Prep work: Identifying automations

This article assumes that you already have an idea of what workflow you would like to build into an automation. Please reach out to your point of contact on Assembled’s AI Deployments team if you need additional resources on how to identify good candidates for workflow automations!

Step 0: Navigate to the Workflows page

Within Assembled, use the left navigation bar to click to the AI support tab and select Workflows. This page is where you will view and manage all the automations you've already built, and their respective activation statuses.

To create a new automation, click on the blue Create workflow button on the top right of the page. You will be redirected to the automation workflow builder page.

Step 1: Create the workflow

When you click the Create workflow button, you'll be taken directly to the workflow builder interface. This is where you'll define your workflow's selection criteria and build out the action steps.

The workflow builder has a visual interface showing workflow steps connected by lines, with a plus (+) button to add new steps. At the top of the screen, you'll see options for:

• Undo/Redo: Reverse or reapply changes
• Preview: Test your workflow with sample tickets
• Manage activation: Configure how the workflow is deployed
• Save changes: Save your current workflow configuration

Step 1A: Define the trigger criteria

The first step in creating a workflow is to configure the trigger. When you click on the trigger step labeled "Start of workflow”, a panel opens on the right side where you can define your workflow's selection criteria.

When setting up your trigger, you’ll specify the:

• Workflow title: Provide a descriptive name that indicates what this automation aims to do. This title will also be used to reference this automation from the Workflows overview page.
• Selection filters:
  • Case type: Choose which type of interaction (any interaction, initial interactions, follow-ups, or webhook event) the automation will run on
  • (Optional) Case filters: Add specific rules to filter cases based on ticket properties (e.g., brand, status, or ticket tag)
  • You can check "Run this workflow whenever all case filters are met, ignoring any selection criteria" if you want the workflow to run based solely on the case type and filters specified
• Selection criteria:
  • Selection criteria instruction: Describe the primary goal or intent of this workflow and what you want it to achieve for your customers
  • (Optional, but strongly encouraged) Types of cases to include: Clearly describe which types of cases this workflow should handle, with examples
  • (Optional) Types of cases to exclude: Specify which cases should be excluded from this workflow
  • Here are some tips when defining your selection criteria:
    • Explain to Cal like you would to a coworker or as you would while onboarding a new agent
    • Provide example topics or intents of what types of content should be selected in this path (and what types should not!)
    • Feel free to use bullet points to structure your description!

Once you've completed the trigger configuration, click ‘Update step’ to return to the main workflow builder view. Always remember to click Save changes after making any edits to the automation.

Step 1B: Build out the steps

Once your trigger criteria is complete, you can start building your workflow by adding steps. Click the "+" button that appears after the trigger to add a new step. When you click the "+" button and Add step, a panel appears showing different categories of steps you can configure.

We currently support the following steps:

Condition Steps

Condition

This step allows you to create branching logic in your workflow based on evaluated conditions:

• Create different paths based on variable values (from analyze case, lookups, etc.)
• Set up branch names to clearly identify each path
• Define conditions using operators like equals, contains, does not equal, etc.
• Add multiple conditions to create complex routing logic

Example use case: Route cases with negative sentiment to one path and positive/neutral sentiment to another path.

Lookup Steps

Analyze case

This step extracts information from the customer's message as variables:

• Create variables with specific names that can be referenced later in the workflow
• Configure variable types (Text, Boolean, etc.)
• Provide clear instructions for Cal on how to analyze the content
• Extract sentiment, categorize requests, identify specific information, etc.

Example use case: Analyze sentiment as "positive," "neutral," or "negative" to route frustrated customers to tailored replies or specific workflow paths.

Zendesk data lookup

This step imports data from Zendesk as variables into your workflow:

• Pull information about the ticket, requester, or organization
• Access Zendesk ticket fields, custom fields, tags, etc.
• Reference these variables in conditions or reply steps

Example use case: Pull the customer's organization information to personalize responses or route to specialized teams.

Google sheets CSV lookup

This step allows you to access data from Google Sheets:

• Connect to spreadsheets shared with your Assembled account
• Look up information based on ticket properties
• Import data as variables to use in your workflow

The sheet must follow these requirements:

• First row must contain column headers
• All rows must have the same number of columns
• Data should be arranged in a tabular format
• No merged cells or complex formatting
• Sheet must be shared with Cal's email address

Custom API lookup

This step retrieves data from external systems through API calls:

• Map API response data to workflow variables
• Create branches based on API response status codes
• Handle success and error cases differently

This is particularly useful for retrieving customer data from CRMs, product information from catalogs, or any external system data needed to process the ticket effectively. Note: To configure your API endpoints, go to Setup in the left-hand navigation, then select Integrations > API Endpoints.

Reply Steps

Reply

This step allows Cal to compose and send customer-facing responses:

• Choose between Cal-generated responses or fixed templates
• Provide detailed instructions for response composition
• Reference variables from earlier steps to personalize content
• Configure knowledge sources for Cal to reference
• Specify macro(s) for Cal to reference

For Cal-generated responses, you can:

• Describe the tone, style, and content you want
• Provide example responses or templates for reference
• Specify what information should be included or excluded

Collect information

This step enables Cal to interact with customers to gather specific details:

• State clear questions for Cal to ask to obtain needed information
• Define variables to store the collected information
• Provide instructions on how to phrase follow-up questions
• Create a conversational flow for information gathering

Intelligent information gathering:

• Collects multiple variables defined in a single step
• Asks for each variable one at a time in a natural conversation
• Automatically skips questions for variables already answered earlier in the chat
• Creates a smooth experience while ensuring all needed information is gathered

This is particularly useful for gathering customer details - like order numbers, account info, or preferences - that can’t be retrieved automatically, in a way that feels natural and conversational.

