What's my plan?
Add-on AI agents - Advanced

An advanced email AI agent automates repetitive support tasks so your staff can focus on solving complex issues.

For example, with CRM actions an email AI agent can perform the following tasks:
  • Add tags and internal notes to Zendesk Support tickets.
  • Create personalized messages based on user information.
  • Update ticket fields.
  • Merge tickets from the same requester.
  • Route tickets to agents.
If you’re using an expression-based email AI agent, it will require some training based on existing data. This article outlines the steps to get an expression-based email AI agent trained and ready to recognize each intent.

This article covers the following activities:

  • Step 1: Importing historical data
  • Step 2: Connecting the email AI agent
  • Step 3: Creating replies and conversation flows
  • Step 4: Testing the email AI agent
  • Step 5: Iterating on the email AI Agent
  • Step 6: Launching and maintaining the email AI agent

Step 1: Importing historical data

Use your existing data to help train the email AI agent to understand your customers.

  • If you already use Zendesk Support, use a dedicated tool to import conversations.
  • If you don't use Zendesk Support, import your data from another CRM or manually from a TSV file.
  • If you don't have historical data, learn about other sources of training data.

Step 2: Connecting the email AI agent

Your email AI agent needs its own Zendesk account. For details, see Connecting your advanced AI agent to Zendesk Support.

Step 3: Creating replies

Use the dialogue builder to compose the replies you would like to use for each intent.

Here are some guidelines for crafting replies:
  • Keep replies aligned with your brand identity, tone of voice, and customer persona.
  • Use markdown to style your replies.
  • Review the actions available for automation to plan how to replace manual customer support processes.

Step 4: Testing the AI agent

Once your email AI agent is connected to Zendesk and you have created your replies, you can begin testing it.

Here are some recommendations for testing:
  • Ask testers to take screenshots to accompany their feedback.
  • Don't deliberately confuse, provoke, or “troll” the email AI agent.
  • Ask the same question in multiple different ways. Gather feedback in one place.
  • While test conversations are recorded in conversation logs, having consolidated feedback will make it easier to iterate.
  • Ensure that actions and internal notes align with your established support processes.
  • If launching the email AI agent includes changes to established processes, use the testing phase to take screenshots for training your human agents.

Step 5: Iterating on the email AI agent

Once you’ve gathered feedback from the tests, make changes to fix and improve your AI agent. For example, use different content for a reply, or use actions to add tags that allow you reduce your backlog by merging tickets.

Step 6: Launching and maintaining the AI agent

After your email AI agent starts working in your team, monitor its performance by auditing conversation logs and reviewing the goals and KPIs set during project preparation. Continue to train the email AI agent and make adjustments to your replies as needed.

Powered by Zendesk