Call QA scorecard template for contact centres: three gates and eleven lines

A call QA scorecard for a contact centre needs three parts: a few gates that fail a call outright, about ten yes-or-no lines grouped into customer, company and engagement, and a rule for lines that do not apply. The template below has three gates and eleven lines worth 100 points, split 40, 30 and 30. Replace the gates with what your regulator or client requires, then calibrate the form on five shared calls before anyone is scored.

Andrey TikhomirovCRO, Spellit7 min read

The grouping follows a trade guide to contact centre quality scorecards, first published in 2019 and last modified in August 2026. It builds a voice scorecard on three areas: what matters to the customer, what the company and its regulators require, and how the agent engages.

Three gates and eleven lines fit a call QA scorecard on one page

Read the table by block. Gates come first and sit outside the points. The eleven scored lines are yes, no or not applicable, and a yes needs the sentence from the call written beside it.

BlockYes-or-no lineCounts as yes whenPoints
GateWas the caller identified before any account detail was shared?The identity check is passed before the first account detail is saidFails the call
GateWas the disclosure your regulator or client requires given when due?The required wording is said at the point your script setsFails the call
GateDid every commitment stay inside policy?No refund, credit or deadline is promised beyond what policy allowsFails the call
CustomerDid the agent say the reason for the call back in their own words?The agent paraphrases the problem and the caller accepts it8
CustomerDid the caller leave with an answer, an action done, or a named next step with a time?One of the three is said before the call ends12
CustomerWas the explanation clear the first time?The caller does not repeat the same question10
CustomerDid the agent check that nothing else was open?An "anything else" question comes before the close10
CompanyDid the agent open with the greeting and name the script requires?Both are said5
CompanyWas the case logged with the correct reason?The reason code matches the conversation when checked against the record10
CompanyWas the promised follow-up recorded in the system?A callback, ticket or refund exists with an owner and a date10
CompanyWere holds and transfers announced?The caller is told why and for how long before the hold or transfer5
EngagementDid the agent acknowledge the caller's concern in words before moving to the fix?One sentence names the problem or the feeling10
EngagementDid the agent let the caller finish the first explanation?No interruption in the caller's opening turn10
EngagementDid the agent close with a summary the caller confirmed?The next step is repeated and the caller agrees10

Four things are recorded beside the score and never inside it: the reason for the call, the channel and length, whether the query was resolved on first contact, and any line flagged outside the agent's control. Spellit Actions builds the call quality rubric from the review form a team already uses, in the team's own wording, and each item is marked mandatory or optional, so a table like this one maps onto the way the product keeps a rubric.

Customer, company and engagement carry 40, 30 and 30 points, and the split is a default

The guide notes that many contact centres weight the three areas equally, a third each. It adds that heavily regulated sectors such as banking may weight the company area more, while retail may weight customer and engagement more. This template tilts toward the customer block, 40 against 30 and 30, because that block holds the question a form covers worst: whether the caller got what they called about.

The tilt is a default worth starting from, not a standard. Move at most 10 points between blocks, and write one sentence next to the change that says why. The wider imbalance between what gets measured on every call and what gets read by a person is covered in QA sampling in contact centres, and it is the reason the customer block is worth weighting up.

Three gates is a ceiling, and each gate has to be checkable in the transcript

Gate counts follow regulation more than taste. Voxjar's published call center templates, checked on November 16, 2026, show the range: its support and customer experience template has 13 lines worth 90 points and one gate, identity verified before account details, while its collections compliance template has 17 lines and five gates.

Keep a gate only if the regulator or the client demands it and a second listener can find it in the recording. Where the regulator requires exact wording, check the wording. Everywhere else, check the step: script adherence is better measured as whether the step happened than as whether the sentence matched.

If most calls trip a gate, the list is too broad, and the gate has become a scored line that was never weighted. Bring your gate list to a demo if you want to see which of them a week of your calls would trip.

Not-applicable lines leave the denominator, and the report should print how many did

A line that does not fit the call comes out of the score. The same guide recommends an N/A box on every question and excludes ticked N/A items from the final score. Take a call where the acknowledgment line (10 points) and the hold line (5 points) do not apply. Only 85 points are available, and an agent who earns 68 reads 80, not 68.

The share of lines that do not apply varies widely between teams. Across nine teams at six companies in September 2026, on 83,671 scorecard rows in Spellit Actions, the share of yes-or-no lines left blank ran from 2.1% to 81.6%. The top figure rests on 1,475 rows after that team's form changed mid-month. The full table is in the piece on not-applicable lines.

