AI Sandbox

Overview

The AI Sandbox allows you to precisely control which companies and content are shared with the AI. By defining scopes, request types, and output formats, you can extract insights, generate summaries, or retrieve structured information — all driven by your own prompt.

This guide walks you step by step through a typical workflow in the AI Sandbox.

Step 1: Select Companies

Once you land on the AI Sandbox page, start by selecting the companies you want to analyze.

You can choose companies in two ways:

  • Using the dropdown list to manually select one or multiple companies.
  • Selecting a predefined company list by clicking the dedicated icon.

You can select one or several companies depending on your analysis needs.

Step 2: Define the Report Scope

Next, choose the Report Scope using the dropdown menu.

This step defines which reports or documents from the selected companies will be included in the analysis.

Step 3: Select Content Scope

You now define what type of content the AI should analyze.

Option A: Content Search

You can choose between:

Regular search

Enter a specific keyword to search within the selected document.

When a user initiates a search:

  • The system scans all relevant documents from our database of 50,000+ reports, based on the filters applied (by default, or by the user).
  • Matches are located within fragments (also called chunks), which are sections of text or tables.
  • A highlighting mechanism visually indicates the occurrences of the search term within each fragment.

Unlike traditional search tools, Regular Search groups results by fragments:

  • A single fragment may contain one or multiple search matches.
  • Fragments can be:
    • Text sections
    • Tables

Example

If the term “TCFD” appears multiple times within the same paragraph, it will be returned as a single fragment, with all matches highlighted. The number of fragments returned may differ from the total number of individual keyword matches. This approach optimizes readability and contextual relevance.

(More information: LENS - Disclosures - Regular Search - Documentation)

AI search

Describe in natural language what you are looking for. The AI will retrieve the most relevant data based on your description. Unlike traditional keyword-based search, semantic search aims to understand the meaning behind a query and relies on natural language processing (NLP) and deep learning.

(More information: LENS - Disclosures - AI search - Documentation)

Initial Search in the Database

The user enters a query (e.g., greenhouse gas). The semantic search engine analyses the query and compares it to all reports in the database.

  • By default, filters (year, report type) are applied, and only matching documents are considered. Users can modify these filters and add other ones depending on the analysis.
  • The algorithm selects the top 300 results based on semantic relevance.
  • Once results are retrieved, they are grouped by company.
  • Each company may have multiple relevant fragments (results).
  • A company score is calculated based on:
    • The scores of the fragments.
    • The number of fragments found for the company.

(More information: LENS - Disclosures - AI search - Documentation)