Skip to main content

AI-Powered Email → Inventory Order Automation

Overview​

This document describes a zero-code automation built with the Visual Workflow Builder that:

  1. Listens for incoming vendor emails via the inbound_emails module
  2. Uses an LLM to extract structured order data from the email body
  3. Automatically creates an inventory order if the email is an order confirmation
  4. Marks the email as processed

Trigger event: inbound_emails.inbound-email.created

No custom code is required — the entire flow is configured through the Visual Flow Builder admin UI.

The Complete Flow​

ASCII Diagram​

[Trigger] inbound_emails.inbound-email.created
    │
    ▼
[ai_extract]  key: "extracted"
  → extracts email_type, order_number, vendor, items[], total
    │
    ▼
[condition]
  → success if extracted.object.email_type IN [order_received, confirmation]
  │                                  │
  │ success branch                   │ failure branch
  ▼                                  ▼
[execute_code]                   [log]
  key: "order_input"               "Not an order email, skipping"
  → builds order_lines from
    extracted.object.items
    │
    ▼
[trigger_workflow]
  → createInventoryOrderWorkflow
  → input: {
      order_lines,
      total_price,
      status: "Pending",
      metadata: { order_number, vendor, inbound_email_id }
    }
    │
    ▼
[update_data]
  → module: inbound_emails
  → record: $trigger.id
  → sets status: "processed"
  → sets metadata.action_result.order_id: $last.result.order.id

Step-by-Step Table​

StepTypeKeyWhat It Does
1ai_extractextractedCalls Gemini Flash to extract email_type, order_number, vendor, items[], total from subject + HTML body
2condition—Checks extracted.object.email_type is order_received or confirmation; routes to failure branch otherwise
3execute_codeorder_inputMaps extracted.object.items into order_lines array with quantity/price fields
4trigger_workflow—Runs createInventoryOrderWorkflow with the built order input
5update_data—Marks the inbound email as processed and stores the created order ID

Follow-Up: Shipped Email Flow​

A second flow handles shipping confirmation emails. It is also triggered on inbound_emails.inbound-email.created.

[Trigger] inbound_emails.inbound-email.created
    │
    ▼
[ai_extract]  key: "extracted"
  → extracts email_type, order_number, tracking_number, carrier
    │
    ▼
[condition]
  → success if extracted.object.email_type === "shipped"
  │                         │
  │ success                 │ failure
  ▼                         ▼
[read_data]              [log] "Not a shipping email, skipping"
  key: "found_order"
  → module: inventory_orders
  → filter: metadata.order_number = extracted.object.order_number
    │
    ▼
[update_data]
  → record: found_order.result[0].id
  → sets status: "Shipped"
  → sets metadata.tracking_number: extracted.object.tracking_number
  → sets metadata.carrier: extracted.object.carrier

Variable Reference Cheatsheet​

VariableSourceExample Value
{{ $trigger.subject }}Inbound email subject line"PO Confirmation #2024-001"
{{ $trigger.html_body }}Inbound email HTML bodyFull HTML string
{{ $trigger.from_email }}Sender email address"[email protected]"
{{ $trigger.id }}Inbound email record ID"email_01HXYZ..."
{{ extracted.object.email_type }}ai_extract output"order_received"
{{ extracted.object.order_number }}ai_extract output"PO-2024-001"
{{ extracted.object.vendor }}ai_extract output"Acme Textiles"
{{ extracted.object.items }}ai_extract outputArray of line items
{{ extracted.object.total }}ai_extract output1250.00
{{ extracted.usage.totalTokens }}ai_extract token usage165
{{ order_input.result.order_lines }}execute_code outputMapped order lines array
{{ $last.result.order.id }}Most recent operation resultCreated order ID
{{ found_order.result[0].id }}read_data outputExisting order record ID

ai_extract Configuration for Order Emails​

{
  "model": "google/gemini-2.0-flash-exp:free",
  "input": "Subject: {{ $trigger.subject }}\n\n{{ $trigger.html_body }}",
  "system_prompt": "Extract order info from this vendor email. Be precise. Return only the requested fields.",
  "schema_fields": [
    {
      "name": "email_type",
      "type": "enum",
      "enumValues": ["order_received", "confirmation", "shipped", "delivered", "other"],
      "required": true
    },
    { "name": "order_number", "type": "string" },
    { "name": "vendor", "type": "string" },
    { "name": "items", "type": "array", "description": "List of items with name, quantity, unit_price" },
    { "name": "total", "type": "number", "description": "Total order value" }
  ],
  "fallback_on_error": true
}

Condition Filter Rules​

The condition operation uses filter rules to check the extracted type:

{
  "field": "extracted.object.email_type",
  "operator": "in",
  "value": ["order_received", "confirmation"]
}

For the shipped flow:

{
  "field": "extracted.object.email_type",
  "operator": "eq",
  "value": "shipped"
}

Notes​

  • Both flows run concurrently on the same trigger event. The condition in each flow ensures only the relevant flow takes action.
  • The fallback_on_error: true on ai_extract prevents the flow from crashing on malformed emails — it simply skips through the condition's failure branch and logs the skip.
  • The execute_code step is needed to reshape the raw items array (which may have inconsistent field names from the LLM) into a normalized order_lines format expected by createInventoryOrderWorkflow.
Last updated on