Using Box Extract

Box Extract is Box’s AI-powered solution for automatically and accurately extracting structured data from unstructured enterprise content at scale. Extracted data can be saved as metadata in Box or delivered through APIs to downstream systems and custom applications. Box Extract also enables users to create Custom Extract Agents, complete with sample documents, metadata templates, field settings, AI-powered instructions and prompts, choice of the Standard or Enhanced Extract Agent, extraction policy, and more. Custom Extract Agents can be applied to specific root/parent folders, as well as the subfolders within root/parent folders, to automate data extraction at scale. Once activated, the Custom Extract Agent automatically processes content within its assigned folders and saves extracted values as metadata in Box. Use Custom Extract Agents to automate repeatable, high-volume extraction of information such as invoice numbers, dates, supplier IDs, contract terms, and other critical business data.

• Files supported:
  • Images: bmp, cr2, crw, dcm, dicm, dicom, dng, gif, heic, jpeg, jpg, nef, png, raf, raw, svg, tga, tif, tiff, webp
  • Design/CAD: ai, dwg, eps, ps, psd, xbd, xdw
  • Documents: boxcanvas, boxnote, doc, docx, gdoc, msg, odt, pages, pdf, rtf, webdoc, wpd
  • Presentations: gslide, gslides, key, odp, ppt, pptx
  • Spreadsheets: csv, gsheet, numbers, ods, tsv, xls, xlsm, xlsx
  • Code/text: as, as3, asm, bat, c, cc, cmake, cpp, cs, css, cxx, diff, erb, groovy, h, haml, hh, htm, html, java, js, json, less, log, m, make, md, ml, mm, php, pl, plist, properties, py, rb, rst, sass, scala, scm, script, sh, sml, sql, txt, vi, vim, xhtml, xml, xsd, xsl, yaml
• Enablement: An Enterprise admin must enable Box AI before you can use Box Extract
• Agent types: Standard Extract Agent and Enhanced Extract Agent
• Sources per agent: Up to 10 folders
  • For each folder, choose Selected folders only to extract from files in the root folder, or Selected folders and all subfolders to extract from files throughout the folder hierarchy.
• Agent limit per user: 100
• You cannot share Custom Extract Agents with other users. Only the user who creates a Custom Extract Agent can create, configure, and deploy it at scale.

Confirm Box AI and Box Extract are enabled for your organization. For admin enablement, visit Configuring Box AI.

​

Create a Custom Extract Agent

1. Navigate to Automate and select the Extract tab.
2. Click New+ and select Custom Extract Agent.
3. Select up to 10 sample files to reference and test your Custom Extract Agent with.

Admins and co-admins have the ability to perform the following actions when creating a Custom Extract Agent:

• Automatically generate a new metadata template with AI using up to 10 sample files for reference.
• Create a new metadata template and configure metadata fields and field properties manually.
• Use an existing metadata template.

End users only have the ability to use an existing metadata template.

​

Generate a new metadata template with Box AI

Using this option, Box AI analyzes the files in the source folders and generates a new metadata template based on the files you selected. You can then review the fields generated, adjusting any settings within the fields in Configure the agent. Admins and co-admins can select up to 10 sample documents for Box AI to analyze when generating a new metadata template. Based on the content and structure of these documents, Box AI automatically generates relevant metadata fields, including field names, field types, and extraction instructions. Admins and co-admins can review and edit the generated fields and their properties before saving and applying the template within the Custom Extract Agent. For more information about modifying metadata fields and extraction instructions, see Configure the agent. Once the template is created, you can select the template and fields you want to use in Configure the agent. The information shown after the template is created is an example of the data extracted from the files. AI-generated metadata templates are saved in the Admin Console and can be used by other users in Box Extract as well as other workflow products unless admins or co-admins hide them. Field instructions generated by Box AI are not automatically saved in the template, so users create their own field instructions when they use the template in a Custom Extract Agent.

​

Create a new metadata template manually

If you choose to create a new metadata template manually, you can add fields to the template by clicking the + Add field button and adding from the following field types:

• Text
• Number
• Date
• Dropdown
• Taxonomy

​

Use an existing metadata template

From the configuration page of the Custom Extract Agent, choose a Metadata Template from a list of available templates. This option is available for all users and groups.

• By default, all fields are selected. However, you can select or deselect which fields from the selected metadata template you want to extract data to.
• On existing Custom Extract Agents, managed users do not see the Add field and Save template buttons on the side panel.
• After a template is saved, the field name and field type are read-only for managed users. To change field definitions, an admin or co-admin must edit the template in the Admin Console.

​

Configure the agent

1. After you choose a metadata template and fields, rename the Custom Extract Agent by clicking the ellipsis (…) and then selecting Rename. The character limit for the name is 255 characters.
2. Select the AI Agent you want to use:
   1. Box AI Standard Extract Agent: Recommended for high-volume, structured or semi-structured documents with 50 pages or less and fewer than 20 extraction fields.
   2. Box AI Enhanced Extract Agent: Recommended for complex, large, or unstructured documents with 50 pages or more and more than 20 extraction fields for advanced use cases requiring deep insights and precision.
3. Enable the metadata fields you want to extract. At least one field must be enabled to save the Custom Extract Agent. You can also toggle all metadata fields on for extraction. Supported field types include:
   1. Text
   2. Number
   3. Date
   4. Dropdown
   5. Taxonomy
4. Add AI Instructions (recommended) for more accurate and precise extraction results. Attach short, structured prompts per field including details such as where the field is located, expected format, validation rules, edge cases, and more. The current prompt limit is 5,000 characters.
5. Select the extraction policy that determines whether the extraction process preserves existing metadata or overwrites it entirely.

