A call center QA checklist is the list of items a reviewer checks on every call, from identity checks to resolution. Use the one below as a starting point, grouped by call phase, and adapt it to your rules.
In short
- Write each item so the call itself answers it.
- Keep the auto-fail list short and truly severe.
- Calibrate reviewers on it or scores won’t match.
- Run it on every call, not a sample.
Why a checklist, and what makes one work
A scorecard defines what a good call is. A checklist is how a reviewer applies it without drifting. The difference matters because the most common failure in call center QA is not a missing criterion, it is two reviewers scoring the same call differently, which is what teaches agents that their score depends on who reviewed them rather than on how they performed.
A checklist fixes that by making each item a specific, answerable question. Not was the agent professional, which two people will judge differently, but did the agent acknowledge the reason for the call before moving to a solution, which they either did or did not. Phrasing every item so it can be answered from the evidence is the single biggest lever on scoring consistency, and it is the same discipline as writing a rubric an AI can score. This checklist is call-specific; if you support tickets and chat as well, pair it with the broader customer service QA checklist.
The checklist, by call phase
Grouping by phase keeps reviews fast and complete. Adapt the wording, but keep each item answerable.
| Phase | Check | Auto-fail candidate? |
|---|---|---|
| Opening | Agent greeted and identified themselves per standard | No |
| Opening | Identity or account verified before any account action | Yes |
| Opening | Agent acknowledged the reason for the call before solving | No |
| Handling | Agent asked the questions needed to diagnose correctly | No |
| Handling | Information given was accurate and current | Yes, if wrong on a material point |
| Handling | Policy stated correctly, not invented or overstated | Yes, if a false commitment was made |
| Handling | Agent set clear expectations for timing and next steps | No |
| Handling | Customer was not made to repeat themselves unnecessarily | No |
| Resolution | The customer’s actual issue was resolved or a clear path set | No |
| Resolution | Agent confirmed the resolution with the customer | No |
| Compliance | Required disclosures or consent language given | Yes |
| Compliance | Sensitive data handled correctly | Yes |
| Close | Agent summarized outcome and next steps | No |
| Close | Any handoff carried full context to the next person | No |
Deciding which items auto-fail
The right-hand column flags candidates, not conclusions. An auto-fail zeroes the whole evaluation on one item, so it should be reserved for failures that are non-compensable: a security or compliance breach, a false commitment, mishandled sensitive data. Keep the list short, because a long auto-fail list stops the rest of the scorecard from carrying information. The full test for what earns that power, and how to run auto-fails without losing agent trust, is in auto-fail criteria in QA. Everything not on the auto-fail list is a weighted item: it lowers the score, sometimes heavily, but does not destroy it.
Making the checklist produce trustworthy scores
A checklist on its own does not guarantee consistency; two people can still read did the agent set clear expectations differently. Two things close that gap. The first is calibration: reviewers score the same calls, compare, and sharpen any item they disagreed on until the checklist reads the same way to everyone.
The second is coverage. A checklist applied to a handful of calls per agent gives you a noisy read that agents can dismiss. Applied to every call that lands in your helpdesk, it becomes a full picture of where the team is strong and where a specific item keeps failing. Whatever tool you use, make every checklist result point back to the exact moment in the call, so an agent can see precisely what a failed item refers to, rather than being handed a score they have to take on faith. On the written side, at UiPath, Kaizo automated 100% of QA on support conversations with 200% ROI and an 8% lift in quality score, which is what running a checklist at full coverage makes possible.
Frequently asked questions
What should be on a call center QA checklist?
Items grouped by call phase: opening (greeting, identity verification, acknowledging the reason), handling (correct diagnosis, accurate information, correct policy, clear expectations), resolution (issue actually resolved and confirmed), and compliance and close (required disclosures, data handling, a summary, and a clean handoff). Each item should be answerable from the call itself rather than from reviewer opinion.
Which checklist items should be auto-fails?
Only the non-compensable ones: failing to verify identity before an account action, giving materially wrong information, making a false commitment, missing a required disclosure, or mishandling sensitive data. Keep the auto-fail list short, because once many items can zero a call, the rest of the checklist stops carrying useful information.
How do you keep two reviewers scoring a checklist the same way?
Phrase every item as a specific, evidence-answerable question, then calibrate reviewers by having them score the same calls and reconcile any item they disagreed on. Calibration is what turns a shared checklist into consistent scores, so an agent’s result reflects how they performed rather than who reviewed them.
Can a call center QA checklist be applied to every call?
Yes. Manual review limited teams to a sample, but automated scoring can apply the checklist to every call that lands in the helpdesk. Full coverage turns the checklist from a spot check into a complete picture of where a specific item keeps failing, with each result traceable to the exact moment in the call.
Run your QA checklist on every conversation, not a sample
Bring your current checklist and a week of chat, email and ticket conversations. We will show you how it scores across every one of them rather than a handful, which items your reviewers disagree on, and how each result traces back to the evidence in the conversation so agents can trust and act on it.
