Role Guide: QA Reviewer
For Narqix specialists and leads who score support work and coach agents. After reading you will know how to sample, score fairly, write useful feedback, run calibration, and turn QA findings into better processes.
Track: Customer Support · Role: QA Reviewer · Time to complete: 25 min
What this role is
QA is how Narqix proves quality to clients, instead of just claiming it. A QA Reviewer reads, listens and scores real customer interactions against the QA Scorecard, finds what needs to improve, and coaches agents to get there.
Good QA is fair, consistent and useful. Fair: every agent is judged by the same rules. Consistent: two reviewers give the same score to the same ticket. Useful: every piece of feedback tells the agent exactly what to do differently.
What you own
| Area | Your job |
|---|---|
| Sampling | The right number and mix of interactions reviewed every week |
| Scoring | Each sample scored on the QA Scorecard with evidence |
| Auto-fails | Identified, reported the same day, and acted on |
| Feedback | Written feedback within 2 business days |
| Coaching | 15-minute sessions for every score below pass, within 3 business days |
| Calibration | Monthly session where reviewers score the same items and compare |
| Insights | Patterns sent to leads: training gaps, weak macros, process gaps |
| Reporting | QA scores per agent and per account in the weekly and monthly reports |
Sampling rules
| Sample | How many |
|---|---|
| Standard, per agent per week | 5 interactions (mix of email, chat and phone if the agent does all three) |
| New agent, first 30 days | 10 per week |
| New agent's first 20 customer-facing replies | All 20, reviewed before or right after sending |
| Every negative CSAT rating | All |
| Every escalated complaint | All |
| Agent on an improvement plan | 10 per week |
Pick standard samples at random. Do not choose only easy or only hard tickets.
How to score
- Read the whole interaction first, including internal notes, order history and the customer's past tickets. You cannot judge a reply without knowing what the agent could see.
- Check auto-fails first. If one applies, the score is zero. Stop and report it.
- Score each criterion using the definitions in the QA Scorecard. Use full, half or zero marks as defined. Do not invent your own scale.
- Write evidence for every lost point: quote the exact line.
- Write the better version of what should have been said or done.
- Note what was done well. Specific praise teaches as much as correction.
Writing feedback that works
| Weak feedback | Strong feedback |
|---|---|
| "Tone could be better." | "Line 2: 'As per our policy you must…' reads cold. Try: 'Our return window is 30 days, and here is what I can do…'" |
| "Missed a question." | "The customer asked about the gift receipt in the last line. Not answered. Add: 'I have added a gift receipt to the package.'" |
| "Good job." | "Your first line ('A birthday gift arriving late is the worst timing') showed real empathy and set up the rest of the reply." |
Coach's note: use the "one big thing" rule in coaching. Even if you found five issues, focus the session on the one that matters most. Agents fix one thing well; they forget five things quickly. The written scorecard still records everything.
The coaching session (15 minutes)
- Open with something they did well (1 minute).
- Show the one big thing: the exact line, why it matters to the customer (4 minutes).
- Practice: the agent rewrites the line or role-plays the moment (6 minutes).
- Agree the action and how you will check it next week (2 minutes).
- Record it: date, issue, action, follow up date (2 minutes).
If the agent's mistake was caused by an unclear process, a wrong macro or missing knowledge, the fix is the process, not the agent. Send it to the lead the same day.
Calibration
Once a month, every reviewer on the account scores the same 5 interactions alone, then the group compares. Where scores differ by more than 5 points, discuss until you agree what the rule means, then write that decision into the scorecard notes. Target: reviewers within 5 points of each other on 90% of items.
Auto-fail handling
An auto-fail is serious (see QA Scorecard). When you find one:
- Score zero and mark the auto-fail reason.
- If it is a security or data issue, report it to the Narqix lead immediately as an incident.
- Tell the agent the same day, privately and factually.
- If the customer was affected, agree with the lead how to fix it with the customer.
- Check the agent's other recent work for the same pattern.
How you are measured
| KPI | Target |
|---|---|
| Samples completed on time | 100% |
| Feedback delivered within 2 business days | 100% |
| Coaching within 3 business days for below-pass scores | 100% |
| Calibration variance | Within 5 points on 90% of items |
| Process fixes raised from QA findings | Tracked monthly |
Never do this: never change a score because an agent is a friend, is struggling, or argues. Fairness is the whole value of QA. If you are unsure, take it to calibration.
Never do this: never share an agent's QA results with other agents, or customer details from reviewed tickets outside the QA tools. QA records are confidential.
Common mistakes
| Mistake | Fix |
|---|---|
| Scoring without reading the history | Read everything the agent could see |
| Vague feedback | Quote the line; write the better version |
| Only negative feedback | One specific strength in every review |
| Blaming the agent for a broken macro | Fix the macro and tell the lead |
| Skipping calibration when busy | Calibration keeps scores trustworthy; never skip |
Check yourself
- How many interactions do you sample for an agent in their first 30 days?
- What is the first thing you check when scoring?
- Describe the five parts of a 15-minute coaching session.
- Two reviewers score the same ticket 92 and 81. What happens next?
© Narqix Academy