Routing settings

Routing decides what happens when a new conversation arrives: who gets it, how fast it has to be answered, and whether the AI agent drafts or sends first. Three things sit under Settings → Routing: rules, SLAs, and the AI default-assignee mode.

Routing rules

A list of conditions and actions evaluated top to bottom on every inbound message. The first rule that matches wins. Each rule looks like:

> If channel = email and subject contains "billing", assign to Finance Group, set tag billing, set priority high.

You can match on channel, customer fields, message content, sender domain, time of day, and more. Actions include assignment (to a person or a group), tagging, priority, and SLA selection.

For the full ruleset language and examples see routing-rules.

SLA policies

A policy is a pair of clocks attached to a conversation:

• First response: how long before a customer first hears back.
• Resolution: how long before the conversation is closed.

Each clock pauses outside business hours. Each clock has a warning threshold (you get an internal notification) and a breach threshold (the SLA is violated and reports flag it).

You can have multiple policies. Routing rules pick which one applies. Common pattern: VIP customers get a faster SLA than free-tier.

For the full SLA model see sla-policies.

AI default-assignee mode

This single toggle determines whether the AI agent ships replies on its own or always defers to a human first. Two modes:

Human-first

The AI agent drafts a reply, parks it as a pending draft, and assigns the conversation to a human. The human reads the draft, edits if needed, sends. Customers do not see the AI's draft until a human approves it.

Use human-first when:

• You are new to AI and want a human gate on every send.
• Your replies are high-stakes (legal, regulated, financial).
• You want CSAT scores attributable to humans, not AI.

AI-first

The AI agent sends directly when it has high enough confidence. Lower-confidence drafts still get parked for human review. Confidence threshold is configurable.

Use ai-first when:

• You have tuned the AI for a few weeks in human-first mode.
• Your replies are mostly informational (FAQ-style).
• You want to deflect volume from your team.

You can switch modes any time. Switching to AI-first does not un-park existing human-first drafts; those still need human approval. Switching to human-first does not un-send already-sent AI replies.

For more detail see autopilot, confidence-thresholds, and drafting-modes.

How rules, SLAs, and AI mode interact

Order of evaluation when a new message arrives:

1. Routing rules evaluate. The first matching rule sets assignment, tags, priority, and SLA.
2. SLA clock starts (or stays paused if outside business hours).
3. AI agent considers the conversation. If AI-first and confidence is above threshold, AI replies. If human-first or confidence is low, AI drafts and the human is notified.

If a routing rule explicitly sets the assignee to a human, the AI agent will still draft (in human-first mode) or send (in AI-first mode if it owns the conversation per the rule). You can write rules that exclude AI entirely for sensitive segments.

Common patterns

VIP fast-lane. Rule: customer.plan = enterprise → assign to Senior Support, SLA vip-2hr, AI mode override human-first.

Billing always to finance. Rule: subject contains "invoice|refund|charge" → assign to Finance Group, tag billing.

After-hours auto-reply. Rule: time outside business_hours → send auto-reply, set status pending. SLA does not start until business hours resume.

AI handles known FAQs. Rule: tags include "faq" plus AI-first mode → AI sends if confident, otherwise drafts.

Common mistakes

Forgetting rule order matters. Rules evaluate top to bottom and the first match wins. Drag your most specific rules to the top.

Setting SLAs without business hours. SLAs pause outside business hours. If you have not configured business hours in workspace-settings, SLAs run 24x7.

Switching to AI-first too early. Spend a couple of weeks in human-first to see how the AI behaves on your data before letting it send unsupervised.

What to do next

• Build your first rule in routing-rules.
• Define an SLA in sla-policies.
• Tune AI behavior in confidence-thresholds.
• See routing-overview for the big picture.