Expense Processing Skill

Overview

This skill handles the complete expense management workflow from receipt capture through reimbursement, including OCR extraction, policy validation, and Odoo integration.

Capabilities

• Receipt image OCR via PaddleOCR
• AI-powered data extraction
• Policy compliance validation
• Automatic expense categorization
• Odoo hr.expense creation
• Approval workflow routing
• Reimbursement tracking

Workflow

Step 1: Receipt Capture

Input Methods:

• Mobile app upload
• Email forwarding (receipts@insightpulseai.net)
• Telegram bot
• Manual upload via Odoo

Supported Formats:

• Images: JPG, PNG, HEIC
• Documents: PDF (first page)
• Max size: 10MB

Step 2: OCR Extraction

PaddleOCR Endpoint: https://ocr.insightpulseai.net/api/ocr/receipt

Request:

{
  "image": "<base64_encoded_image>",
  "language": "en"
}

Response:

{
  "text": "JOLLIBEE STORE 1234\n...",
  "confidence": 0.95,
  "regions": [...]
}

Step 3: AI Data Structuring

Claude Prompt:

Extract receipt data and return JSON:
- merchant_name: string
- merchant_tin: string (if visible)
- receipt_date: YYYY-MM-DD
- receipt_number: string
- subtotal: number
- vat_amount: number
- total_amount: number
- payment_method: cash|credit_card|gcash|paymaya|other
- expense_category: meals|transportation|supplies|communication|accommodation|other
- description: brief description

Step 4: Policy Validation

Approval Thresholds (PHP):

Policy Rules:

rules = [
    Rule("amount_limit", lambda e: e.amount <= 100000, "Amount exceeds maximum"),
    Rule("valid_category", lambda e: e.category in ALLOWED_CATEGORIES, "Invalid category"),
    Rule("has_receipt", lambda e: e.receipt_image is not None, "Receipt required"),
    Rule("or_required", lambda e: e.amount < 1000 or e.has_official_receipt, "OR required > 1000"),
    Rule("not_weekend", lambda e: not e.is_personal, "Personal expenses not allowed"),
]

Step 5: Odoo Integration

hr.expense Creation:

{
  "name": "Jollibee - Client Meeting Lunch",
  "product_id": 1,  // Meals category
  "unit_amount": 1250.00,
  "quantity": 1,
  "date": "2026-01-06",
  "employee_id": 77,  // RIM
  "payment_mode": "own_account",
  "reference": "OR-12345678",
  "description": "Client meeting with ACME Foods project team"
}

Category Mapping:

Step 6: Approval Workflow

Odoo States:

draft → reported → submitted → approved → posted → done
                 ↘ refused ↙

Notifications:

• On submit: Notify approver via Mattermost
• On approve/refuse: Notify employee
• On post: Update reimbursement queue

n8n Workflow Integration

The expense_receipt_capture.json workflow:

1. Webhook receives image
2. PaddleOCR extracts text
3. Claude structures data
4. Policy check (IF node)
   • Pass: Create expense in Odoo
   • Fail: Alert to #finance-general
5. Respond with result

Mattermost Notifications

Policy Violation Alert:

#### :warning: Policy Violation Alert
| Field | Value |
|:--|:--|
| **Amount** | PHP 7,500 |
| **Threshold** | PHP 5,000 |
| **Merchant** | Restaurant ABC |
| **Category** | meals |

**Action Required:** Manager approval needed

Expense Created:

#### :receipt: Expense Submitted
| Field | Value |
|:--|:--|
| **Employee** | Rey Meran (RIM) |
| **Amount** | PHP 1,250 |
| **Category** | Meals |
| **Status** | Pending Approval |

Receipt Requirements

Valid Receipt Elements

• Date of purchase
• Merchant/vendor name
• Itemized list
• Total amount
• OR number (for amounts > ₱1,000)
• VAT details (if applicable)

Common Issues

Reimbursement Schedule

• Cutoff 1: Submit by 10th → Reimbursed on 15th
• Cutoff 2: Submit by 25th → Reimbursed on 30th

Error Handling

errors = {
    "OCR_FAILED": "Could not extract text. Please re-upload clearer image.",
    "INVALID_DATE": "Receipt date could not be parsed. Please enter manually.",
    "CATEGORY_UNKNOWN": "Could not determine expense category. Please select.",
    "EMPLOYEE_NOT_FOUND": "Employee not found in system. Contact HR.",
    "DUPLICATE_RECEIPT": "This receipt may already exist. Please verify.",
}

Commands

Related Skills

• Finance Month-End
• BIR Tax Filing

Changelog

• v1.0.0 (2026-01-06): Initial release