Skip to main content
Data extraction means the agent reads a document and pulls out the information you care about. For example, an agent can read an invoice and return the vendor, invoice number, date, line items, and total.
Data Extraction result panel showing extracted invoice fields and a line-items table

A Data Extraction result: structured fields (invoice number, vendor, dates, total) plus a line-items table you can review and download.

​
What files can be used

Agents can work with common business files, including:
  • PDFs
  • Images
  • Word documents
  • Excel spreadsheets
  • CSV files
  • PowerPoint files
  • Text and Markdown files

​
How extraction works

1

Add the document

Upload a file or let the agent receive it from email, a folder, or another connected app.
2

Tell the agent what to find

List the fields you need, such as invoice number, vendor name, due date, total, and line items.
3

Review the result

The agent returns the data in a table or structured result. When available, it also shows where the value was found in the document.
4

Send the data where it belongs

The agent can add the data to a spreadsheet, create a file, update a business system, or return it to the user.

​
Define the fields you need

Be specific about the fields the agent should extract. Good examples:
  • “Extract invoice number, vendor name, invoice date, due date, subtotal, tax, total, and currency.”
  • “Extract each line item with description, quantity, unit price, and line total.”
  • “If a purchase order number is missing, mark it as missing and ask the user.”
Use the field names your team already uses. This makes the result easier to review and export.

​
Use source checks

For many documents, the agent can show where a value came from in the original file. This helps reviewers quickly answer:
  • Did the agent read the right value?
  • Is the value missing from the document?
  • Did the agent confuse two similar fields?

​
Add quality rules

Rules help the agent catch common problems. Examples:
  • “The invoice total must equal subtotal plus tax.”
  • “The due date must be after the invoice date.”
  • “Every line item must have a quantity and unit price.”
  • “Ask for review if the total is above $10,000.”

​
Improve over time

When you correct an extraction, turn that correction into a rule. For example, if a vendor writes their name in three different ways, teach the agent the final name your team wants to use. Learn more in Rules.