Want help applying this guide?Ask the DripTell team
A team lead opens a quality review and sees that an agent scored poorly on clarity. The easy response is to book a meeting and say, “Be clearer next time.” The agent leaves knowing the score but not what to do differently.
Useful customer service QA coaching starts later than most teams think. First confirm that the finding is a repeated, agent-controllable behavior rather than a broken policy, missing permission, weak knowledge source, or unusual case. Then describe one observable change, practise it in the real situation, and recheck comparable work. A score points to a question. It is not the coaching plan.
A score is only the start
A customer service QA scorecard should identify material failures and show the evidence behind a rating. It still cannot explain the cause on its own. A low accuracy score might come from an agent ignoring a clear rule. It might also come from a knowledge article that contradicts the refund policy.
- Pattern confirmedComparable work shows more than one isolated miss.
- Rule is stableReviewers and policy owners agree on the expected action.
- Evidence is completeThe full conversation, policy date, access, and outcome are visible.
- Behavior is controllableThe agent could perform the action with the tools and authority available.
- Customer impact is clearThe missed behavior changed safety, accuracy, effort, or the next step.
Microsoft’s current quality evaluation guidance describes evaluation outputs as scores, summaries, recommended actions, and follow-up opportunities. It also warns that the feature is not intended to make employment decisions. That is a useful boundary. Treat QA as evidence for improvement, not an automatic verdict about a person.
Look for a pattern across comparable work. Use a sampling plan that keeps ordinary work separate from targeted risk cases. One difficult conversation may justify a check-in, but it rarely proves a stable behavior.
Check whether coaching is the right response
Before writing an individual plan, reconstruct what the agent could actually see and do. Was the correct policy available on that date? Did the agent have the required account access? Was ownership clear after a transfer? Did the interface show the promised action as complete?
If several people miss the same step, investigate the system first. Repair the knowledge, routing, macro, permission, or policy. The roles and permissions guide can help separate an access problem from a judgment problem. Coaching someone to work around a missing permission only hides the defect.
Calibration matters too. If reviewers disagree about the criterion, resolve that before asking an agent to change. The reviewer calibration method shows how to preserve independent scores, diagnose disagreement, and test a clarification on fresh work.
Describe one observable behavior
“Improve empathy” is not coachable. “Before explaining the refund rule, name the customer’s specific inconvenience in your own words” is observable. So is “state the owner and expected next step before closing the conversation.”
Write the finding as situation, behavior, and customer impact. For example, suppose an agent closes five delivery complaints after saying the warehouse will investigate. The missing behavior is not “ownership.” It is failing to name who will update the customer and when. The customer has to ask again, and the queue receives avoidable follow-up.
Choose one behavior for the cycle. If the review found accuracy, tone, and closure problems, start with the issue carrying the greatest customer or compliance risk. Trying to coach three habits at once makes the next review impossible to interpret.
| Finding cause | Primary owner | Corrective action | Evidence of closure |
|---|---|---|---|
| Clear rule missed by one person | Team coach | Practise one observable behavior | Comparable conversations show the behavior |
| Knowledge is outdated | Knowledge owner | Correct the source and notify the team | New answer matches the approved policy |
| Policy is ambiguous | Policy owner | Decide the rule and effective date | Reviewers apply the same interpretation |
| Access or workflow is broken | System owner | Restore permission or repair the flow | Agent can complete the action without a workaround |
| Causes are mixed | Coach and system owner | Fix the blocker before practising | Recheck separates system and behavior results |
Practise the real situation
Practice should resemble the work. If the gap appears in refund chats, rehearse refund chats. Ask the agent to rewrite one weak response, annotate one strong example, and role-play one realistic exception. The coach should model the target behavior, then let the agent try it without reading a script.
Use customer language, policy context, and the tools the agent will have live. A generic communication course is unlikely to fix a missing handoff sentence in a billing queue. Keep the meeting private and developmental. Ask what made the correct action hard. The answer may expose another system issue.
Record the behavior, practice method, owner, and review date. Inbox reporting can locate the right conversation cohort, but the coach still needs to inspect the actual evidence.
Recheck comparable work
Set the recheck when the plan is agreed. Review fresh conversations from the same issue type, channel, and policy period. Do not make the result look better by mixing in easier work.
Measure the target behavior as present, absent, or not applicable before looking at the total QA score. Then check the customer outcome. Did the clearer next step reduce repeat questions? Did the agent complete the action safely? A score can improve while the practical failure remains.
If the behavior does not change, do not simply make the warning louder. Check whether practice matched the job, the standard was understood, the agent had enough opportunities, and the system blocker was actually removed.
Keep development separate from punishment
Quality evidence may uncover serious conduct or compliance concerns, but everyday coaching should not feel like a disguised disciplinary process. Keep the development record focused on evidence, behavior, practice, support, and recheck. Apply formal employment procedures separately and under the organization’s own law and policy.
A shared customer inbox can preserve conversation history, ownership, notes, and status for review. It does not decide whether coaching is fair. That judgment still belongs to accountable people using a stable standard and complete context.
The aim is not to move an agent from 78 to 85. It is to make a customer-facing behavior more reliable and to prove that the operation no longer creates the same failure.
Frequently Asked Questions
How many QA findings should one coaching plan include
Usually one primary behavior. A plan can mention related evidence, but the practice and recheck should focus on the highest-impact change so both the agent and coach can tell what improved.
Should a manager coach after one failed conversation
A serious safety, privacy, or compliance issue may need immediate action. For ordinary quality gaps, first check comparable conversations and the surrounding system. One case can start a conversation without proving a repeated habit.
How soon should QA coaching be reviewed
Review soon enough that the agent can remember and practise the situation, but only after enough comparable work exists. Agree the date in advance and keep the issue type, channel, and rule version consistent.
DripTell Editorial
Practical guidance reviewed by the DripTell product and customer workflow team.
See how DripTell checks product claims, uses primary sources and handles corrections.
Editorial and source policy