StyloDocs

Creating a Workflow

Step-by-step guide to building your first Response Workflow.

View as Markdown

This guide walks through creating a Response Workflow from scratch. By the end, you'll have a working workflow that generates contextual responses for a specific type of customer interaction.

Before you start, you need an admin role. Connecting any integrations and knowledge bases the workflow will use first (steps 4–5) makes setup smoother, but you can add them later.

Build your workflow

Create the workflow

Navigate to Response Workflows in the Stylo dashboard and click New Workflow.

Give your workflow a name that describes the situation it handles — for example, "Order Refund Request" or "Shipping Delay Response." The name is visible to agents when they browse available workflows.

Write your response content

This is where you decide how the AI generates responses. Choose a mode:

Write strategy instructions that tell the AI how to handle this type of interaction. Think of these as the playbook you'd give a new agent on their first day. See Strategy Instructions for detailed guidance.

Use the visual template editor to design the exact structure of the response. Add dynamic placeholders to pull in data:

  • Ticket data — customer name, email, ticket subject, status, priority
  • Order data — order number, total, fulfillment status, tracking number (requires Shopify or WooCommerce connection)
  • AI blocks — sections where the AI generates text based on a prompt you provide

Build a template and write strategy instructions. The template controls structure; the strategy instructions make the AI blocks smarter. See Response Modes for when to use each approach.

Configure "when to use"

Write a description of when this workflow should fire. Stylo uses it to automatically match workflows to incoming tickets. Be specific about the scenario:

  • Good: "Use when a customer is asking about the status of an order they've already placed, including questions about shipping, tracking, or delivery timeline."
  • Too vague: "Use for order questions."
  • Too narrow: "Use when a customer says 'where is my order' exactly."

The "when to use" description is evaluated by AI, so write it in natural language. Describe the intent, not the keywords.

Connect your tools

If your workflow needs external data (like order details from Shopify), select the tools it should use. The workflow automatically fetches the relevant data before generating a response.

Available tools depend on which integrations you've connected in your Stylo settings.

Select knowledge bases

Choose which knowledge bases the AI should search when generating responses. This grounds responses in your actual documentation — return policies, FAQs, product guides.

If you don't select specific knowledge bases, the AI searches all available knowledge bases for your organization.

Set up escalation rules

Define conditions where the workflow should not respond and instead flag the ticket for human review. See Escalation Rules for details. Common escalation rules:

  • Order value above a threshold
  • Disputed or fraudulent orders
  • Specific product categories that require specialist knowledge

Configure actions

Choose what happens after a response is sent:

  • Set ticket status — automatically change the ticket to "pending," "solved," etc.
  • Add tags — tag the ticket for reporting and routing (e.g., "auto-responded," "refund-handled")

Test and publish

Once you're satisfied with the workflow configuration:

  1. Test it — execute the workflow against sample tickets to verify it generates appropriate responses.
  2. Publish — make the workflow available for use.

Publishing creates an immutable snapshot of the workflow. You can continue editing the draft without affecting the published version. When you're ready to update the live version, publish again.

Expected result: the workflow is live and selectable by agents (or runs automatically, depending on how you configure it next).

Next steps

After creating your workflow, configure how it's used:

Or leave it as manual — agents select and execute it when they need it.

On this page