SenseTask User Guide

View Categories

AI Copilot for data extraction

Contents

SenseTask AI Copilot enables advanced entity extraction and transformation using prompts. This guide explains how to configure Copilot to detect and process data from your documents.

Access the settings of the respective workflow folder Workflow folder settings.

  1. Select Labels settings > AI Copilot from the left-hand panel.
  2. Click on Add new AI Copilot prompt.

Scope #

Each prompt operates at one of three levels:

  • Label value – Targets a specific field value
  • Table row – Processes a single row in a table
  • Entire page – Uses the full page context

Example: To extract product codes from table lines, use the “Table row” scope.

Actions #

Copilot supports two actions: Detect and Transform.

Detect #

This identifies and extracts entities based on your prompt description.

To configure a detect action:

  1. Select the Scope (e.g. Table row)
  2. Choose Action: Detect
  3. Set Applies to → Select the field to populate
  4. Add a Prompt description that defines the expected pattern

Example #

Prompt: A string starting with 'W' followed by a number of digits.
Applies to: Product code

Best practices for Detect #

  • Provide clear examples
  • Include common variations and edge cases
  • Specify expected formats (e.g. date, code, etc.)

Transform #

This modifies an existing value based on rules in your prompt.

To configure a transform action:

  1. Choose Scope (typically Label value)
  2. Set Action: Transform
  3. Select a Source entity
  4. Write a prompt describing the format to apply

Example #

Prompt: Convert all dates to yyyy.mm.dd. If already correct, leave unchanged.

Best practices for Transform: #

  • Add input/output examples
  • Be explicit about edge cases
  • Specify how to handle already-correct values

Example setup #

To extract a product code from line items:

  • Scope: Table row
  • Action: Detect
  • Applies to: Product code
  • Prompt: A string starting with 'W' followed by a number of digits.

You can add additional prompts for related fields like batch series or expiration date.

Troubleshooting #

  • Make sure the scope and label match your document layout
  • Check that prompts are specific enough
  • Ensure label names are correctly configured
  • Re-test with a few samples to improve accuracy