SelfHostSelfHost

n8n Hands-On Guide 9: Building an Invoice Processing Bot (Part 1)

10 months ago

What Can This Workflow Do?

For finance professionals who frequently process reimbursement invoices, manually entering invoice information is tedious.

This workflow automatically recognizes invoice images, extracts key information, stores it in a spreadsheet, and finally sends an email notification.

  • Traditional Method: Accountants manually review each invoice → Manually input invoice number, date, amount → Then verify the tax rate.
  • Current Process: AI automatically recognizes → JSON data cleaning → Directly enters Google Sheets, eliminating manual entry. The system automatically sends an email (with clearly formatted HTML) to ensure the relevant person receives it promptly.

https://appstore.lazycat.cloud/#/shop/detail/cloud.lazycat.app.n8n

Practical Steps

Open n8n and create a new workflow.

image.png

First, Add All Nodes

Step 1: I want to trigger via chat, so create a new 'On Chat Message' node.

image.png

Step 2: Search for OpenAI.

image.png

Select this node for analyzing images:

image.png

Step 3: Add a Code node to clean the data.

image.png

Step 4: Save to Google Sheets.

image.png

Select 'Add a new row'.

image.png

Step 5: Extract the email address from the chat message, add another Code node.

image.png

Step 6: Add an If node to check if the email is empty.

image.png

Step 7: Send the email, search for Gmail.

image.png

Select 'Send a message'.

image.png

These are all the workflow nodes.

image.png

Next, detailed configuration instructions for each node.

1. Chat Message

Double-click the node, add the file upload property.

image.png

Turn the switch on.

image.png

Now you can test it. For example, I have this invoice image on my computer:

1605491102752717.jpg

In the chat box, type "Send to your.email@gmail.com", click the paperclip icon, and select your invoice image.

image.png

Click send. You can see the output result; the name is data0, which we will use later.

image.png

2. OpenAI Configuration

Go to the official website openai.com, select the API platform.

image.png

In the top right dashboard, go to API Keys, and create a new key.

image.png

Give it a name, for example, 'n8n'.

image.png

Note: You need to have funds in your account first, otherwise the call will fail. Or use another API platform.

For example, I found my OpenAI account had no funds, so I temporarily switched to using Gemini. The steps are the same.

image.png

Select the model. Use the prompt below as a reference.

## Role

You are a professional "Invoice Information Extractor". The input is a piece of invoice text (possibly from OCR, email body, or chat paste), and the output is strictly valid JSON, suitable for direct writing to a database/Google Sheets.

## Task
Identify and extract fields from the given invoice text, return a single JSON object, placed inside a markdown code block (`json ... `). Do not output any explanations, comments, or extra characters.

## Output Requirements

Output only a single JSON code block, no additional text.

Use `null` for any values that do not exist or are uncertain.

Dates must be in YYYY-MM-DD format (ISO 8601).

For amount/quantity numerical fields, use number type (not strings), preserve precision but do not add currency symbols or thousand separators.

Use decimals for tax rates (e.g., 1% → 0.01), and also provide a display field (e.g., "1%").

The `items` array must exist; if there are no line items, output an empty array `[]`.

## Field Definition (Schema)

{
  "invoice": {
    "title": "string|null",
    "number": "string|null",
    "issue_date": "YYYY-MM-DD|null",
    "currency": "string|null",
    "buyer": {
      "name": "string|null",
      "tax_id": "string|null",
      "note": "string|null"
    },
    "seller": {
      "name": "string|null",
      "tax_id": "string|null"
    },
    "items": [
      {
        "name": "string|null",
        "spec": "string|null",
        "unit": "string|null",
        "quantity": "number|null",
        "unit_price": "number|null",
        "amount_without_tax": "number|null",
        "tax_rate": "number|null",
        "tax_rate_display": "string|null",
        "tax_amount": "number|null"
      }
    ],
    "totals": {
      "amount_without_tax": "number|null",
      "tax_total": "number|null",
      "amount_with_tax": "number|null"
    },
    "issuer": "string|null"
  }
}

Select the input type as 'Bytes', the name is data0 from above. image.png

Execute it, and you'll see the output result on the right.

image.png

Data Cleaning

This step is to format the result from the previous step to make it cleaner and more presentable.

Reference input code:

// n8n Code Node - Invoice Data Parsing
// Get input data (handle possible string format)
const inputData = $input.first().json.choices[0].message.content;

// If input is a string, parse it as JSON first
let invoiceData;
if (typeof inputData === 'string') {
  // Remove possible markdown code block markers
  const cleanData = inputData.replace(/^```json\s*/, '').replace(/\s*```$/, '').trim();
  invoiceData = JSON.parse(cleanData);
} else {
  invoiceData = inputData;
}

// Extract invoice data
const invoice = invoiceData.invoice;

