Custom Field Extraction with Directions

Custom Field Extraction allows administrators to define how specific data points are pulled from scanned invoices using AI instructions. This is particularly valuable when your organization needs to create custom invoice identifiers or track recurring invoice types, such as utility bills with meter numbers. By combining field selection with clear direction prompts, you can instruct Invoice AI to extract values based on patterns or positions on the document — helping ensure accuracy and consistency across your NetSuite records.

This article walks you through how to configure Custom Field Extraction in the Create AI Direction page of Invoice AI. It explains how to add fields, write extraction directions, and enhance data capture with examples and location-based hints.

Configure Custom Field Extraction with Directions

The Custom Field Extraction section lets you define how specific fields are captured from invoices using targeted directions. The process involves selecting a field, defining how it should be interpreted, and applying logic or examples to shape the extracted result.

Follow these steps to define how individual fields are extracted using custom directions:

1. Navigate to the bottom of the Create AI Direction page and locate the Custom Field Extraction section:

   

   This area is designed to define how specific fields should be extracted from an invoice.

2. Next, click the + Add Field button to open the Add Custom Field Extraction pop-up window:

   

   

3. On this page, you can use the Field dropdown to select from a set of core fields:

   

   

   You can also choose a body or transaction field that supports number, date, or text values.

4. For example, you can select Body: Invoice Number from the dropdown to configure extraction rules for the invoice number field:

   

5. Next, in the Field Extraction Direction section, you can enter your custom logic. For example, (concatenate invoice month, year, and meter number):

   

   This method is often used with utility bills, where invoices are linked to multiple meters. If your organization manages several meters across different locations, this approach allows you to construct a unique identifier in NetSuite using the invoice’s billing month, year, and meter number.

6. To make the instruction more specific, include an example, such as concatenate invoice month, year, and meter number for instance, (Dec 2024-24645757):

   

7. Avoid using the word “extract” in your direction:

   

   Invoice AI already understands that it needs to extract data, so explicit terms like “extract” are unnecessary.

8. Whenever possible, provide reference points to help the system locate the data. You might include phrases like “in the upper left-hand corner” or “first column” to direct the model’s attention. Think of this process as prompting any AI model, such as ChatGPT. Clear examples, logical phrasing, and reference cues all contribute to better recognition and accuracy.

9. Custom Field Extraction also supports body and column fields, enabling advanced data capture for non-standard layouts:

   

This concludes the overview of Custom Field Extraction with Directions in NetSuite. For more guidance, refer to the Invoice AI Directions documentation.