sparklesAutomate extraction of agreement data with AI

Formality uses AI to automatically analyze your agreements and extract key information from them.

This automation is entirely based on the configuration of the Properties : categories, fields and AI settings. When well configured, they help reduce manual entry and provide reliable, directly usable data.


Prerequisites

Before enabling or optimizing AI analysis, make sure that:

  • your agreement categories are defined;

  • your sections and fields are configured;

  • you have Admin rights on your workspace.


How AI analyzes an agreement

When an agreement is imported or reprocessed, the AI works in two steps:

  1. Context understanding The AI identifies the final category of the agreement in order to understand what type of document it is.

  2. Targeted data extraction The AI extracts only the fields visible for that category, relying on their AI settings.

👉 The AI strictly follows the structure you defined: it analyzes no more and no less.


Configure a field for AI analysis

Each field has dedicated settings that allow you to refine extraction quality.

AI label (optional)

  • Used by the AI and the Assistant

  • Written in English

  • If not provided, the field’s default label is used


AI description

  • Serves as the primary guide for extraction

  • Written in English

  • Can be customized by category

👉 The clearer the description, the more reliable the extraction.


Field captured by AI

The option “To be captured by AI” indicates whether the field’s value should be automatically extracted from documents.

  • Enabled: the value is extracted automatically

  • Disabled: the value is entered manually

⚠️ Fields of type contact and user group cannot be extracted by AI.


Analyzed sources

You can define where the AI should look for information:

  • main contract ;

  • attachments ;

  • amendments.

Additional settings:

  • limit the number of sources analyzed;

  • set a display priority order.


Test and improve extraction

Each field can be tested to check extraction quality.

If the result is not satisfactory, you can:

  • adjust the AI label;

  • clarify the AI description;

  • modify the analyzed sources.


Result

A good AI configuration enables:

  • reducing manual entry;

  • obtaining consistent data;

  • improving the Assistant’s responses;

  • making alerts and workflows more reliable.


Last updated

Was this helpful?