Print the count of applicable lines beside every score. Without it, two agents with the same 80 may have been scored on very different forms.

Calibrate on five shared calls before the first score counts

Five calls and a second reviewer are a default worth starting from, not a measured minimum. The steps:

  1. Pick five recorded calls: two routine, one with an upset caller, one with a transfer, one where a gate could be tripped.
  2. Have each reviewer score all five alone, with the sentence from the call beside every yes.
  3. Compare line by line. A line where reviewers differ on two of the five calls gets rewritten, because the wording is the problem and the reviewers are not.
  4. Set one cut line. Start at 80 and move it after the first month, once you can see what share of calls falls under it.
  5. Agree who decides disputes. Calibration of this kind is written up for sales calls, and the same sessions work for a service queue.

Send the form to the agents before the first scored week. Ask for a demo with your current form if you want to see it run against your own calls.

Four mistakes make a scorecard unfair to the agent

A scorecard that agents consider unfair turns coaching into an argument about the form. Four causes are worth ruling out first:

  • Scoring what the agent does not control, such as a queue wait or a system outage. Flag those lines as outside the agent's control and keep them out of the points.
  • Gates that need judgment. "Professional tone" is a scored line. A gate has to be something a second listener can find in the recording.
  • One number per agent with no count of applicable lines behind it.
  • Stopping at the score. A weak result is useful when it becomes a task: in Actions it turns into a task for a named rep instead of a number in a spreadsheet, and a rule picks one to three tasks per week from the rep's own calls.

For contact centres, call quality review across every call is where those tasks come from, and the contact centres solution page describes the same setup for service queues.

What to do next

Copy the table into a spreadsheet this week. Replace the three gates with the ones your regulator or client actually asks for, delete any line your channel cannot produce, and score five calls with a second reviewer. Count how many lines each call left not applicable before you read a single total.

Bring the form you use today. A rubric built from your own review form, scored on every call, with each score opening into the quote behind it, is what the call quality page describes. See your own calls scored against your form.
See the demo now. Then run it on yours.

A live demo of our products and a real conversation about the growth problem you need to solve.

Book a demo

FAQ

How many lines should a call QA scorecard have?

Eleven scored lines plus up to three gates is a size one reviewer can hold in their head, and it is the size of this template. Voxjar's support template has 13 lines worth 90 points, checked on November 16, 2026, which puts it in the same range. Add lines only when a gap shows up in disputes or in calibration, not because a section looks thin. Every line you add makes the form slower to apply and raises the share of lines that will come back not applicable on some calls. If a block feels too short, check whether its lines are really one question asked twice. If a block feels too long, the line that never changes a score is the first to cut.

What is a good pass mark for a call QA scorecard?

No published standard sets one, so start with a single cut line at 80 and treat it as a default to move. Score the first month without acting on the line, then look at the share of calls that fall under it. A cut line that fails most calls is set wrong or sits on lines the agents never saw in advance. A cut line nobody falls under teaches nothing. Move it until the calls under it are a number a team leader can actually review in a week, and write the date of each change on the form. Share the cut line with the agents before the first scored week, so nobody finds out the bar from a result.

Should a gate fail the whole call or only subtract points?

A gate should fail the whole call, and the list should stay at three or fewer. Gates cover what a regulator or client requires, for example identity verification before account details, and a high total elsewhere must not offset a miss there. Record the failed gate next to the score and send the call for review, so a person looks at it instead of the formula deciding. If a line would not justify failing the call on its own, it belongs in the scored blocks with a weight. Review the gates once a quarter and remove any that no regulator or client still requires.

Can one scorecard cover phone, chat and email?

Not without edits. The guide's template is for the voice channel only, and several lines here, such as interruption in the opening turn and announced holds, exist only on a call. Keep the three blocks and the weights, and swap the channel-specific lines for ones that fit the channel: response time and message clarity in chat, for example. Score each channel against its own copy, and compare scores only inside a copy. A single blended number across channels hides the difference between a form that fits the work and one that does not. Name each copy by channel, so a score never travels without the form it came from.

Can an agent dispute a call QA score?

Yes. In Spellit Actions any rep can appeal a score, and a person makes the final decision. The appeal names the call and, if needed, the rubric item, states the score the rep thinks is right and carries a written justification. The person deciding sees the AI's verdict, its reasoning and the quote on one side and the rep's argument on the other, then applies the correction or keeps the score and writes down why. An administrator in your organization decides, and every step goes into an audit log. On your own form, the same habit works with a spreadsheet: no score is final until the agent has seen the quote.