Action Steps

Zendesk update ticket

This step modifies ticket fields and statuses in Zendesk without sending a customer response:

• Set ticket status (New, Open, Pending, Solved, etc.)
• Add or remove tags
• Assign to specific groups or agents
• Update custom fields
• Change ticket form type

This is useful for routing tickets to specialized teams, updating ticket status after automated responses, or adding tags for reporting and analytics purposes.

Zendesk leave internal note

This step adds private comments visible only to your support team:

• Compose notes with case analysis
• Include variables from previous steps
• Provide instructions to team members
• Document actions taken by the automation

Example use case: After analyzing a complex billing issue that requires manual review, Cal might add an internal note explaining what was found, what API data was retrieved, and recommendations for how the agent should proceed.

Custom API action

This step triggers actions in external systems:

• Configure API endpoints for POST, PUT, PATCH requests
• Set up request headers and body
• Use variables from earlier steps in the request
• Handle different response codes appropriately

Example use case: After collecting confirmation from a customer about canceling an order, trigger an API call to your order management system to process the cancellation, then respond to the customer differently based on whether the API call was successful.

Configuring each step

When you add a step, a configuration panel opens on the right where you can:

1. Input the step name (optional)
2. Configure the specific parameters for that step type
3. Click "Update step" to save your step

As you build more complex workflows, you can create branching paths with conditions and connect multiple steps together to handle different scenarios.

Best practices for building workflows

• Start simple: Begin with a basic flow and add complexity as needed
• Use descriptive names: Give clear names to steps and variables
• Test thoroughly: Use the preview or chat testing function after adding new steps
• Create modular workflows: Build separate workflows for distinct functions rather than one complex workflow
• Review variable usage: Ensure variables are properly defined before being referenced
• Consider edge cases: Create branches to handle unexpected input
• Save your work: Always remember to click Save changes after creating or editing a step

Step 2: Preview and test the workflow

Once you've built your workflow, you should test it using the Preview function before activating it. This allows you to see exactly how your workflow will behave with real or simulated tickets.

Using the Preview panel

Click the Preview button at the top of the workflow builder to open the preview panel. From there, you can either:

• Zendesk ticket ID: Paste the ticket ID of an existing Zendesk ticket you would like to test, and click Run preview to view the output.

• Case plain text (email or chat): Select the channel, enter the ticket subject and/or content, and click Run preview to view the output.

Note: API execute actions won’t run in preview mode. To test these, you’ll need to manually select the output variables and provide sample values in the preview workflow panel.

Understanding Preview results

When you run a preview, you'll see a detailed breakdown of how your workflow processes the ticket:

• Customer Case Details: Ticket content and timestamp
• Path Selection: Indicates which workflow was selected and why
  • Example: "The ticket references GCal, which matches the criteria for the resolution path designed to respond to tickets about the Google Calendar integration.”
• Step-by-Step Execution: Each step in your workflow is displayed with:
  • Step type (Lookup, Reply, Action, etc.)
  • Results of the step execution
  • Green checkmarks indicating steps that executed successfully
• Analyze Case Results: Any variables extracted from the ticket
  • Example: sentiment: neutral
• Condition Results: Shows which branch was followed and why
  • Example: "Condition: Positive/neutral sentiment" (selected); "Condition: Negative sentiment" (not selected)
• Reply Preview: For reply steps, you'll see:
  • Cal's analysis of the case
  • The composed reply that would be sent to the customer

Step 3: Activate the workflow

Once testing is complete, you'll need to activate your workflow. Click the Manage activation button at the top of the workflow builder to configure activation settings.

To allow for maximum comfort and control, we've provided multiple levels of activation that let you gradually increase automation as your confidence in the workflow grows.

Activation settings

The Manage activation settings let you control how Cal handles interactions for each workflow across different channels:

• Email agent:
  • Inactive: No automation for email interactions matching this workflow
  • Agent review: For matching tickets, Cal executes the whole workflow but any action steps (e.g. send reply, update Zendesk ticket, custom API) are left unexecuted for agents to review in the Copilot side panel before executing
  • Automated: Cal handles email interactions matching this workflow without agent intervention. You also have the option to configure:
    • Autosend percentage: Set what percentage (1-100%) of matching tickets Cal should process automatically
    • Apply tag: Add a specific tag to tickets handled by this workflow for tracking purposes
• Chat agent:
  • Inactive: No automation for chat interactions matching this workflow
  • Automated: Cal handles chat interactions matching this workflow without agent intervention
• Voice agent:
  • Inactive: No automation for voice interactions matching this workflow
  • Automated: Cal handles voice interactions matching this workflow without agent intervention

After configuring your activation settings, click "Update activation" to apply the changes.

Activation recommendations

Here are our recommendations for when to use which level of activation:

• Lightweight testing completed:
  • You’ve done a little bit of lightweight testing (e.g. tested 5 individual cases) and you want to start running the automation on incoming live cases.
  • Recommendation: Agent review
• Heavier testing completed or you’ve reviewed many live responses:
  • You’ve bulk tested a significant number and range of cases (e.g. 20+ cases) or you’ve been on Agent review for a while and have been monitoring responses and feel satisfactory about the output.
  • Recommendation: Automated with up to 50% autosend percentage
• Full confidence in the automation:
  • You’ve bulk tested or monitored live responses for a large number and range of cases (e.g. 40+ cases or monitored live responses consistently for 1-2 weeks) and feel satisfactory with the output.
  • Recommendation: Automated with up to 100% autosend percentage

Step 4: Review live responses

Once activated, the automation will run on any matching incoming cases. You can track its performance and see which cases it has run on in the QA review page under the Quality tab.

This page is useful for monitoring the execution and quality of this automation on actual incoming cases and keep track of which cases this automation has run on since being activated.