Quality, CSAT and coaching

Use this when the team needs a report card, not a feeling. Score real threads against a rubric, read CSAT and handle time from the records, and draft the one fix per rep.

What you need first

The thread sample with dates and reps, the CSAT and average handle time (AHT) records with source and period, the QA rubric or talk track in play, and the last review's fixes.

Build or confirm the rubric

Score on weighted dimensions the team agreed: acknowledgement speed, accuracy against docs and policy, resolution or a clear next step with a date and owner, tone and positive language, and logging. Calibrate before scoring — two reviewers score the same three threads and settle the differences — or the numbers are opinion. A rubric you cannot find gets the five dimensions above, marked INFERENCE, with a request for the official one.

Sample fairly and score with quotes

A fixed sample per rep per week — newest, oldest, one escalated, one refund-adjacent — never cherry-picked wins. Score each thread with the quoted line behind every mark. A thread you cannot read is not scored.

Read the numbers from the records

CSAT with its response rate and sample size — a 4.6 on nine responses is not a trend; AHT beside first-contact resolution, never alone, because faster is not better if the customer writes back; reopen rate as the check on "resolved". Trend against last period with the source and period on every number. Never fill a number from memory.

Coach one fix per rep

One quoted miss, the better line, why it is better, and a re-check date. Praise one quoted win in the same note. The team-wide pattern — the miss three reps share — becomes a macro, a knowledge-base fix, or a talk-track change, staged for the owner.

Output

The scored sample with quotes, the metric read with trends and sources, and the per-rep coaching drafts. Coaching notes go to the owner as drafts — never delivered to a rep or published as a score by you.

When the records are thin

Thin records mean fewer scored threads, not padded scores. Say which record would settle each UNKNOWN.