Workflow Editor

The workflow editor is where you build and configure Response Workflows. This page walks through every section of the editor, explains what each setting does, and provides guidance on how to configure it.

Opening the editor

From the Stylo dashboard, navigate to Response Workflows and either:

• Click New Workflow to create a new one
• Click an existing workflow to edit it

The editor opens with the workflow name at the top and two buttons in the header: Save Draft and Publish.

Editor layout

The editor is organized into five sections:

1. Essentials — name, description, enabled toggle, tags
2. Response Content — where you write strategy instructions and/or build a template
3. Activation & Context — when to use, trigger type, conditions, tools, knowledge bases
4. Guardrails — escalation rules and post-response actions
5. Automation — background processing mode and response delay

Below the main editor is the Test Panel, where you can run your workflow against real tickets without leaving the page.

Each section is a collapsible card. You can work through them in any order.

---

Essentials

Basic information about the workflow.

---

Response Content

This is the core of the workflow — where you define what the AI generates. This section has two tabs: Strategy Instructions and Template.

Strategy Instructions tab

A text editor where you write guidelines for how the AI should handle this type of interaction. The AI uses these instructions along with the ticket context, tool data, and knowledge base articles to generate a complete response.

Think of strategy instructions as the playbook you'd give a new agent:

• What data to check
• What decisions to make based on that data
• What to include in the response
• What tone to use
• What to avoid

See Strategy Instructions for detailed guidance on writing effective instructions.

Template tab

A rich text editor where you design the exact structure of the response. You can add:

• Static text — paragraphs, headings, formatting that appear the same in every response
• Ticket placeholders — dynamic values like customer name, ticket subject, pulled from the ticket
• Tool placeholders — values from connected tools like Shopify order number, tracking info
• AI blocks — sections where the AI generates text based on a prompt you provide

Use the {{ shortcut to open the placeholder menu, which shows all available placeholders grouped by source.

Using one or both tabs

You can use either tab on its own, or both together:

Most teams start with strategy instructions (it's faster to write and more flexible), then add a template later if they need more control over formatting. See Response Modes for a deeper comparison.

---

Activation & Context

This section controls when the workflow runs and what information it has access to.

When to Use

A natural language description of the situations this workflow handles. This text is evaluated by AI when:

• Ranking Assist suggestions for agents
• Deciding which workflow to execute in background automation
• Classifying incoming tickets

Write it as if you're explaining to a colleague when to use this workflow:

"Use when a customer is asking about the status of an order they've already placed, including questions about shipping, tracking, or delivery timeline. Don't use for pre-purchase questions about shipping policies."

Be specific about what the workflow handles AND what it doesn't handle.

Trigger Type

For background automation, this controls which ticket events trigger the workflow:

This setting only matters for workflows that run automatically in the background. A workflow an agent runs manually doesn't use a trigger.

Conditions

Optional rules that the ticket must match for this workflow to run. Unlike the "When to Use" description (which is evaluated by AI), conditions are evaluated deterministically — they either match or they don't.

Available condition types:

• All of — every condition must match
• Any of — at least one condition must match
• None of — no condition can match

Each condition checks a ticket field (subject, status, priority, tags, custom fields) against a value using operators like equals, contains, starts with, etc.

Example: A "Refund Request" workflow might have:

• All of: status equals "open"
• Any of: tags contain "refund" OR tags contain "return"
• None of: tags contain "escalated"

Tools

Select which integrations the workflow should pull data from before generating a response. The available tools depend on which integrations you've connected in your Stylo settings.

When a tool is enabled, the workflow automatically fetches the relevant data (e.g., order details from Shopify) and makes it available to both the AI and your escalation rules.

Knowledge Bases

Select which knowledge bases the AI should search when generating responses. The AI will find relevant articles and use them to ground its response in your actual documentation.

