How to Run a Bulk Assessment from the Chats Table or a List of IDs?

A bulk AI QA assessment evaluates many closed chats at once. The checklist of every chat is taken by its topic from the auto-assessment settings, and before the start you see how many chats go to which checklist and which are not assessed. Chats can be selected in the operator panel chats table, loaded as a list of IDs or opened from a protected link a colleague shared.

When It Will Be Useful

• You need to assess chats found with the chats table filters.
• You already have a list of chat IDs — in a CSV file or as text.
• Every topic should be assessed with its own checklist, and topics set to "Do not assess" should be left out.

What's Important to Know

• When this window is used. The "Bulk AI QA assessment" window opens when the AI QA auto-assessment settings are enabled on your instance (the main "For all bots" settings are created). Until then "Evaluate dialogs" opens the usual "Bulk AI evaluation" window.
• Checklist by topic. For every chat the checklist comes from the auto-assessment settings of its bot (the bot's own settings or "For all bots"): first the row of the topic, then the row of its family, then the "All other topics" row. Chats without a topic go by the "No topic" row. You can change a row for this run — the settings themselves do not change.
• "Do not assess". Chats in "Do not assess" topics are not assessed in bulk. To assess such a chat, switch the row to a checklist or assess the chat manually from its card.
• Repeat assessment. A chat already assessed with the same checklist is assessed again like any other — there are no separate marks or checkboxes. Every run creates a new assessment.
• Skip rules. The "Apply skip rules" switch appears when at least one bot of the selected chats has "Apply skip rules to bulk assessment" on in its auto-assessment settings. Only the rules enabled in the bot settings apply. The skip reason is shown in the run history, not in the dialog card.
• What is never assessed. Open chats and chats where no operator replied are rejected before the start with the reason "Chat is open or no operator replied".
• Limit. One run holds up to 30 chats; an administrator can raise the limit to 100. The limit counts every chat that can be assessed — including those not assessed by topic or skipped by rules. For a larger selection, create several runs.
• File or IDs. A CSV with a chat_room_id column, or chat IDs separated by commas, spaces or new lines. Up to 1,000 IDs, file up to 5 MB. XLSX is not supported — save the spreadsheet as CSV. Duplicates are removed automatically.
• Selection link. A checked chat list can be passed to a colleague as a protected link — see Share a chat selection with a link.

Before You Start

• Your operator panel role has the right to launch quality assessments, and the chats table is available to you.
• The AI QA auto-assessment settings are enabled on the instance and the main "For all bots" settings exist.
• The Quality Assurance module has at least one active checklist.

Step-by-Step Instructions

1. Choose the chats

Chats from the table:

1. Open the Operator Panel and its chats table.
2. Filter the chats with the table filters and select the ones you need.
3. Click "Actions" → "Evaluate dialogs". If more chats are selected than the limit allows, the item is disabled — reduce the selection.
4. The "Bulk AI QA assessment" window opens with the "Selected in the table" source.

File or IDs:

1. In the chats table click "Actions" → "Bulk assessment". The item is available even with no chats selected.
2. In the window choose the "File or IDs" source.
3. Upload a CSV file or paste the IDs into the field.
4. Click "Check".

2. Review the check result

1. The window shows how many chats can be assessed and how many are rejected.
2. The rejected table lists the chat ID and the reason: "Invalid ID", "Chat not found", "Duplicate in the list" or "Chat is open or no operator replied".
3. If there are more chats than the limit, the window says "{count} chats selected — more than the limit of {limit}. Reduce the selection." and the run cannot be started.

3. Check the checklists by topic

1. The "Checklist by topic for this run" block shows a row for every topic of the selected chats and a "No topic" row: the topic, the number of chats and the checklist.
2. The caption "via topic “…”" means the checklist comes from the row of the topic's family; "via “All other topics”" means the topic has no row of its own.
3. If bots plan the same topic with different checklists, the topic is shown in a separate row for every bot.
4. To change the checklist for this run, pick another one in the "Checklist" column or "Do not assess". A changed row is marked; to undo the change, click "Revert".

4. Check the skip rules

1. If the window has the "Apply skip rules" switch, it is on by default.
2. Turn it off if this run does not need skip rules. Topics set to "Do not assess" are not assessed regardless of the switch.

5. Start the assessment

1. Check the summary, for example: "3 → Demo main · 2 not assessed". Chats assessed before are counted in the summary like any other.
2. Click "Start".

What Happens Next

• The window closes, the table selection is cleared, and the page of this run opens in the Quality Assurance module.
• Chats are assessed in the background. When all are processed, a completion notification appears. If some chats were skipped, it contains the line "Skipped by rules: … Not assessed by topic: …".
• All runs are listed in the history — see View the bulk assessment history.

How to Make Sure Everything Worked

• On the run page the source is "From the chats table", "File or IDs" or "Link".
• Chats of different topics are assessed with their own checklists; chats of "Do not assess" topics have the status "Not assessed by topic".
• The number of chats in the run matches the summary in the window.

If the Selection Is Larger than the Limit

1. Split the selection into parts within the instance limit. Count every chat that can be assessed, including those not assessed by topic.
2. Create a separate run for every part.
3. Add a comment with the part number to every run, for example "Period 01–07.08, part 1/3".

Possible Messages

Related Materials

• View the bulk assessment history and add a comment
• Share a chat selection with a link
• View assessment details
• View assessments list