// Return flattened JSON data
return [{
  json: {
    title: invoice.title,
    invoiceNumber: invoice.number,
    issueDate: invoice.issue_date,
    currency: invoice.currency,
    issuer: invoice.issuer,
    buyerName: invoice.buyer.name,
    buyerTaxId: invoice.buyer.tax_id,
    buyerNote: invoice.buyer.note,
    sellerName: invoice.seller.name,
    sellerTaxId: invoice.seller.tax_id,
    itemName: invoice.items[0].name,
    itemSpec: invoice.items[0].spec,
    itemUnit: invoice.items[0].unit,
    quantity: invoice.items[0].quantity,
    unitPrice: invoice.items[0].unit_price,
    amountWithoutTax: invoice.items[0].amount_without_tax,
    taxRate: invoice.items[0].tax_rate,
    taxRateDisplay: invoice.items[0].tax_rate_display,
    taxAmount: invoice.items[0].tax_amount,
    subtotal: invoice.totals.amount_without_tax,
    totalTax: invoice.totals.tax_total,
    grandTotal: invoice.totals.amount_with_tax
  }
}];

Execute it, and you might find an error.

image.png

Don't panic, use AI to help us fix it. I used Claude and asked like this: "This is my input in n8n:" [Copy the code from the left side] "This is the code I want for cleaning:" [Copy the code above] "But I encountered this problem:" [Copy the error message]

Claude directly gave me the corrected code:

image.png

// n8n Code Node - Invoice Data Parsing

// Get input data
const inputItem = $input.first().json;

// Extract text content based on the actual data structure
let textContent;
if (inputItem.content && inputItem.content.parts && inputItem.content.parts[0]) {
  textContent = inputItem.content.parts[0].text;
} else if (inputItem.choices && inputItem.choices[0]) {
  // Compatible with other possible data structures
  textContent = inputItem.choices[0].content?.parts?.[0]?.text || 
                 inputItem.choices[0].message?.content;
} else {
  // If neither matches, try using it directly
  textContent = inputItem;
}

// Parse JSON data
let invoiceData;
if (typeof textContent === 'string') {
  // Remove possible markdown code block markers
  const cleanData = textContent.replace(/^```json\s*/, '').replace(/\s*```$/, '').trim();
  invoiceData = JSON.parse(cleanData);
} else {
  invoiceData = textContent;
}

// Extract invoice data
const invoice = invoiceData.invoice;

// Handle multiple item entries - return each item as an independent record
const items = [];
for (let i = 0; i < invoice.items.length; i++) {
  const item = invoice.items[i];
  items.push({
    json: {
      // Basic invoice information
      title: invoice.title,
      invoiceNumber: invoice.number,
      issueDate: invoice.issue_date,
      currency: invoice.currency,
      issuer: invoice.issuer,
      
      // Buyer information
      buyerName: invoice.buyer.name,
      buyerTaxId: invoice.buyer.tax_id,
      buyerNote: invoice.buyer.note,
      
      // Seller information
      sellerName: invoice.seller.name,
      sellerTaxId: invoice.seller.tax_id,
      
      // Item details
      itemIndex: i + 1,  // Item sequence number
      itemName: item.name,
      itemSpec: item.spec,
      itemUnit: item.unit,
      quantity: item.quantity,
      unitPrice: item.unit_price,
      amountWithoutTax: item.amount_without_tax,
      taxRate: item.tax_rate,
      taxRateDisplay: item.tax_rate_display,
      taxAmount: item.tax_amount,
      
      // Invoice totals
      subtotal: invoice.totals.amount_without_tax,
      totalTax: invoice.totals.tax_total,
      grandTotal: invoice.totals.amount_with_tax
    }
  });
}

// If there are no items, at least return basic invoice information
if (items.length === 0) {
  return [{
    json: {
      title: invoice.title,
      invoiceNumber: invoice.number,
      issueDate: invoice.issue_date,
      currency: invoice.currency,
      issuer: invoice.issuer,
      buyerName: invoice.buyer.name,
      buyerTaxId: invoice.buyer.tax_id,
      buyerNote: invoice.buyer.note,
      sellerName: invoice.seller.name,
      sellerTaxId: invoice.seller.tax_id,
      subtotal: invoice.totals.amount_without_tax,
      totalTax: invoice.totals.tax_total,
      grandTotal: invoice.totals.amount_with_tax
    }
  }];
}

return items;

I copied the code above again, executed it, and found the result was now correct.

PS: Remember this? This is how you debug from now on.

image.png

Google Sheets

For Google Sheets configuration, you can refer to my previous guide.

I created a new sheet on Google: 'n8n Invoices'

image.png

Reference column names for the sheet:

Invoice Title	Invoice Number	Issue Date	Buyer Name	Buyer Tax ID	Buyer Note	Seller Name	Seller Tax ID	Item Name	Specification Model	Unit	Quantity	Unit Price	Amount Without Tax	Tax Rate	Tax Rate Display	Tax Amount	Subtotal Without Tax	Total Tax	Grand Total With Tax	Issuer

image.png

Drag the corresponding fields from the left side to their respective column name positions

Author
天天
n8n Hands-On Guide 9: Building an Invoice Processing Bot (Part 1)