• Select specific knowledge bases to scope the search to relevant content (e.g., only search the returns policy KB for a refund workflow)
• Leave empty to search all available knowledge bases for your organization

---

Guardrails

Escalation Rules

Define conditions where the workflow should not generate a response and instead leave the ticket for a human agent.

Each rule checks a field from your connected tool data against a condition. The field picker shows all available fields based on which tools you've enabled — for example, if you've connected Shopify, you'll see fields like order total, fulfillment status, and financial status.

Rules are evaluated after tool data is fetched but before the AI generates anything. If any rule matches, the workflow stops immediately.

See Escalation Rules for common patterns and best practices.

Post-Response Actions

What happens after a response is sent. These apply to all execution modes (manual, cache, background).

Actions are applied after the response is posted. If the workflow escalates (an escalation rule matches), no actions are applied.

---

Automation

Controls how the workflow runs in the background. See Background Automation for the full guide.

Background Processing Mode

Response Delay

For internal and public modes, how many seconds to wait after generating the response before sending it. This gives agents a window to respond first — if they do, the automated response is discarded.

Recommended settings:

• Public reply: 30-60 seconds minimum
• Internal note: 0 seconds (internal notes don't compete with agent responses)

---

Saving and publishing

The editor header has two buttons:

• Save Draft — saves your changes without making them live. Use this while you're still iterating. Your draft is preserved even if you navigate away.
• Publish — saves your changes AND creates an immutable published snapshot. The new version goes live immediately — Assist suggestions and background automation will use it.

When to use each:

• Use Save Draft while you're building and testing. This lets you iterate without affecting the live version.
• Use Publish when you're confident in your changes and ready to go live.

Published versions are immutable snapshots. If you make a mistake, just edit and publish again — the new version replaces the old one. Previous versions are preserved in the execution history for audit purposes.

---

Test Panel

The test panel sits below the main editor sections. It lets you run your workflow against real tickets without leaving the page — the fastest way to iterate on your instructions or template.

How to test

1. Enter a Zendesk ticket ID in the test panel
2. Click Run Test
3. Stylo fetches the ticket's full context from Zendesk (customer info, conversation history, channel) and runs your workflow against it — using your current unsaved editor state, not the published version
4. The panel shows the result:

If the workflow generates a response:

• Generated Response — the full text the AI produced, displayed exactly as it would appear to the customer
• Tool Data — a collapsible section for each connected tool showing the data that was fetched (e.g., order number, fulfillment status, total price). Expand to see the full data the AI had access to.
• Metadata — execution time, which AI model was used, token usage (input/output), and which response mode ran (template, guided, or hybrid)
• Unresolved Placeholders — if any template placeholders couldn't be resolved, they're flagged here

If an escalation rule fires:

• The panel shows an escalation alert with the rule that triggered and the reason — no response is generated, just as it would work in production

You don't need to save before testing

The test panel runs against whatever is currently in the editor — your unsaved strategy instructions, template changes, escalation rules, and tool selections. This means you can:

1. Change a line in your strategy instructions
2. Hit Run Test on the same ticket
3. Instantly see how the output changed

No saving, no publishing, no leaving the page.

Testing tips

• Test with different ticket types. Run 3-5 tickets that represent different scenarios your workflow should handle. Does the AI adapt correctly?
• Test edge cases. Try a ticket where the customer has no order, or where the order is in an unusual state. Does the workflow escalate when it should?
• Compare before and after. When you change your instructions, re-run the same ticket ID to see how the output changed. This is the fastest way to refine your strategy instructions.
• Check the tool data. If the response is wrong, expand the tool data section to see what the AI was working with. The issue might be missing data, not bad instructions.

Moving to production

Once you're confident in the test results:

1. Publish your workflow to make it live
2. Enable Cache mode to start generating Assist suggestions — agents review before sending
3. Monitor how agents interact with the suggestions — do they use them as-is, edit them, or dismiss them?
4. When ready, move to Public reply mode for full automation

Related

• Creating a Workflow
• Conditions
• Actions