Quality score for AI and human replies
Quality score for AI and human replies
Every sent reply scored on an editable checklist, with per-agent views and the conversations that need attention.
Each reply TR-AI sent and each reply a person sent on a solved ticket gets a 0–100 score on accuracy, completeness, clarity, tone, policy and resolution. If the customer reopened the ticket, gave low satisfaction or replied that it did not help, the resolution part is capped.
- Edit the checklist: up to 8 criteria with weights.
- Sample rate: score every reply or a percentage; daily caps are 300 (Bronze) and 2,000 (Growth Pro).
- About one AI call per scored reply.
- Agent names are available for Zendesk; on other help desks human scores appear under one “Human agent” row.
Tips & tricks
- Start with the “Needs attention” list: flagged replies (promised refunds, wrong information) are the fastest wins.
- Compare AI vs human by criterion — if humans score higher on completeness, their past answers are good material for articles.
- Lower the sample rate if you only need trends; keep 100% while you are tuning.