05. AI Extract Usage Guide

Overview

The AI Extract feature is designed to extract structured data from invoices using natural language prompts. It returns results in a format compatible with downstream processing.

Example Usage

If your prompt is:

extract this document and read ABN number

The system will process the document, and the extracted data will be returned accordingly.

Note: The default ExtractType is INVOICE.

Data Handling Behavior

• When ExtractType = INVOICE or RECEIPT, the extracted data will be saved into EzzyBills’ internal invoice data structures.
• If you have previously used an EzzyBills Extract task to process invoice data, that data will be overridden by the new AI Extract result.

Avoiding Data Overwrites

If you wish to extract data without overriding the existing invoice data, you must instruct EzzyBills to perform a generic extract.

Example Prompt:

generic extract, extract this invoice and read ABN

In this case, the result will not be saved into the internal invoice structures.

Result:
The ABN on the invoice is 26 008 672 179.

To retain the extracted values, you must save the result as JSON:

generic extract, extract this invoice and read ABN, save to json

The returned JSON will resemble standard invoice extraction output, but because the json does not contain a field “ExtractType”: “INVOICE”, it will not overwrite existing data.

Here's the extracted information from the invoice, including the ABN, saved in JSON format:

```json
{
  "invoice_number": "8172/02123818",
  "invoice_date": "15/08/2018",
  "supplier": {
    "name": "Bunnings Group Limited",
    "abn": "26 008 672 179",
    "address": "285 Burleigh Connection Road, Burleigh Waters QLD 4220",
    "phone": "07 5507 0700"
  },
  "client": {
    "name": "EZZYTEST PTY LTD"
  }
}

Conclusion

To prevent AI Extract from overriding existing invoice data, always begin your prompt with:

generic extract

This ensures a non-destructive, read-only extraction suitable for custom processing or separate data storage.