Skip to Content
FeaturesAI Tools and Workflows

AI Tools and Workflows

Resly comes with built-in automation tools to take actions, not just answer questions. Configure workflows to automate common support tasks.

Built-in Actions

The AI can perform these actions automatically based on customer needs:

The default tooling uses Resly’s own orders, refunds, and tickets tables so you can demo flows immediately. When you’re ready to hit your production systems, mirror your data into Supabase or wire up a connector that calls your API.

1. Gather Information

Purpose: Ask clarifying questions to get all the details needed

When to use:

  • Customer request is vague or incomplete
  • Multiple solutions exist and AI needs context
  • Required information is missing

How it works:

  • AI detects incomplete information
  • Asks targeted follow-up questions
  • Gathers details systematically
  • Proceeds once all information is collected

The AI won’t guess or assume—it asks questions to get the full picture before taking action.


2. Check Orders

Purpose: Look up order status and details

Data Source:

  • Uses the orders table inside your Supabase project
  • Seed data is included for demos; you can sync your own orders via a script or connector

What customers can ask:

  • “Where is my order?”
  • “What’s my order status?”
  • “When will my package arrive?”
  • “Track order #12345”

Response includes:

  • Order status
  • Tracking number (if available)
  • Estimated delivery date
  • Order details

To use order checks with live production data, either mirror your orders into the orders table or add a connector that calls your order system. Enterprise plans can help you design and maintain these custom integrations.


3. Process Refunds

Purpose: Issue refunds automatically

How it works today:

  • Writes to the refunds table so you get an audit trail
  • Updates the matching order status (refunded or partial_refunded)
  • You can still call an external payment API by adding a connector or workflow step

How it works:

  1. Customer requests refund
  2. AI verifies order eligibility
  3. Asks for confirmation before processing
  4. Issues refund if approved
  5. Confirms with customer

Important: Always set maximum refund amounts and approval rules. Large refunds should escalate to human review.

Safety features:

  • Confirmation required before processing
  • Maximum refund amount limits
  • Escalation rules for high-value transactions
  • Full audit trail of all refunds

4. Create Tickets

Purpose: Escalate complex issues to your support team

When it happens:

  • AI can’t find an answer in the knowledge base
  • Customer explicitly requests human support
  • Issue requires specialized expertise
  • Automated action fails or needs approval

Ticket includes:

  • Full conversation transcript
  • Customer information (if collected)
  • Priority level (low, normal, high, urgent)
  • Timestamp and context

Email notifications:

  • Configure under Account → Organization → Ticket Notifications
  • Automatic email sent to your primary notification address (optional)
  • Includes ticket details and link to admin dashboard

Tickets preserve full conversation context so your team never starts from scratch.


How Actions Work

AI Decision Making

The AI determines when to use tools based on:

  1. Customer intent - What is the customer trying to achieve?
  2. Available information - Is enough context provided?
  3. Tool availability - Which actions are configured and enabled?
  4. Confidence level - Is the AI certain about the next step?

Confirmation Flow

For sensitive actions (refunds, data changes):

  1. AI proposes the action
  2. Asks customer for explicit confirmation
  3. Waits for “yes” or similar affirmative response
  4. Executes action only after confirmation
  5. Confirms completion with customer

Customers must explicitly confirm before the AI takes any irreversible action like processing a refund.

Action Logging

Every action is logged for your review:

  • What action was taken
  • When it occurred
  • Who requested it (customer info)
  • Result (success or failure)
  • Full context (conversation leading to action)

Access logs in the ANALYZE tab.


Workflow Automation

Create multi-step workflows for common scenarios.

What are Workflows?

Workflows are automated sequences triggered by:

  • Keywords - Specific words or phrases
  • Customer intent - Detected goals or needs
  • Conditions - Rules you define

Example Workflows

Password Reset Workflow:

  1. Detect “forgot password” or similar phrase
  2. Verify customer identity
  3. Generate reset link
  4. Send email with instructions
  5. Confirm sent

Order Cancellation Workflow:

  1. Detect cancellation request
  2. Look up order status
  3. Check if cancellation is possible
  4. Confirm with customer
  5. Process cancellation
  6. Send confirmation email

Billing Question Workflow:

  1. Detect billing-related question
  2. Check question complexity
  3. If simple: Answer from knowledge base
  4. If complex: Create high-priority ticket
  5. Notify billing team immediately

Workflows remember context across the conversation, so customers don’t need to repeat information.


Integration Marketplace

Connect Resly to external APIs so the AI can fetch or update live data.

Available Templates Today

  • Shopify – look up orders or products from your store
  • Stripe – retrieve customer or subscription info before responding
  • Gmail – send notification or follow-up emails through your workspace account

You can also create custom connectors by describing the HTTP request (URL, method, headers, body, parameters) and the response mapping. The AI will call those connectors via the call_connector tool whenever your workflow triggers it.

Need Zendesk, Salesforce, Slack, or another provider? Those templates aren’t bundled yet, but you can build them yourself or work with us on an Enterprise engagement.

Meta channels (WhatsApp + Messenger) are included on Pro and Enterprise. Custom API connectors and white-glove builds remain Enterprise-only.

Setting up WhatsApp or Facebook Messenger? Follow the dedicated Meta channels guide for embedded signup, webhook verification, and testing.

How Integrations Work

  1. Navigate to Admin → Connectors
  2. Select integration from marketplace
  3. Authenticate securely (OAuth or API key)
  4. Configure integration settings
  5. Test connection
  6. Enable for production use

Security

  • All integrations use secure authentication (OAuth 2.0 or API keys)
  • Credentials encrypted at rest
  • Connections validated regularly
  • Can be revoked anytime

Testing Workflows

Before deploying, test all workflows thoroughly:

  1. Open the chat widget preview from Install Widget → Widget
  2. Trigger each workflow with test questions
  3. Review event log to see execution steps
  4. Verify actions complete correctly
  5. Check edge cases - What if something fails?
  6. Iterate based on results

Always test workflows in preview mode before enabling in production. Verify that actions complete successfully and handle errors gracefully.


Monitoring Automated Actions

Track all automated actions in the ANALYZE tab:

Metrics to monitor:

  • Action success rate
  • Failed actions and why
  • Average time to complete
  • Customer satisfaction after actions
  • Escalation rate

Use insights to:

  • Identify failing integrations
  • Improve workflow logic
  • Add error handling
  • Optimize performance

Best Practices

Start Simple

Don’t automate everything at once:

  1. Start with information gathering
  2. Add ticket creation for escalations
  3. Enable order checking if applicable
  4. Finally, add transactional actions like refunds

Set Clear Boundaries

Define what the AI should and shouldn’t do:

Do automate:

  • Simple, repeatable tasks
  • Information lookups
  • Ticket creation
  • Low-risk actions

Don’t automate (escalate instead):

  • Complex troubleshooting
  • High-value transactions without human review
  • Policy exceptions
  • Sensitive account changes

Monitor and Iterate

  • Review action logs weekly
  • Look for patterns in failures
  • Update workflows based on real usage
  • Add error handling for edge cases

The best automation strategies evolve over time based on real customer interactions and team feedback.


Next Steps

Last updated on