Last updated: September 9, 2026
What a call center QA scorecard template is
A call center QA scorecard template is a fixed list of scored criteria that a reviewer applies to every evaluated call. Sections carry weights, individual criteria carry point values, and a short list of auto fail items zeroes the score outright. The template below runs to 100 points across 17 criteria.
The 100 point scorecard
Six sections, 17 criteria, 100 points.
| Section | Criterion | Points |
|---|---|---|
| Opening and identification (15) | Answered with the approved branded greeting, including company name and agent name | 5 |
| Opening and identification | Verified the caller's identity to the level the account policy requires | 5 |
| Opening and identification | Delivered the recording or monitoring notice where the jurisdiction requires it | 5 |
| Discovery and listening (25) | Asked at least two probing questions before proposing a solution | 8 |
| Discovery and listening | Let the caller finish without interrupting | 6 |
| Discovery and listening | Restated the caller's issue accurately before acting on it | 6 |
| Discovery and listening | Captured caller name, callback number, and reason for the call | 5 |
| Communication quality (20) | Held an audible pace with clear articulation | 5 |
| Communication quality | Used an empathy statement where the caller expressed frustration | 5 |
| Communication quality | Stayed inside the filler word and slang tolerance | 5 |
| Communication quality | Framed a limitation in positive terms with an alternative | 5 |
| Resolution and accuracy (25) | Gave information that matches current policy | 10 |
| Resolution and accuracy | Resolved the call without an avoidable transfer or hold | 7 |
| Resolution and accuracy | Set an accurate next step with a timeframe | 8 |
| Compliance and process (10) | Read the required disclosure lines in full | 5 |
| Compliance and process | Documented the interaction in the system of record | 5 |
| Closing (5) | Offered further assistance and used the approved sign off | 5 |
| Total | 17 criteria | 100 |
Take the whole rubric as a spreadsheet: download call-center-qa-scorecard.csv. The file is ungated, so it costs you one click.
One hundred points is a convention rather than a finding. Nobody has shown that a 100 point rubric coaches better than a 90 point one. It divides cleanly, which is the entire argument for it.
The weights are there because a wrong policy answer and a missed sign off are not the same mistake. Auto fails cover the handful of behaviors bad enough that the arithmetic stops mattering.
Auto fail items
Each row below zeroes the whole score no matter how many points the agent earned elsewhere.
| Auto fail | Why it is absolute |
|---|---|
| Skipped a required legal or recording disclosure | Creates regulatory exposure that no other score offsets |
| Gave actionable information that contradicts current policy | The caller acts on the wrong instruction |
| Used profanity or a hostile tone toward the caller | Brand damage independent of the outcome |
| Disconnected a live caller without cause | The interaction never had a chance to succeed |
| Released account or protected information to an unverified caller | Data exposure |
How to score a call with this template
- Pick the call. Random selection inside a window beats letting a supervisor choose, because chosen calls tend to be the ones somebody already complained about.
- Score against the transcript rather than your memory of the call. Scoring from memory a week later sucks, and it produces numbers nobody will defend in a coaching session.
- Mark any auto fail before you total anything. Totaling first tempts a reviewer to soften the flag so the number looks reasonable.
- Total the sections, then the whole card.
- Record the timestamp behind every point you deducted. Programs skip this one, and it is what makes a disputed score resolvable.
How the score is calculated
Points earned divided by 100.
Say a reviewer scores an inbound call on a property management account. The agent opens with the branded greeting, delivers the monitoring notice, verifies the caller, asks three probing questions, and books the maintenance visit with a date. Two lines come off: the agent never restated the caller's issue before acting on it, worth 6, and took the callback number without reading it back digit by digit, worth 5. The call scores 89.
Eighty nine sits above the 85% that SQM Group puts at the industry average and just under the 90 to 99% band the same source calls good. It also hands the supervisor two timestamps to play back, which does more for the agent than the number does.
How to adapt the weights to your program
The section weights are the part you should change. Criteria travel across programs reasonably well; weights are where your own risk lives.
Compliance heavy programs move points out of communication quality and into compliance and process. A Medicare agency scoring TPMO disclaimers and a Scope of Appointment collected before the appointment needs more than 10 points sitting on compliance. Sales programs add a discovery section and push resolution down.
We started this rubric with 20 points on compliance and walked it back to 10. The auto fail list already handles the breaches that end up in a regulator's file, and scoring a missed disclosure twice, once as a deduction and once as a zero, tells a supervisor nothing the zero did not already say. Run without auto fails and those points belong back in compliance.
An answering service scoring after hours work against an account script has a different problem again. Whether the agent followed the on call schedule and made the right dispatch decision matters more than pace, so resolution and accuracy absorbs points from communication quality.
If you are starting from an empty page, building a scorecard from scratch covers the order to do it in.
What makes a criterion checkable
Most scorecards die on one line. Somebody writes "showed empathy," an agent scores 3 out of 5, and the coaching session becomes an argument about whether the agent sounded like they cared. Neither person can win that argument, because there is nothing to point at.
