Quality scoring
Every reply gets a score. This module surfaces the worst ones with the why, so you can fix patterns instead of chasing individual conversations.
What it shows
A ranked list of the lowest-scoring replies in your window, regardless of who wrote them (human agent or Fin Agent). Each row shows the conversation, the reply, the score, the breakdown across the four criteria, and the reason flags.
The four criteria
- Relevance — does the reply address what the customer actually asked? Catches off-topic answers and template misfires.
- Accuracy — is the information correct against your docs and product? Catches stale pricing, deprecated steps, wrong limits.
- Tone— does it match your brand voice and the customer's emotion? Catches robotic replies on emotional tickets.
- Completeness — does it resolve the question or leave gaps? Catches partial answers that force a follow-up.
The overall score is a weighted average. Tone weighs less than Accuracy and Completeness by default. You can adjust the weights in Settings → Scoring policy.
Reason flags
Beyond the numerical breakdown, each low score comes with one or more human-readable flags:
stale-info— the reply quotes a value or step that no longer matches your current docs.off-topic— the reply answers a different question than the one asked.tone-mismatch— apologetic where assertive was needed, or vice versa.link-only— the reply just dropped a help center link without context.unresolved— the customer had to ask again. Strong predictor of negative CSAT.
Calibrating the AI
On every low-scoring row you can validate (yes, this was bad) or correct (no, the score is off) with one click. These feedback signals re-train the scoring weights for your workspace nightly. The reliability score on the Insights module reflects the result.
Tip
Workflows
Weekly QA
Filter to your team's replies for the last 7 days, sorted by lowest score. Skim the top 20. Validate the obvious cases, correct the AI on edge cases. Identify one repeating reason flag and brief the team in your weekly meeting.
Onboarding new agents
Filter to a new agent's replies in their first 30 days. The list of bad replies + reasons is the most concrete coaching material you can give them.
Fin Agent vs. humans
Toggle the Authorfilter between Fin Agent, Team, or both. Comparing scores between Fin and your team on similar topics is the single best way to decide where to extend Fin's coverage and where to keep humans in the loop.