# Bot quality: what the bot resolves on its own and what it doesn't know

> Share of questions resolved by the bot, reasons for handoffs to people, staff response time, customer ratings, knowledge gaps and AI conversation usage under your plan.

Источник: https://jecheck.io/en/docs/bots/quality

**Customers → Bot quality** shows how well the bots are working and suggests what to improve. The period is 7, 30 or 90 days; if there are several bots, you can pick one.

![Bot quality: what the bot resolves on its own and what it doesn't know](/docs-media/en/bots/quality.jpg)

## Metrics

| Tile | What it means |
|---|---|
| **Conversations** | how many conversations started in the period; below the number — how many are with employees right now |
| **Resolved by the bot** | the share of closed conversations where no employee was needed |
| **Handed to people** | the share of conversations where an employee stepped in |
| **Staff reply** | the median time from handoff to the employee's first reply |
| **Customer rating** | the average rating from 1 to 5 and the share of satisfied customers (4–5) |

Below: a chart of conversations by day, **Why the bot called people** (the most frequent handoff reasons), a breakdown by **Channels** and **Bot actions** — how many tasks, requests and queries it performed and with what errors.

## What the bot doesn't know

A list of questions for which the bot found no reliable answer in its knowledge. Similar questions are grouped, with how many times they were asked and links to the conversations.

How to close a gap:

- **Quick reply** — write the answer right away; it's added to the bot's draft. Publish the bot so that customers get it.
- **Instruction** — open [Instructions](/docs/instructions/create-instruction) and describe the topic in detail in a folder open to customers.

> [!TIP]
> Review this list once a week: 5–10 new quick replies noticeably raise the share of questions resolved by the bot.

## Plan usage

At the bottom — **AI conversations this month**: how many conversations have already been counted and how many the plan includes. A conversation is counted once — when the bot first replies in it. When the allowance runs out, new conversations go straight to employees, and the bot finishes the ones already in progress. Learn more in [Plans and billing](/docs/account/pricing).

## Who can see it

The **Chatbots: quality and usage** permission — see [Roles and permissions](/docs/account/access-roles).

## Frequently asked questions

### Why don't "Resolved by the bot" and "Handed to people" add up to 100%?

"Resolved by the bot" is calculated from closed conversations — while a conversation is open, it's unknown whether a person will be needed. "Handed to people" is calculated from all conversations in the period, including open ones.

### Do test conversations from the sandbox count in the statistics?

No. The sandbox isn't counted in the metrics or in the plan allowance.