifteen minutes on the day a weight moves, and nobody can produce it truthfully in arrears.
Sources
- Trinsic feature sheet and call notes, 2026-08-27 and 2026-09-03. Verified internally.
- Call Centre Helper on roughly 65 evaluations per agent per month at 90% confidence for 1,400 calls. Third party, unverified.
- Aircall and StackAI on rule based scoring holding up better than tone scoring. Third party, unverified.
- Random selection over supervisor chosen samples. Practice guidance, our editorial judgment.
- A 30 call quarterly audit sample. Editorial judgment for this page.
- The nineteen program items and the ten areas they cover. Editorial judgment, mapped to no external standard.