Skip to main content

Document Insights

You have 50 vendor contracts and one question: which ones auto-renew next quarter? Reading all 50 takes a day. Document Insights reads them for you and puts the answer in a table.

You tell Docana what to pull out of each document. It reads every document, extracts those fields, and lays the results out as a grid: one row per document, one column per field. What was 50 PDFs becomes a spreadsheet you can sort, filter, and ask questions about.

How it works​

Create a Document Insights application by selecting Documents Insights in the type picker. Open it, create a new insight with a name and instructions, then build the table in three steps.

1. Define your columns​

A column is one thing you want from every document. Click Add Column and give it:

  • Name: what to call it, like "Renewal Date" or "Payment Terms"
  • Prompt: what to extract, written as an instruction, like "The date the contract automatically renews"

Add as many columns as you need. Each one becomes a column in the final table.

Define a column with a name and a prompt that tells the agent what to extract.
Define a column with a name and a prompt that tells the agent what to extract.

2. Add your documents​

Click Add Documents and pick the files or collections to analyze. Each document becomes a row, and analysis starts for the new rows.

3. Generate​

Wait for analysis to finish, or use Generate Insights to run it again. Docana reads each document, pulls out every column you defined, and fills in the grid. A 50-contract, 4-field table is 200 values it extracts for you.

Two articles appear as rows under the Main topic column. Analysis is pending in this example; extracted values appear when processing finishes.
Two articles appear as rows under the Main topic column. Analysis is pending in this example; extracted values appear when processing finishes.

Work with the results​

Once the table is filled in, you can:

  • Sort and filter by any column to find what matters, like every contract renewing this quarter
  • Chat with it: open the playground and ask questions in plain English, like "which three contracts have the highest liability cap?"
  • Download it as a CSV, or save it to a collection

Turn insights into a database​

The table you get is structured data, so it works like any other dataset. Save it to a collection and an agent can query it with SQL, the same way it queries an uploaded spreadsheet. Ask "what's the average contract value by vendor?" and the agent runs a real query against the extracted numbers.

So Document Insights closes the loop: unstructured documents go in, a queryable table comes out.

Example uses​

  • Contracts: pull renewal dates, payment terms, and liability caps from every agreement, then sort by what's due soon
  • Resumes: extract years of experience, key skills, and certifications across a stack of applicants
  • Invoices: pull vendor, amount, and due date from each invoice into one table
  • Research: extract the method, sample size, and finding from a folder of papers

Best practices​

  1. Write prompts like instructions: "The total contract value in USD" extracts better than "value".
  2. One fact per column: separate "Renewal Date" and "Payment Terms" rather than one "Key Terms" column.
  3. Start small: define your columns on a handful of documents, check the results, then run the full set.
  4. Save it to query later: keep the output in a collection so an agent can answer number questions about it.

Next steps​