Document Intelligence

Extract structured data via natural language from uploaded documents and images

Feathery Document Intelligence automatically extracts structured data from your user-uploaded or provided documents. You first define questions using natural language. Then Feathery AI will read the document to answer those questions and store them within form or hidden fields. Finally, Feathery's AI results tab allows you to review the places in the document where the AI got the answers to each question from. The extracted data can be connected to your forms, workflows, and custom systems.

Note: this feature is only available on our Business plan. Reach out to book a demo to get access.

How to set up

  1. Create a form with the following fields:

    1. A file upload field (at the beginning of the form)

    2. All the fields you want to extract from your documents

  2. Go to the AI tab of your form and click "Create New Extraction". Give your extraction a name and select the file upload field from step 1 as the file source. You can leave "Correct Image Rotation" turned off unless you expect many of your inputted documents/images to not be oriented correctly (ex: turned sideways, upside down, etc)

  3. For each piece of data you want to extract from the document, you can add a new query which tells Feathery's AI what to extract. If no suitable value is found, the field will be left blank

  4. For each query, you can define the following components:

    1. Data Query: the natural language question you want answered. For example, "What is the business address of the company mentioned in this bank statement?". You can leave instructions such as "Only return XYZ".

    2. Frequency: this is how many values you want to extract for a specific query. One Value returns a single string. Multiple Values returns an array (for example, "return a list of all items purchased in this bank statement" would be a good use of Multiple Values)

      1. If you select multiple values, you will need to define a Page Query (which tells the AI which pages to focus on and which to ignore)

    3. Filter Pages (optional): for documents with many pages, you can enter a Page Query to tells the AI which pages to focus on and which to ignore. When Filter Pages is enabled, the AI will first filter for pages that match your query and then only use those identified pages to answer your query (ex: "only select pages that list out the transactions in a table format"

    4. Save To: which field to save the extracted value to (including hidden fields)

  5. In your form, add a button with the "AI Document Extraction" action and select the Extraction you just defined.

  6. You can optionally take the user to a step of your form with the extracted field values so that they can review before submitting

  7. After extractions are run, the results are available in the "AI" tab by clicking on the Extraction we just defined. You can click into each previously run extraction and view each document. Within the AI results UI, you can see where in the document the AI extracted values from and make edits to the extracted values

Last updated