A criterion is checkable when two reviewers scoring the same call separately land in the same place, because the evidence sits in the recording instead of in their heads. In practice that means the criterion names an observable event: a phrase, a number read back, a hold, a transfer, a disclosure line delivered in full.
Every criterion in the table above was written to pass that test. The rewrite usually looks like this.
| Vague criterion | Checkable rewrite | What a reviewer points at |
|---|---|---|
| Showed empathy | Used an empathy statement where the caller expressed frustration | The phrase and its timestamp |
| Was professional | Stayed inside the filler word and slang tolerance | A count of flagged terms in the transcript |
| Handled the call well | Resolved the call without an avoidable transfer or hold | Hold and transfer events on the timeline |
| Followed the script | Read the required disclosure lines in full | The disclosure text matched line by line |
CallSniper scores against the transcript for exactly this reason. Every line carries a timestamp and a per line confidence score, the vocabulary library detects the categories most rubrics already ask about (branded greetings, empathy phrases, probing questions, filler words and slang, offers to assist, caller ID capture), and a critical fail flag zeroes the card the same way the auto fail rows above do. When a supervisor disputes a deduction on one of our call scorecards, they open the call at the second the criterion was scored and listen to it.
That works better on some lines than others, and pretending otherwise would be a bad way to start a QA program. Detecting whether a callback number was read back digit by digit is close to arithmetic. Judging whether an empathy statement actually landed, on a caller who had just been told their dispatch slipped four hours, is a human read, and automated scoring is measurably less reliable on tone than on rule based criteria (Aircall, StackAI). Machine scoring on the binary lines with a human pass over the subjective ones is what holds up. Track calibration variance alongside it so you know how far your reviewers have drifted from each other.
A scorecard also cannot fix a bad account script. When the script has agents capture the issue before the callback number and the account wants the number first, every scored call returns the same deduction, and the fix lives in the script rather than in anybody's coaching.
Five mistakes that make a scorecard useless
- Soft criteria carrying too much weight. Once tone and empathy add up to more than a quarter of the total, the score mostly measures reviewer taste.
- Criteria nobody can evidence. Move anything a reviewer can only justify with "it felt like" into coaching notes.
- Auto fail lists that keep growing. The flag means emergency at five items and means very little at fifteen.
- Weight changes made mid quarter with no note. Whoever inherits the dashboard next quarter has no way to tell a coaching win from a reweighting, and the trend line you were about to walk an agent through stops comparing to last month.
- Scored calls that no supervisor ever plays back. A number sitting in a dashboard has never changed how anybody answers a phone.
Download the template
The CSV holds the same 17 criteria and point values as the table above plus the five auto fail rows, in four columns: section, criterion, points, auto fail.
Download call-center-qa-scorecard.csv
Other rubrics, including the answering service version, live in the full template library.
Frequently asked questions
What should a call center QA scorecard include?
A scorecard needs weighted sections, criteria written so a reviewer can point at evidence for each one, and a short list of auto fails. Keep the total at a round number so scores compare cleanly across teams and across months.
How is a QA score calculated?
Add the points earned across every criterion and divide by the maximum, which is 100 in this template. An auto fail overrides the arithmetic and sets the call to zero.
What is an auto fail in QA?
An auto fail is one behavior serious enough to zero the whole call, usually a compliance or safety breach. Most programs hold the list at five items or fewer so the flag keeps its meaning.
How many calls should QA review per agent?
Most contact centers review five or more calls per agent per month, and 60% land at or above that number (SQM Group). Say an agent takes 1,400 calls a month: a statistically defensible read on that agent needs roughly 65 evaluations at 90% confidence (Call Centre Helper). The number moves with volume, so run yours through how many calls to review per agent.
Can AI score this scorecard?
Automated scoring handles the binary criteria well, such as whether a disclosure was read or a callback number was captured. Tone and empathy criteria still need a human read. The practical setup is machine scoring across every call with human calibration on the subjective lines.
Run this scorecard against calls you already have
Most programs get eyes on 1 to 3% of their calls (Verint, Gistly, ViciStack), which is a thin base to argue from. An afternoon of running this rubric against calls already sitting on your servers settles more than a month of debating weights.
Bring a folder of recordings and an agent roster, and the paid pilot does the rest: six months of your own historical audio scored against this template, then 30 minutes on Zoom where your team watches their own calls come back scored. It runs at a fixed pilot price. Manual bulk upload keeps the phone system out of it. The demo is the shorter version on a handful of calls.
Sources
- SQM Group, on QA benchmark scores and evaluation volume per agent. Third party, unverified by us.
- Call Centre Helper, on sample size for a statistically defensible read at 90% confidence. Third party, unverified by us.
- Aircall and StackAI, on where automated scoring is reliable and where it is weaker. Third party, unverified by us.
- Verint, Gistly, and ViciStack, on the share of calls manual QA reviews. Third party, unverified by us.
- Trinsic Technologies feature sheet, on CallSniper scorecards with points per criterion, critical fail flags, transcript timestamps, per line confidence scores, the vocabulary library, and manual bulk upload. Verified.