Admins or co-admins can create and manage the taxonomy field type in the Admin Console. See Creating and managing taxonomies to learn more.

​

Activate and attach source folders

• After you save the Custom Extract Agent, you can activate or deactivate it and apply it to up to 10 source folders by clicking the ellipsis (…) and selecting Add Source Folder.
  • You can add folders to inactive Custom Extract Agents.
  • Choose Selected folders only to extract from files in root folders, or Selected folders and all subfolders to extract from files at every level in each folder’s hierarchy. You can change this setting for any folder after you add it.
• When the Custom Extract Agent is active and a source folder is attached, Box Extract automatically extracts data from uploaded files and applies that data as metadata.
  • Monitor the source folders for newly uploaded files to confirm extraction completes as expected.
  • Custom Extract Agents can automatically process files that already exist in a folder. This option doesn’t contain a file limit.
• Deactivating the agent stops new extraction processes. Previously extracted data remains as metadata on the associated files.

Each folder displays the Custom Extract Agents attached to it, so you can review which agents cover your content without opening the Extract tab.

​

Enable references and confidence levels

You can enable references and confidence levels for metadata extracted by a Custom Extract Agent. These capabilities can be enabled independently. Both are off by default.

• References help users verify extracted values by identifying where the supporting information is in the source document.
• Confidence levels indicate how confident Box Extract is in each extracted metadata value and can help identify values that might require additional review.

To enable either capability, go to Configurations > Advanced, then turn on Enable references or Enable confidence levels. To learn more about references and confidence levels in Box Preview, see Using confidence levels and references in Box Preview.

​

Resolve folder conflicts

Box Extract processes source folders using metadata cascade policies. A cascade policy is applied using a unique combination of metadata template and folder properties, so only one Custom Extract Agent or cascade policy can use a given template on a given folder. A conflict occurs when:

• Two Custom Extract Agents that use the same metadata template target the same folder
• A Custom Extract Agent and an existing metadata cascade policy use the same metadata template on the same folder

When you add several folders at once and one of them conflicts, Box attaches the folders that do not conflict and displays an error for the folders it blocks. Review your extraction sources after a bulk add to confirm which folders are attached. Nested folders behave differently. When a Custom Extract Agent is applied recursively to a parent folder, it automatically processes files within that folder and all of its subfolders. You can also assign a different Custom Extract Agent directly to a subfolder. In this case, the agent assigned directly to the subfolder takes precedence, ensuring each file is processed only once. When that happens, only the agent on the nearest folder runs, so each file is still processed once. A cascade policy on the subfolder behaves the same way. Example: the Invoices folder has a recursive agent, and its 2026 subfolder has an agent of its own. Files in Invoices/2026 use the 2026 agent. Files anywhere else in Invoices use the Invoices agent. To learn more about cascade policies, see Cascading metadata in folders.

​

Test and run extraction

Select Test Extraction on the Custom Extract Agent configuration page to test the agent against sample files before activating it. For managed users, the button displays on the side panel, outside the template ellipsis (…) menu, so you can add sample files and validate results before running the agent at scale. For admins, the button displays as Run extraction inside the template ellipsis (…) menu. Extraction runs only when you select Test Extraction or Run Extraction. Box does not automatically trigger extraction when you:

• Upload a file to an agent that already has a template selected.
• Replace a template on an existing agent.

Automatic extraction applies only when Box AI recommends a template during agent creation.

​

Troubleshooting

• If the template has been deleted or is inaccessible, select another template or contact your administrator for assistance.
• If enabled template fields have been removed, update and save the agent configuration.
• If the source folder or file has been deleted or you lost permission, restore the content or source manually.
• If activation fails due to missing configuration, ensure at least one field is enabled and that the template is valid.
• If a folder is blocked by a conflict, detach the folder from the other agent or remove the conflicting cascade policy, then add the folder again.

​

Best practices

• Use consistent, descriptive naming when creating or modifying a Custom Extract Agent name to simplify large-scale management.
• In the AI instructions, be as descriptive with as much detail as possible related to the document being extracted for more accurate and precise results.
• If you are not receiving the extraction results you want, try adjusting the AI instructions or revising your prompts until you achieve desired results.
• Use the Box AI Standard Extract Agent for consistent, high-volume extraction or the Box AI Enhanced Extract Agent for more complex documents requiring higher accuracy.
• Confirm folder permissions and template visibility with Admins or Co-Admins before activating agents.
• Use Selected folders only when you only need the files at a folder root. Using Selected folders and all subfolders on a deep hierarchy processes every file it finds, which increases extraction volume and AI unit consumption.