Letterbook Docs

Automations vs AI triage

Understand when to use strict automation rules and when to use AI-based routing, priority, and labels.

Last updated June 25, 2026

What this guide covers#

Letterbook has two settings areas that can organize incoming tickets:

Automations are strict rules and workflow toggles.
AI Triage is AI-based classification for priority, assignment, and labels.

They are complementary, but they solve different problems. Use Automations when the condition is exact and predictable. Use AI Triage when the decision depends on the meaning of the customer's message.

Quick comparison#

Area	Automations	AI Triage
Where to configure	Settings > Automations	Settings > AI Triage
Best for	Exact rules, toggles, and mechanical workflow behavior	Semantic decisions that need message understanding
Input style	Form fields, switches, selectors, sender/domain patterns, subject patterns	Natural-language instructions
Output examples	Round robin owner, static label, auto-response, ignored email, auto-resolve, translation defaults	Priority, assignee, labels chosen by AI
Predictability	Deterministic when the rule matches	Probabilistic and should be tested with real examples
Maintenance	Update the specific rule or toggle	Update the instruction text and examples

Use Automations for strict rules#

Automations are best when you can describe the rule without interpretation.

Examples:

Assign every new ticket to eligible agents in round-robin order.
Apply a Billing label when the sender domain is billing-provider.com.
Apply a label when the subject contains invoice.
Ignore messages from noreply@vendor.com.
Only process forwarded messages sent to support@company.com.
Send an automatic acknowledgement from a specific inbox.
Resolve Pending tickets after a fixed number of days.
Translate non-English conversations by default.

These rules do not need the AI to understand the customer request. If the condition matches, Letterbook applies the configured behavior.

Use AI Triage for judgment calls#

AI Triage is best when the customer message needs interpretation.

Configure AI Triage in Settings > AI Triage. It currently supports:

Priority: the AI sets conversation priority from your instructions.
Assignee: the AI routes conversations to the right teammate or group.
Labels: the AI applies one or more labels from your instructions.

Examples:

Set URGENT when the customer appears blocked by an outage, security incident, or production failure.
Assign billing exception requests to the teammate who can approve credits.
Assign technical integration failures to the teammate who owns developer support.
Apply a Bug label when the message describes broken product behavior, even if the subject does not contain the word "bug."
Apply Churn risk when a customer expresses cancellation intent or severe frustration.

These are not exact string matches. The AI reads the message and decides whether your instructions apply.

Auto labels vs AI labels#

This is the most common overlap.

Use Auto Label Tickets when the rule is exact:

Sender is finance@customer.com
Sender domain is enterprise-customer.com
Subject contains invoice

Use AI Triage labels when the label depends on meaning:

Customer is asking for a refund but does not use the word "refund"
Customer describes a bug in natural language
Customer is upset enough to be a churn risk
Customer asks an account-access question that needs identity review

If both rules apply, use the strict auto-label for the guaranteed signal and AI labels for interpretation.

Round robin vs AI assignment#

Use Round robin assignment when tickets should be distributed evenly across eligible teammates.

Use AI Triage assignment when the owner depends on the ticket content:

Billing questions go to one person.
Technical issues go to another person.
Enterprise escalations go to a named owner or group.
Security or privacy issues go to a designated reviewer.

Round robin answers "who is next in rotation?" AI assignment answers "who is the right owner for this issue?"

Recommended setup order#

Add strict intake filters first, such as ignored senders and allowed recipients.
Add exact labels for known sender domains or subject patterns.
Enable round robin if general tickets should be distributed evenly.
Add AI Triage instructions for priority, assignment, and semantic labels.
Test with realistic conversations from each channel.
Review incorrect results and update the relevant rule or AI instruction.

This order keeps obvious cases deterministic and leaves nuanced decisions to AI Triage.

Troubleshooting wrong outcomes#

When something routes incorrectly, identify which system made the decision:

Symptom	Check
A sender/domain label was wrong	Auto Label Tickets
A subject keyword label was wrong	Auto Label Tickets
A semantic label was wrong	AI Triage > Labels
Ownership rotated to the wrong eligible person	Round Robin Assign Tickets
Ownership should have gone to a subject-matter expert	AI Triage > Assignee
Priority was wrong	AI Triage > Priority
A non-ticket email entered the queue	Ignore Non-Ticket Emails

Update the strict rule when the match condition is wrong. Update the AI instruction when the model misunderstood the ticket.