Outcomes
Read workflow health score and failure hotspots.
Use the Dead Letter Inbox.
Ask the AI debugger why a workflow failed.
Examples
Use these patterns while learning this chapter.
Example workflow
Payment failed recovery
Stripe reports a failed subscription payment and the tenant needs a polite retry sequence before suspension.
Conditions
- Invoice status equals unpaid
- Customer email exists
- Subscription plan is not free
Actions
- Notify billing admin
- Email customer with payment link
- Wait 3 days
- Check invoice status again
- Create follow-up task if still unpaid
Safety checks
- Tenant suspension is behind a human approval node
- Max one reminder email per day
- Dead letter if Stripe token refresh fails
Try it: Paste a sample payment_failed payload and verify the no-email branch catches missing customer email.
Example workflow
Ticket SLA breach escalation
A high-priority support ticket is close to breaching SLA and managers need visibility before the customer gets frustrated.
Conditions
- Priority is high or urgent
- Assigned agent exists
- Ticket status is not resolved
Actions
- Notify support manager
- Create escalation task
- Send customer update from macro
- Ask AI to summarize ticket history
Safety checks
- Customer email uses approved support template
- AI summary is read-only
- Escalation task is assigned to support manager group
Try it: Run Step Test on the customer email action with mocked delivery to confirm the rendered macro is correct.
Step 1
Review run history
Runs show status, duration, branch decisions, step outputs, and errors.
What to do
- Open Automation Studio or Command Center.
- Select the workflow.
- Review recent run status and step output.
Check before continuing
- Failures have enough context to debug.
- Duration is within expected limits.
Step 2
Use Dead Letter Inbox
Dead letters are failed runs that exhausted retries or need human attention.
What to do
- Open Dead Letter Inbox.
- Assign an owner or add notes.
- Retry, ignore, resolve, or convert to a ticket.
Check before continuing
- A failed run is not retried forever.
- Known false failures are closed with notes.
Step 3
Ask AI to debug
The AI Workflow Debugger reviews payloads, failed nodes, config, logs, and permissions to suggest exact fixes.
What to do
- Click Debug on a failed run.
- Read the suggested fix and affected node.
- Apply changes in a draft version before re-enabling.
Check before continuing
- The fix explains the root cause.
- A new preview confirms the branch/action now works.
Pro tips
- Use failure paths for customer updates and internal escalation.
- Pause workflows that fail repeatedly until the root cause is fixed.
Avoid these mistakes
- Do not ignore dead letters without recording why.
- Do not retry external failures if credentials or scopes are missing.