A customer support QA scorecard reviews the whole human interaction, from understanding the question to a clear resolution or transfer. For a small team, this template uses five criteria rated from zero to two, with a separate critical-error flag. The purpose is to choose a specific improvement and agree on standards rather than rank agents from a handful of random conversations.
Define the scale and a comparable sample
Use this suggested scale: zero means unmet, one partially met, and two fully met with conversation evidence; include different types of issue. Everyone knows the criteria and sample scope in advance, and a non-applicable criterion is marked separately rather than receiving automatic full points.
Example: eight points out of ten across five applicable criteria equals eighty percent on your internal scale, not an industry benchmark. Do not compare agents when one gets only easy questions and the other gets complaints and technical failures.
Front describes QA as reviewing conversations against internal standards alongside customer metrics; the five criteria and scale here are our proposal. Front: conversation quality with a QA scorecard.
This evaluates human support; individual AI suggestions belong in AI Copilot draft quality.
Criterion: understanding the actual issue
Check whether the agent established the customer’s goal, read the context, and asked only questions needed to prepare a useful answer. Full marks require addressing the real question rather than selecting a template because one word looks familiar.
Example: the customer asks about cancelling an appointment, so explaining only how to book another one does not address the need. One clarifying question can help; repeatedly asking for information already in the conversation weakens the assessment.
Scroll the table sideways to see all columns.
| Check | Action | Evidence of completion |
|---|---|---|
| Define the scale and a comparable sample | Use this suggested scale: zero means unmet, one partially met, and two fully met with conversation evidence; include different types of issue. | Everyone knows the criteria and sample scope in advance, and a non-applicable criterion is marked separately rather than receiving automatic full points. |
| Criterion: understanding the actual issue | Check whether the agent established the customer’s goal, read the context, and asked only questions needed to prepare a useful answer. | Full marks require addressing the real question rather than selecting a template because one word looks familiar. |
| Criterion: accurate facts and conditions | Compare the answer with approved information and inspect requirements, exceptions, and the boundaries of the agent’s authority. | Two points require correct facts and material conditions; a small stylistic omission is not equivalent to a false policy. |
| Criterion: clear language and the next action | Assess whether the customer can understand the message without internal jargon and perform a specific next step. | Full marks need an action, location, or expected result; politeness and good grammar alone do not replace usable instructions. |
| Criterion: ownership of what happens next | Check assignment, agreed follow-up, and whether a transfer preserved context and identified the owner of the next action. | Two points mean a clear continuation or verified completion rather than leaving the customer with “someone will get back to you”. |
| Criterion: safety and respect | Review the minimum data requested, verification procedure, and communication that addresses the situation without insults or unsupported promises. | Full marks mean following the approved data procedure and communicating respectfully; a serious disclosure is a separate critical error. |
Worksheet template: conversation reference, issue type, five ratings, reasons for any rating below two, critical error, improvement, and review date. For non-applicable criteria, calculate the total against the available points and explain the exclusion. Do not let a high total hide a disclosure or a materially false promise. Conversation organization is covered in a customer support ticket workflow.
Criterion: accurate facts and conditions
Compare the answer with approved information and inspect requirements, exceptions, and the boundaries of the agent’s authority. Two points require correct facts and material conditions; a small stylistic omission is not equivalent to a false policy.
Example: an instruction gives the right deadline but omits a condition for the selected variant, leaving the answer incomplete. A false financial commitment or unauthorized decision receives a critical flag regardless of the other scores.
A calibration example: both reviewers read an answer with the correct deadline but no exception for the customer’s variant. One awards a single point for facts; the other gives two because they checked only the date. They open the approved rule and agree that the relevant exception is part of full accuracy.
Store this example under the existing criterion rather than creating another category. Agent feedback addresses one behavior: checking the variant before sending a deadline. A later case with that condition shows whether the improvement was applied.
Criterion: clear language and the next action
Assess whether the customer can understand the message without internal jargon and perform a specific next step. Full marks need an action, location, or expected result; politeness and good grammar alone do not replace usable instructions.
Example: replace “please verify in the system” with the screen name and what the customer should see there. Do not penalize a different style when it is clear, suitable, and preserves the same meaning.
- Have two reviewers rate the same conversation independently before calibration.
- Discuss differences using the exact sentence and criterion rather than impressions.
- Record the agreed decision and an example for later reviews.
- Give the agent one strength and one specific improvement.
- Use the next review to check that improvement rather than repeat every criticism.
Calibration does not mean averaging opinions. If one reviewer scores zero for a missing exception and another gives two, find the approved rule and clarify the criterion. Make feedback observable: “The message omitted the condition; add it before the link.” Useful internal notes help assess transfer quality.
Criterion: ownership of what happens next
Check assignment, agreed follow-up, and whether a transfer preserved context and identified the owner of the next action. Two points mean a clear continuation or verified completion rather than leaving the customer with “someone will get back to you”.
Example: the agent explains what the team will check, adds a note, and gives an agreed update time they can actually honor. Do not demand an instant resolution when the process correctly requires a decision from an authorized specialist.
The Meetlead inbox lets you inspect conversations, assignments, tags, and notes. An account and configured website support team work. Selected cases become material for your own QA scorecard; this does not imply automatic scoring of these five criteria by the product. Base the review on the actual support standard rather than an arbitrary target number. The guide to customer service standards shows how to write that standard down on a single page.
Criterion: safety and respect
Review the minimum data requested, verification procedure, and communication that addresses the situation without insults or unsupported promises. Full marks mean following the approved data procedure and communicating respectfully; a serious disclosure is a separate critical error.
Example: a ticket reference is enough to identify the issue, so the agent does not request a password or complete document. Do not use the scorecard to enforce artificial friendliness; respect and suitability matter more than a scripted emotional style.
When a customer returns after a highly rated chat, investigate repeat support tickets instead of awarding extra points for fast closure.
Questions that come up in practice
How many criteria should I start with? Five are sufficient for this template: goal, facts, clarity, ownership, and safety. Add another only when it captures a distinct important behavior rather than duplicating an existing criterion.
Does the percentage tell the whole story? No. The total does not replace reasons or critical flags. Inspect individual criteria and case comparability before drawing conclusions about a particular agent.
How should I deliver a low rating? Show the exact sentence, required standard, and possible correction. Give the agent space to explain missing context and agree on how the improvement will be checked.
