Create and Use Custom Extractions

Custom extractions allow you to quickly (and in bulk) pull relevant information from different types of documents (e.g. diagnosis codes from medical records, notice periods from contracts, date fields from forms, etc). For example, if patient records are part of your corpus, you can extract out values from those records — drugs and dosages, medical care providers, etc — that are important to capture. Custom extractions are particularly useful in cases when keyword or regular expression searches are too ineffectual or cumbersome.

This document introduces custom extractions and explains how extractions are generated and managed in a project. 

Requirements

The following permissions are applicable to custom extractions:

• Project Admins, who have the below permissions by default and full access to manage extraction fields on the dedicated Extractions page in Project Settings.
• Users with Generate permissions of Everlaw AI extractions (i.e. Summaries, topics, extractions, and doc Q&A permissions) and Batch actions enabled under their Everlaw AI permissions, who can batch add, edit, and hide saved extraction fields via the results table’s batch action dialog.
• Users with Generate or View permissions for Everlaw AI extractions (i.e. Summaries, topics, extractions, and doc Q&A permissions), who can interact with extractions in the results table and review window. 

In addition, you must be working in a project with the Enable summaries, topics, extractions, and doc Q&A setting switched on from the Project Settings > Everlaw AI page.

Intro to custom extractions

How extractions work

Extractions are pieces of information pulled from documents based on the extraction field that your team defines. They can be viewed in documents’ review windows and in the results table, as columns.

An extraction field defines the category of information to be extracted from the document. It is composed of a name, description, and field type (i.e. Entity, Text, Number, DateTime).

There can be multiple extractions per field. For example, if your field has a name “Dates,” a description “Key dates mentioned in the thread,” and a field type “DateTime,” Everlaw AI will extract all data values it deems relevant from the document according to the definition of the field. 

Extraction fields can be defined and managed from multiple locations in the platform. However, all extraction fields are automatically saved at the project level. This means that each field is available to be run on any document in the project. 

Relevant feature locations

Depending on their permissions, users can interact with and manage extractions from the following locations 

• Extractions page: Located in Project Settings, this page serves as a central repository for all custom extraction field prompts created within a project. Any extraction field prompt defined in the review window or through the batch menu in the results table is automatically saved here. This centralized management allows Project Admins to create, oversee, standardize, and refine the extraction fields available to their team, ensuring consistency and reusability across all documents. To learn more, visit Project-level management of custom extraction fields.
• Review window: In a document’s review window, users can:
  • View existing extractions from the document
  • Add, edit, and delete values for a given field
  • Generate (and regenerate) extractions for a single document
  • Add new extraction fields to the project
  • Reconfigure existing fields
• Results table: In the results table, users can view columns for each extraction field in the project. Columns are added from the View menu.
• Results table batch dialog: From the batch dialog, users can batch generate (and regenerate) extractions from a set of documents, create, edit, copy, and hide extraction field prompts on the project, and batch add columns of fields being generated to the results table for all selected fields.

Workflow recommendations

Include in project setup

While not required, we recommend that Project Admins set up expected extraction fields as part of new project setup. This creates uniformity in how fields are set up and allows reviewers to get up and running faster than if they had to create all extraction prompts themselves. This can be done prior to uploading documents, and at any time after, from the Extractions page in Project Settings.

Test early

We recommend testing your extraction configurations on a few documents before scaling use.

Testing your extraction fields helps you fine-tune the description language to avoid under-extracting or over-extracting. Based on your testing, you may find that, for example, you need to use more precise language or describe what you are not looking for in addition to what you are looking for.

Practice iterative refinement

When refining an extraction field prompt, you can choose from two main approaches: creating a new variation (copy) of the field or editing the existing field. The table below compares these two options:

Ultimately, the best approach depends on your preference for re-running fields and whether you're comfortable with related values potentially appearing in different columns.

Understand credit consumption

When generated in batch for multiple documents at once, extractions consume AI credits. The number of credits required is shown on the generate button. 

The baseline number of credits used depends on the number of documents included in the action and the amount of text in those documents. Then, the baseline is multiplied based on the number of fields you are extracting for, in multiples of 5:

• 1-5 fields = X credits
• 6-10 fields = 2x credits
• 11-15 fields = 3x credits

This means that generating extractions for 5 fields uses the same number of credits as generating extractions for 2 fields, and generating extractions for 6 fields uses twice as many credits as generating extractions for 5 fields. 

Generating extractions on an individual document from the review window does not consume credits. 

Generate extractions for a single document

Required permissions: Generate permissions for Everlaw AI extractions (i.e. Summaries, topics, extractions, and doc Q&A permissions)

To get started generating extractions for a single document in the document review window:

1. Open the Review Assistant context panel. Then select the Extractions tab.
   
2. Select the expand button to expand the Selected fields list.
   
   Here you can create new fields and/or select to run fields that were previously created in the project.

For detailed instructions on how to generate extractions for a single document, see our Work with Extractions in the Document Review Window article.

