What's my plan?
Add-on AI agents - Advanced
AI agents with agentic AI are currently in an early access program (EAP). You can sign up for the EAP here.

Generative procedures are AI-generated procedures that help power the decision-making capabilities of AI agents with agentic AI. Instead of building scripted conversation flows in the dialogue builder, you can simply enter your business policies and let the AI agent generate a procedure to map out the best path to resolution.

This article contains the following topics:

  • About generative procedures
  • Creating generative procedures
  • Testing generative procedures
  • Best practices for creating generative procedures

Related articles:

  • About AI agents with agentic AI (EAP)
  • Getting started with AI agents with agentic AI (EAP)

About generative procedures

By leveraging agentic AI, generative procedures give AI agents the freedom to determine the best way to resolve a customer’s issue in line with your business policies. Generative procedures require less setup and maintenance from you than scripted dialogues do.

Each generative procedure is tied to a specific use case. When that use case is triggered during a conversation with a user, the AI agent follows the associated procedure to resolve the issue.

Also, generative procedures power multilingual AI agents. You can write a procedure once, in a single language, and it can serve all the languages your AI agent supports.

Creating generative procedures

Client admins and client editors can create a generative procedure from within a use case.

For helpful advice, see Best practices for creating generative procedures.

Note: Generative procedures don’t currently support rich-text formatting, such as buttons or carousels.

To create a generative procedure

  1. In AI agents - Advanced, use the AI agent drop-down field to select the AI agent you want to create a generative procedure for.
  2. In the left sidebar, select Content > Use cases.
  3. Click the use case you want to create a procedure for.
  4. Under Reply method, make sure Use procedure when use case is triggered is selected.
    Tip: For more information about this setting, see Configure whether a use case uses a dialogue or generative procedure (EAP).
  5. On the Procedure tab, click Create procedure.
  6. On the Create procedure screen, enter a free-text description of how the AI agent should solve a customer’s issue when this use case is triggered.

    List the steps required to solve this task as you would with a human agent.

    Click the plus icon (+) to insert actions, API integrations, and parameters.

  7. Click Create procedure.

    An AI-generated procedure map appears, showing you exactly what logic the AI agent will follow and what steps it will take during a conversation to solve a customer’s issue for this use case.

  8. If necessary, make adjustments to the text of the procedure and click Update draft.
    Repeat as necessary until the procedure map looks the way you expect.
    Note: You cannot edit the procedure map directly.
  9. Click Publish procedure.

Testing generative procedures

After you create or update a generative procedure, you can test it before making it live for your AI agent.

Testing a procedure always tests the most recently made changes. For example, if you draft updates to a live procedure and test it before publishing the updates, the draft is tested, not the published version.

To test a generative procedure
  1. In AI agents - Advanced, use the AI agent drop-down field to select the AI agent you want to test a generative procedure for.
  2. In the left sidebar, select Content > Use cases.
  3. Click the use case you want to test a procedure for.
  4. Click Edit procedure.
  5. In the top-right, click Test procedure.

  6. In the Session parameters dialog, do one of the following:
    • To test a conditional branch of the procedure, select a Parameter and enter a Value, then click Test.
    • To test the procedure as a whole, click Test without parameters.
      Tip: To remember your selection for next time, unselect Ask me every time.

  7. In the test widget that appears, test the procedure by sending messages to the AI agent.
Tip: To view test conversation data, filter the conversation logs by Display only test conversations.

Best practices for creating generative procedures

When creating generative procedures, follow these best practices:

Best practices Examples and explanation
Do’s Be direct and imperative. Write instructions as direct commands (for example, “Check if the customer…”). This helps the AI agent follow a clear flow of actions without confusion.

Set clear conditions and actions.

 Whenever there’s a conditional (“If… then…”), specify precisely what to do if the condition is met or not met. Clearly define under what circumstances to move to the next step.
Use consistent terminology.

Refer to the same items (for example, plans, billing types, policies) by the same name each time. This prevents the AI agent from mixing up terms or synonyms.
Provide contextual examples, if helpful. If a step involves a common scenario, provide an example of how to respond or what the result should look like. This can reduce ambiguity.

Anticipate edge cases.

 Include instructions for what happens if data is missing, incomplete, or contradictory. Decide in advance what the AI should do or ask in these cases.

Maintain a polite and empathetic tone, if applicable.

 When the AI agent interacts with customers, specify how to acknowledge or reassure them. Provide example phrases, such as, “I understand your concern.”
Update procedures regularly. As policies or plans change, update the procedures. Outdated procedures can lead to errors in AI agent responses.
Don’ts

Don’t use ambiguous language.

 Phrases such as “Handle it if you think it’s necessary” or “Take the best action” are too vague. Always specify exactly what “handling” or “best action” means in context.
Don’t mix multiple instructions in one step. If a step contains more than one major action, break it into separate steps. This keeps the procedure straightforward.
Don’t assume common knowledge. Avoid references to policies or information not included in the procedure. If it’s important, include or summarize it within the text.

Don’t leave data validation as an afterthought.

If certain conditions or data points are essential for a decision, specify the validation or check in the instructions.

For example, “Verify that the requested plan is valid before proceeding.”

Don’t overcomplicate the workflow.

Keep the steps concise and focused.

Overly complex branching can confuse the AI agent or lead to error.
Powered by Zendesk