SWOT Agent (config)
5 min read
This tool is available on request only. Please reach out to your Unique contact for more information. The SWOT agent is part of the following space:
Functionality
The SWOT Agent is an advanced AI-powered research assistant, that performs in-depth analysis and investigation across multiple sources to provide an equity analyst / investor a comprehensive overview about a company focusing on on the areas Strength, Weaknesses,Threats and Opportunities. This is usually analysis jobs running 15-30 minutes and reviewing multiple third party documents and external sources such as Earnings Calls from Quartr.
Use Case
Not recommended for:
Simple factual queries (use standard chat)
Real-time information updates (though web search provides current data)
Questions answerable from a single source
Activation - Feature flags and environment variables
A Quartr API license is required to use the SWOT Agent - please contact your Customer Success Manager as this can be provided via Unique.
Variable | Description |
|---|---|
| Required to have the SWOT agent available in the tool selection. |
Tool Configuring
SWOT Agent is a tool available for Unique AI Chat and will, once enabled, appear in the tool section of Unique AI Space
You can click on the SWOT Analysis tool to update this configuration
By default the tool is set with is exclusive and is OnByDefault. This need to achieve the desired behavior of the SWOT agent. Other than this, the tool follows the normal tool patterns Tools
Configuration Overview
We provide a wide range of configuration options that allow you to customize the behavior of the SWOT generation logic. In the following sections, we present a high-level overview of the purpose of each configuration block, without going into technical details. While modifying these configurations requires technical expertise, this overview is intended to help you understand what aspects of the system are configurable and what can be adjusted if needed.
Source Management
The Source Management configuration is divided into three distinct subcategories, each addressing a different aspect of source handling.
Earnings Call Configuration
This configuration is responsible for rendering and uploading earnings call transcripts obtained from the Quartr API.
The Upload Scope ID is a mandatory configuration field. You can either use an existing scope or create a new one. If the upload scope is not configured, the tool will not be able to use earnings call data. In addition, this field ensures that documents are not fetched and downloaded from Quartr if they have already been ingested.
Any user running the SWOT must have read and write access to the upload scope specified by the Upload Scope ID.
The Docx Renderer allows you to specify the contentId of the DOCX template used to render earnings call transcripts. The template must already be ingested in the system. If no template is configured, a default template will be used.
Quartr does not provide the transcript document directly; instead, it returns a JSON object, which is used to render the transcript before it is ingested into the knowledge base.
Read-only access is sufficient for the scope where the DOCX template is located.
Source Iteration Configuration
The Source Iteration configuration is used by an agent responsible for determining the order in which the selected sources are processed. It works as follows: the agent examines the first Max Number of Selected Chunks from each document and reorganizes them based on a defined objective.
By default, the configuration considers 5 chunks per source, and the objective is recency, ordered from oldest to newest. Both of these parameters can be easily adjusted.
Additional prompt configurations exist, but they are not shown here in order to keep the documentation accessible to all readers.
Source Selection Configuration
In a similar way to Source Iteration, the Source Selection configuration is used by an agent whose goal is to determine whether a selected document is relevant for SWOT analysis. The agent uses the company ticker and name, and examines the Max Number of Selected Chunks from the document to assess whether it should be considered for SWOT generation.
This helps prevent cases where documents are accidentally selected or where content unrelated to the company is incorrectly added to the processing pipeline.
You may increase the number of selected chunks, especially when the first pages of a document contain general or introductory information. As with other configurations, the prompts can be adjusted, but for simplicity they are not covered here.
Report Generation
The Report Generation Config is responsible for extracting SWOT facts from the set of documents you provide (after being sorted by the Source Iteration Agent and passed the check of the Source Selection Agent). For each SWOT component, the workflow follows these steps:
Fact Extraction
A first agent reviews the documents and extracts facts relevant to the specific SWOT component.Analysis and Planning
A second agent analyzes the newly extracted facts, cross-checks them against the facts already extracted and organized, and plans the necessary update steps. This ensures that information is not duplicated and that facts are updated consistently, following the order of the documents.Execution
The multi-agent system is then triggered to execute the plan defined in the previous step.
While updating the Commands, Extraction, and Planning prompts requires technical expertise, you can still review the Definition prompts. These prompts describe what each SWOT component represents and help the model determine which facts are relevant to extract.
Report Summarization
After the SWOT facts have been extracted and organized by the Report Generation agent, a separate Summarization agent is used to synthesize all the collected information into a shorter report. The goal of this summary is to provide the user with a quick, high-level overview, including an introduction to the company.
This summary not only highlights the key insights but also makes the full report easier to read and helps analysts better understand the company.
In the future, we plan to enrich this summarization step with web search capabilities, allowing the system to collect additional information about the company and the technologies it uses. This enhancement will further support analysts by providing broader context.
The configuration allows you to choose which LLM is used for summary generation and also provides the option to customize the prompts.
Report Renderer
The Report Renderer uses a Jinja templates to aggregate all the extracted facts from the different SWOT components, along with the summary, and organizes them into a structured DOCX report.
If you want to customize your docx report, you can provide the template contentId of your DOCX template. This allows the renderer to use your custom layout and formatting styles for the final report.
Example Docx template
You can further customize this template and upload it to your knowledge base. Once uploaded, use the generated contentId as the template_content_id in the DOCX renderer configuration for both the earnings call and report renderer configurations, enabling you to generate outputs with your own theme.