# Reply quality scoring and team workload

Source: https://frosvo.com/docs/reply-quality-and-workload
Updated: 2026-11-05

See who has how many open chats, and let the AI review your team's replies for tone and accuracy so owners can coach with real examples.

Numbers about speed tell you if replies are fast. They do not tell you if they are kind or correct. Reply quality scoring adds that, and the same card shows how much each person has on their plate right now. This page is for owners who coach the team and for agents who want feedback.

## Before you start

- You need the Reports feature and a plan with the AI assistant. Free plans do not get AI reviews.
- Each review is a short AI call, so turn it on only for the websites you want reviewed. See [AI budget and usage](/docs/ai-budget-and-usage).

## How do I turn it on?

1. Open **Settings → Widget** and open the website.
2. Tick **Let the AI review team replies for tone and accuracy (uses your AI budget)**.
3. Save the website settings.

From then on replies on that website are reviewed in the background. It does not look at old replies.

## What does the AI judge?

For each reply it gets the recent conversation and, if you have a knowledge base, the passages that match the customer's question. It scores two things from 1 to 5 and adds one sentence of advice:

- **Tone:** 5 is warm, polite and professional. 1 is rude or dismissive.
- **Accuracy:** 5 is correct and complete and agrees with your own notes. 1 is wrong, made up, contradicts your notes or ignores the question.

Card numbers, passwords and keys are masked before the AI reads anything. It is a guide for a conversation with your teammate, not a verdict, so read the conversation before acting on a score.

## Where do I read the results?

Open **Reports**. The card **Team workload and reply quality** shows, for each person:

- **Open** and **Waiting:** conversations assigned to them right now, and how many of those have something new they have not read.
- **Reviewed:** how many of their replies in the chosen dates were scored.
- **Tone** and **Accuracy:** their averages.

Below it, **Replies worth a look** lists the replies that scored 2 or lower on tone or accuracy, with the AI's sentence and a link to the conversation, so you can coach with a real example.

**Quality over time** shows each person's tone and accuracy averages by day, week or month, so you can see whether coaching is working. It appears once there are reviews in at least two steps.

Owners see everyone. Other team members see only their own row. The website, channel and agent filters narrow the card.

## Common problems

- **The card is empty.** Quality scoring is off for the websites, or no replies were reviewed yet. Reviews run in the background and wait five minutes after a reply.
- **A reply was not reviewed.** It may be shorter than about 25 characters, a private note, older than two days, or the AI budget was used up.
- **A score seems unfair.** The AI only sees the conversation and your knowledge base. If the answer was right for reasons it could not see, ignore the score.

## Related

See [Reports](/docs/reports), [roles and permissions](/docs/roles-and-permissions), and [satisfaction and NPS surveys](/docs/customer-satisfaction-surveys).

## Common questions

### Who can see the scores?

Owners see every teammate. Everyone else sees only their own row, so scores are for coaching and not a leaderboard.

### Does the AI send anything to customers?

No. It reads a reply after it was sent and writes a score and one sentence for your team. Customers never see it.

### Which replies are reviewed?

Replies a teammate sent more than five minutes ago, in the last two days, for websites where you turned it on. Private notes and very short replies such as "ok" are skipped.
