Short answer
A QA score is the grade a support conversation gets when it is reviewed against a quality scorecard, shown as a percentage. Each criterion, such as accuracy, tone and resolution, is scored and weighted, and the scores roll up to agents and teams.- Also called
- QA rating, quality score, internal quality score (IQS)
- Scale
- Usually 0 to 100%
- Built from
- A weighted scorecard of quality criteria
- Coverage
- A sample of conversations, or all of them
QA stands for quality assurance. In customer service, a QA score is how a team turns "was that a good reply?" into a number it can track: for each person, each request type and the operation as a whole. The number is only as good as the scorecard behind it and the share of conversations it covers.
How a QA score is calculated
Each criterion on the scorecard has a weight. A reviewer scores the conversation on each one, and the weighted scores add up to a percentage.
| Criterion | Weight | Score |
|---|---|---|
| Correct answer | 35 | 35 |
| Problem solved | 25 | 25 |
| Followed policy | 20 | 10 |
| Checked the order and the customer | 10 | 10 |
| Tone and clarity | 10 | 8 |
| Total | 100 | 88% |
Here the agent answered a return question correctly and kindly but skipped the store's photo-of-damage step, losing half the policy points. Many scorecards also have auto-fail criteria: sharing another customer's details, or promising a refund the policy does not allow, sets the whole score to zero.
What goes on a QA scorecard
A scorecard lists what good looks like; a rubric explains how to score each line.
- Accuracy. Was every fact right: the order status, the policy, the product detail?
- Resolution. Did the customer get what they needed, or a next step with a date?
- Policy and process. Were the required checks done, in order?
- Communication. Clear, warm, the right length, in the store's voice.
- Efficiency. No needless back and forth or repeated questions.
Weighted scorecards put most of the points on accuracy and resolution, because a polite wrong answer is still wrong. Calibration sessions, where reviewers score the same conversations and compare, keep different reviewers' scores consistent.
In Aide, the same kind of inconsistency surfaces on its own: when teammates answer the same question differently, Aide asks which answer is policy as part of its nightly knowledge gaps.
Sampled QA vs 100% QA
Most teams score a small sample, often a few conversations per agent per week, because reviewing takes time. A sampled score is an estimate: a problem that repeats in the unreviewed conversations can hide for weeks.
100% QA means scoring every conversation, usually with AI doing the first pass against the same scorecard and people checking the failures and a share of the passes. It catches repeating problems sooner and removes the luck of the draw from individual scores. It is only trustworthy if the AI's scores are regularly checked against human reviewers.
QA scores for AI agents
Conversations handled by an AI agent need QA as much as human ones, scored on the same scorecard so the two can be compared. Two criteria matter more for AI. Accuracy, because an AI can state a wrong fact fluently. And actions, because a wrong cancellation or refund costs more than a clumsy sentence.
The practice this page rejects is QA as a policing exercise on a tiny sample, read once a month. A QA score exists to find what to fix: a help article, a policy, a step in a process.
With Aide, every reply is checked before it is sent: a wrong fact, a promise the policies don't back or a leftover placeholder gets one rewrite, and the reply goes to the team if that doesn't fix it, so some failures a scorecard would mark are stopped before a customer sees them.
Where Aide fits
Aide runs QA on its own replies, not on your team's. The Quality page reviews a random daily sample of Aide's drafts and replies against what the team actually sent, plus every one that stands out: a draft rewritten or discarded, a handoff, a knowledge gap. Each is acceptable or not, with no partial credit, and the headline is pass or fail against a bar of 90% acceptable over at least 100 reviews, so a handful of reviews can't flip the result. A teammate can mark any draft Good or Wrong, and their verdict overrides Aide's own review.
For reviewing the team's conversations, Aide gives reviewers the record to score from: each customer message shows the topics Aide detected, each reply shows the sources it used, and order actions offered, run or sent for approval are logged in the conversation's Action Trace. Aide's analytics show CSAT for Aide and for the team separately, and every number opens the conversations it counts, which makes it easy to pull the ones worth reviewing.