Evaluating the performance of AI agents using Zendesk QA

What's my plan?

Zendesk QA can help you evaluate how well your AI agents perform in conversations with your customers. You can then analyze the results and use this information to update your AI agents and customer service workflows as necessary.

This article contains the following topics:

• Understanding AI agents
• Manually evaluating AI agent conversations
• Automatically evaluating AI agent conversations

Related articles:

• Manually identifying AI agents in Zendesk QA
• Using the BotQA dashboard to understand bot escalations and performance

Understanding AI agents

Zendesk QA automatically detects the following types of AI agents:

• Conversation bots
• Ultimate messaging bots
• Sunshine Conversations bots

Tip: You can also manually report other users as AI agents so they can be reviewed using the correct resources. See Manually identifying AI agents in Zendesk QA

To view the bots Zendesk QA has detected

1. In Zendesk QA, click your profile icon in the bottom-left corner.
2. Select Users, bots, and workspaces.
3. Select Bots.

The list of bots appears, including the following columns:

• Bot name: The name of the bot.
• Last chat: When the last conversation with the bot took place.
• Reviewable: If the bot is included in reviews.

Manually evaluating AI agent conversations

You can use Zendesk QA to evaluate the performance of your bots across various categories just like you can for human agents.

To do so, you must set up a scorecard for the categories you want to evaluate the bot on.

To manually evaluate a bot’s performance

1. Click the Conversations icon () in the sidebar.
2. Select an existing, or create a new filter (public or private) to identify the bot conversations that you want to review. For example, you might use any of the following filter conditions:
   • Participant | is | <name of your bot>
   • Bot | is | <name of your bot>
   • Bot type | is | <workflow or generative>
   • Bot reply count | more than | 0

     Alternatively, use a Spotlight filter to find bot conversations.

3. Select the conversation you want to review.
4. In the Review this conversation panel, select the bot you want to review in the Reviewee field, then select the Scorecard.
5. Rate the bot’s performance for each category. See rating scale.

   

6. (Optional) In the free-text field, enter comments about the bot’s performance.
7. Click Submit.

Tip: Use the Reviews dashboard to analyze the results of a bot’s performance evaluation.

Automatically evaluating AI agent conversations

If you’ve set up AutoScoring, your bots can be automatically evaluated for the following AutoQA categories:

• Greeting
• Empathy
• Spelling and grammar
• Closing
• Solution offered
• Tone
• Readability
• Comprehension

Note: Automatic evaluations are present only for conversations that were created after AutoScoring was turned on for each relevant AutoQA category.

You can view the evaluation results in the Review this conversation panel in a conversation or in the Reviews section of an assignment.

Automatically scored ratings are marked with a hologram ( ) icon.