Guide to Creating a Workflow

• 🎯 Define the workflow goal (e.g., qualify leads, reactivate contacts, support, collections).

• 🧭 List the necessary stages and which conditions will move a conversation forward.

────────────────────────────────────────

1) Go to the Pipelines section.

2) Click Edit (top right) and then Create Pipeline.

3) Fill in:

- Pipeline Name

- Description

- Pipeline Type (available options): Sales Qualification, Lead Reactivation, Satisfaction, Support, Post Sales, Collection, Orchestrator, or Internal Assistant

4) Click Save.

💡 Tip: use descriptive and consistent names.

────────────────────────────────────────

You can create stages in two ways.

📍 Option A: from Pipelines

1) Inside the pipeline, click Add stage.

2) Fill in Name, Description, and Type.

Available types:

- 🤖 Automated: the AI handles this stage automatically.

- 👤 Human Intervention: the stage is handled by a person (on the same channel as the AI).

- 🚨 Escalation (recommended in all pipelines): for conversations escalated due to SEVERE_ANGER (severe anger), HUMAN_TRANSFER (request to speak with an agent), PROMPT_INJECTION (attempt to manipulate the AI), or manually escalated.

- 🛑 Checkpoint: lets you select another pipeline for quick access (pipelines can connect to each other).

🧰 Option B: from Workforce > Stage Automations

1) Go to Workforce > Stage Automations.

2) Find the pipeline you created and select Create new stage (configured the same as in Option A).

⚠️ Important: always include an Escalation stage to ensure proper handling of sensitive cases.

────────────────────────────────────────

1) Go to Workforce and select the corresponding worker.

2) Open Stage Automations.

3) In the first stage you created, click “Handle this stage with [AI Name]”. You’ll see these sections:

🎯 a) Objective

Describe precisely what the AI must do in this stage: actions, questions, validations, and constraints.

Suggested template (copy and adapt):

Stage objective:

- Purpose: describe the main objective (e.g., qualify the lead with 3 questions).

- Actions: detail the tasks (e.g., request monthly budget, model of interest, province).

- Constraints: indicate limits (e.g., the monthly budget amount must be provided in USD).

- Tone and style: specify the tone (e.g., formal, concise, and empathetic).

🛠️ b) Tools

If you have a catalog, a Python function, or an MCP, select it so the AI can use it in this stage. In the Objective section, explicitly indicate when and how it should be used.

Example instruction within Objective:

Whenever the person asks about products, use the 'catalogo_productos' tool.

🔗 c) Next stages

Connect the current stage to the following ones via conditions.

- Connect stage when: specify the condition that allows progression (e.g., “when the human has mentioned their monthly budget, model of interest, and province”).

- Next stage: choose one from the dropdown. If it doesn’t exist yet, select Create new stage.

You can configure multiple outbound connections according to different conditions.

🧪 d) Quick stage test (without saving)

Before saving, you can test what you wrote in the stage directly on this screen in the section called “Test this stage’s behavior.” The test:

- Runs with the configuration you see on screen.

- Doesn’t persist changes and doesn’t require saving first.

- Lets you validate prompts, logic, and tool usage.

Suggested steps:

1) Select Test stage.

2) Enter a sample message and run the test.

3) Review the simulated conversation and tool usage.

4) Adjust Objective/conditions if needed and test again.

5) When satisfied, save the changes.

────────────────────────────────────────

• Verify that the Objective is met and that the AI asks/acts as expected.

• Check that Tools are invoked when appropriate.

• Confirm that conditions in Next stages route correctly.

• Use the Quick stage test to iterate before saving.

• You can test the full workflow from the “Try Now” menu section.

────────────────────────────────────────

• One intent per stage: keep each stage focused on a specific goal.

• Consistent naming: makes future maintenance easier.

• Clear messages: state constraints and success criteria explicitly for the AI.

────────────────────────────────────────

• Pipeline: the workflow container (overall goal and stages).

• Stage: a step within the pipeline (Automated, Human Intervention, Escalation, Checkpoint).