Skip to main content

Email AI

Overview

The Order Tracking AI Agent autonomously handles order tracking and order-status enquiries on the Email channel, in real time, out of the box. It understands queries like "When will my order arrive?" or "Why is my order delayed?" and takes the conversation to resolution — pulling live status directly from your store and, where connected, from third-party shipping partners.

For queries it isn't meant to resolve on its own — order modification, cancellation, or refund requests — the agent recognizes intent and hands the conversation off to your support team.

What's New in This Release

  • Revamped onboarding with improved customization at every step

  • Smarter, more nuanced responses — better handling of edge cases and ambiguous queries

  • New handoff rules to your support team

  • Third-party shipping partner integrations — the agent can coordinate directly with providers like AfterShip, ShipStation, Shiprocket, and others

How It's Structured

Setting up the Order Tracking AI Agent involves five steps:

  1. Channel Connection

  2. Agent Behaviour Configuration

  3. Order Handling Configuration

  4. Agent Handover (Escalation + Enrichment)

  5. Testing

  6. Go Live


1. Channel Connection

Connect your Email channel to activate the AI agent. Once connected, the agent monitors incoming order-related emails and responds in real time

2. Agent Behaviour Configuration

Customize how the agent sounds and behaves before it goes live:

Setting

What It Controls

Agent Tone

The overall tone of voice used in responses

Model Selection

Which AI model powers the agent

Personality

The agent's conversational character

Custom Instructions

Merchant-specific rules the agent should always follow

Brand Description

Context about your brand the agent uses to stay on-brand

Response Length

How concise or detailed responses should be

Guardrails

Boundaries on what the agent will and won't say or do

Customer Message Limit

Cap on how many messages the agent exchanges with a customer before escalating

3. Order Handling Configuration

Configure where the agent pulls order and shipment data from — your store's native order source and, optionally, third-party shipping partners such as AfterShip, ShipStation, and Shiprocket.

All supported order sources are applicable to the Order Tracking AI Agent except BigCommerce.
(See the full list of supported sources: Connect External Order Sources)

4. Agent Handover

Agent Handover determines when and how a conversation moves from the AI to your support team. It has two parts: Escalation and Enrichment.

a. Escalation

Escalation Rules define when the AI should stop responding and hand the conversation to your support team. These rules apply across all supported channels and take precedence over individual AI journeys and campaigns.

Default system rules (always on, not editable):

  • Customer asks to speak to a human — e.g. "Can I talk to an agent?" triggers an immediate handoff

  • Customer sentiment — frustration, repeated questions, or aggressive/all-caps language triggers an automatic escalation

Custom escalation rules (merchant-configurable):

  • Customer Intent — choose up to three intents (e.g. returns, cancellations, payment disputes) that should always route to a human instead of the AI

  • Trigger Messages — specific phrases (e.g. "I want a refund") that always trigger a handoff; configurable per channel

  • Conversation Length — a maximum number of AI messages before a conversation auto-escalates, configurable per channel

When multiple rules could apply, the conversation escalates as soon as the first rule is triggered — the AI doesn't wait for others. Once escalated, the AI stops responding and your Helpdesk's assignment/routing rules take over.

4b. Enrichment

Enrichment ensures that when a conversation does go to a human agent, it arrives as a structured, qualified ticket rather than a raw one.

  • Found under Settings → Smart Agent Handover. The AI identifies customer intent, collects any missing required information, and hands off an enriched ticket instead of an under-qualified one.

  • Configurable per channel (WhatsApp, Email, Instagram) — merchants define their own intents and the required questions per intent (e.g. Return Request → Order ID, product, reason). Six default intents ship out of the box via Manifest.

  • Before asking anything, the AI scans the full ticket (last 24 hours / 20 messages), so it never repeats a question the customer already answered.

  • "Talk to an agent" interception — if a customer asks for a human, the AI first tries to resolve or capture the query before escalating. A second request (count is configurable) triggers immediate assignment.

  • Merchants can restrict Enrichment to "Only Outside Working Hours", so tickets raised during live hours go straight to agents.

  • Agents see the identified intent and a collected-answers summary directly in the ticket, plus a Nudge option to re-ask an unanswered question after a merchant-set delay.

  • The AI communicates with customers in their own language throughout.

5. Testing

Before going live, test the full setup end-to-end using a test email address to confirm responses, order data retrieval, and handover rules all behave as expected.

6. Go Live

Once testing is complete, use the Go Live trigger to activate the Order Tracking AI Agent for real customer traffic on your Email channel.


Active Email Channels — turning Email AI on/off

At the top of the page you will see a table of all the email inboxes connected to your account, for example [email protected] and [email protected].

Each row has a toggle:

  • Toggle ON → Active: Email AI will read and reply to customer emails coming into that inbox.

  • Toggle OFF → Inactive: Email AI is switched off for that inbox and fallback to static flows.


Emails for Testing

This section lets you add specific email addresses that Email AI will reply to even if Email AI is turned off (Inactive) for that channel.

Why this is useful: say you've turned off Email AI for [email protected] because you don't want it live yet, but you still want to test how it responds. Add your own test email address here — you can then email in from that address and get live AI replies, without affecting real customers.

How to use it:

  1. Type the email address in the box (e.g. [email protected])

  2. Press Enter to add it

That email can now send messages and get AI replies regardless of the channel's Active/Inactive status.


When Not to Reply

Sometimes you don't want the AI replying at all — for example, spam, internal team emails, or a specific type of customer query that always needs a human. This is what When Not to Reply is for.

How it works:

  1. Click Manage filter (or the Edit link) next to "When not to reply" (or open it from the Spam/inclusion filter section).

  2. Pick the Email account this rule should apply to (e.g. [email protected]) — rules are set per inbox.

  3. Under Whenever an, choose the trigger — currently Email is received.

  4. Add one or more conditions using where. You can filter on things like:

    • Customer Mail Id — don't reply to emails from a specific address

    • Merchant Email Id — match against your own store's email

    • Email Subject — skip emails with a certain subject line

    • Email Body — skip emails containing certain text

    • Is Same Thread — skip replies within an existing thread

  5. Chain multiple conditions with And — click + Add event to add more.

  6. Set the value to match against in the Select an option field.

  7. Use the trash icon to remove a condition.

Once saved, any incoming email that matches these conditions will be skipped — Email AI simply won't respond to it


FAQs

Which order sources does the agent support?
All supported sources except BigCommerce. See Connect External Order Sources for the full list.

Can the agent handle cancellations or refunds directly?
No — it recognizes these intents and hands the conversation off to your support team rather than resolving them itself.

What happens if a customer asks for a human agent?
The AI first tries to resolve or capture the query. If the customer asks again, it escalates immediately. Customers can also be routed straight to a human via Customer Intent or Trigger Message rules.

Can I turn off the default escalation rules?
No — "customer asks for a human" and "customer sentiment" are built-in and always on.


Did this answer your question?