Generate extractions in batch

Required permissions: Generate permissions for Everlaw AI extractions (i.e. Summaries, topics, extractions, and doc Q&A permissions) and have Everlaw AI Batch actions permissions enabled.

You can generate extractions with up to 75 fields at a time on up to 250,000 documents at a time.

To generate extractions in batch:

1. In the results table, select the set of documents from which you want to extract information.
   You may select up to 250,000 documents at a time for extraction.
2. Select Everlaw AI > Extractions.
   
   This opens the Batch generate extraction fields dialog.
   
    
3. [Optional] In addition to letting you select existing extraction field prompts to batch generate extractions from, this dialog also allows you to optionally:
   • Create new fields:
     1. Select + New field.
        A new blank field entry appears.
     2. In the Name field, enter a name with which to identify the extraction field (e.g. contract date, patient name, indemnity language).
     3. In the Type field, choose the field type that should be used to define the expected format of the generated extractions. Your options are:
        • Text: The extractions are expected to be multiple words or sentences in length
        • Number: The extractions are expected to be numbers or references to numerical information
        • DateTime: The extractions are expected to be in datetime format or are references to dates and times
        • Entity: The extractions are expected to be names of entities, such as persons, organizations, or locations. 
     4. For the Description, enter a prompt that guides the AI on what information the extraction field is intended to extract.
        
     5. Select Save.
   • Edit fields:
     1. Select the field’s Edit field button. This makes the prompt editable.
     2. Edit the field name and description of the new extraction field as needed.
     3. Select Save.

• Copy fields: 
  1. Select the extraction fields Copy field button.
     A new extraction field appears ready to be edited. It has the same field type and description as the original. The original’s name is appended with a number (e.g. “Dates” becomes “Dates (2)”).
  2. Edit the field name and description of the new extraction field as needed.
  3. Select Save.
• Hide fields from the project: Toggle the field’s Visible in project switch.

4. Select the fields you want to generate extracted values for from the list.
   You can select a max of 75 extraction fields at a time.
   See the Understand credit consumption section, above, to understand how the number of fields you select determines the number of credits the batch action requires. 
5. [Optional] If you do not want the selected extraction fields to appear as columns in the results table, uncheck Add selected fields as columns in your results table view, which is selected by default.
6. [Optional] To regenerate values for all extraction fields with updated prompts, select Regenerate fields with updated prompts.
7. Select Generate.

View extraction results in the results table

Required permissions: Generate or View permissions for Everlaw AI extractions (i.e. Summaries, topics, extractions, and doc Q&A permissions),

To select which extraction fields to include as individual table columns:

1. Open the results table.
2. Open View > Add or Remove columns.
   
   This opens the Add or remove columns dialog.
3. Select the fields you want to add to your results table. They are listed under the heading EXTRACTION COLUMNS. 
4. Select Save.
   The dialog closes, and the table updates per your selections.
   

Edit, copy, or hide existing extraction fields

Required permissions: Generate permissions for Everlaw AI extractions (i.e. Summaries, topics, extractions, and doc Q&A permissions) and have Everlaw AI Batch actions permissions enabled.

If you are a Project Admin, you can edit, copy, or hide extraction fields from the Project Settings > Extractions page or from the result table’s Batch > Extractions dialog.

All other users can perform these actions from the result table’s Batch > Extractions dialog or when reviewing a single document.

To perform any of these actions from the result stable, follow the instructions in this article’s section about generating batch extractions, as they are performed through the same dialog as batch extractions. 

To perform any of these actions while reviewing an individual document, see our Work with Extractions in the Document Review Window article.

If you follow the aforementioned instructions with the intent to create, edit, or hide extraction fields, and you’re not ready to generate extractions, select Cancel to close the dialog. Your changes will still be saved to the project.

Search on extracted values

AI extraction fields can be used as search terms, allowing you to leverage document extractions to more effectively and efficiently identify documents based on specific identifiable data.

Before you start:

• Extraction field availability is project dependent. Before extraction fields can be used as search terms, they must exist and be visible on the project.
• Search queries only pick up extracted values if they exist, meaning you must run extraction fields on documents before using the fields in the search.

To search on existing extractions:

1. Go to the Search page.
2. Select Find a term. Then type the extraction field name in the input field that appears in the search builder. It is listed under EVERLAW AI.
   

3. Select or enter criteria to search on.
   

   Note

   The input field(s) vary depending on field type (i.e. Text, DateTime, Entity, or Number)
   

Additional considerations

Extractions may not fully and accurately capture each value in a document. Extractions may be wrong in four ways:

• The extraction does not exist in the text
• The extraction exists in the text, but shouldn't have been extracted based on your instructions
• There's relevant text that should have been extracted, but wasn't
• The extraction is generally correct, but may not exactly match the exact document text (this is more of an issue for longer extractions)

To remedy these errors, you can add, edit, and delete extracted values on an individual document level. To learn how to do so, see our Work with Extractions in the Document Review Window article.