6 Configuring AI Agents

Launch users have access to three agents: Offer Similarity Agent, Offer Integrity Agent, and Rule Integrity Agent. This chapter provides an overview of each agent’s capabilities.

Offer Similarity Detection AI Agent

The Offer Similarity Detection AI Agent is an AI-powered tool that helps you identify similar product offerings in your catalog. As you search, the system automatically checks for:
  • Product offerings with similar component structures and attributes
  • Offers that are similar in terms of their features, attributes, and other relevant factors

The goal is to quickly and accurately find similar offers so you can make informed decisions and manage your offers more efficiently.

How it Works

As you search for similar offers, the AI agent:
  • Reviews product offerings in real time
  • Compares them with other offers in the catalog based on component structure, characteristics, and attributes
  • Identifies similar offers and ranks them by similarity score
  • Provides a list of similar offers with explanations for their similarity
Each similar offer is clearly explained with:
  • What makes it similar
  • Why it is relevant
  • How you can use this information to make decisions

How to Invoke

The system can be triggered in two ways:
  • Schedule-based invocation: Used by the Similarity Ingestion Assistant, provided as a template for bulk ingestion and re-indexing.
  • Event-based invocation (use case–driven): Used by the Similarity Detection Assistant and the Similarity Cleanup Assistant, which are provided as seeded agents.

This enables the platform to support both continuous updates and on-demand comparisons.

Similarity Ingestion Assistant

Responsibility: Prepare offerings for similarity search

Flow:
  1. Retrieve eligible offerings from the Enterprise Catalog Product Offerings business object.
  2. Generate structured summaries using well-defined prompts.
  3. Write summaries to Vector Storage using the Vector Writer tool.
Key design points:
  • Separate summarization rules for Atomic vs. Bundled offers
  • Ensures consistent, high-quality embeddings
Similarity Detection Assistant

Responsibility: Find and rank similar offers

Flow:
  1. Read the summarized input (structured mode).
  2. Query Vector Storage via the Vector Reader tool.
  3. Retrieve candidate similar offers.
  4. Apply offer-type–specific scoring logic:
    • Atomic vs. Atomic
    • Bundle vs. Bundle
    • Weighted criteria with penalties

Output: Ranked list of similar offers with deterministic scores.

Similarity Cleanup Assistant

Responsibility: Maintain index consistency

Triggered on:
  • Offer deletion from the Launch application
  • Offer summary reset: Triggered directly through Agent Studio as described in Configuring the Similarity Cleanup Assistant
Actions:
  • Retrieve all related vectors
  • Clean up stale entries
  • Reset ingest status where required

This ensures no orphaned or outdated embeddings remain in the system.

Configuring the Agent

Configuring the Similarity Ingestion Assistant

To configure the Similarity Ingestion Assistant:
  1. Launch Fusion AI Agent Studio: Open the Fusion AI Agent Studio application.
  2. Search for the Similarity Ingestion Assistant Template: Search for the workflow template named “Similarity Ingestion Assistant.”
  3. Create a New Workflow: Click Copy template to create a new workflow from the template.
  4. Access Workflow Settings: Open the workflow settings by clicking the gear icon next to the workflow name.
  5. Configure Triggers: Go to the Triggers section.
  6. Schedule the Workflow: Configure the schedule:
    • Interval: minutes
    • Value: 15 (to run every 15 minutes)
    • Start time: Set the start time in UTC (For example, 23:00:00)
    • End time: Set a suitable end date and time in UTC
  7. Save and publish the workflow: Save the changes and publish the workflow.

Note:

The agents and other artifacts included within an agent template are optimized to provide the best usage of the agent team. We recommend not to change the basic functionality of the artifacts, because that might impact your agent team's performance.

Verification

Once the workflow is published, it will start ingesting records into vector storage at the scheduled start time. You can verify the workflow’s status and the ingestion process through Launch API services to ensure that it is working as expected.

By following these steps, you can successfully configure the Similarity Ingestion Assistant to ingest records into vector storage at regular intervals.

Configuring the Similarity Detection Assistant

No configuration steps are required, as the agent is seeded.

Configuring the Similarity Clean up Assistant

No configuration steps are required as the agent is seeded. Additionally, the Similarity Cleanup Assistant can be run from Studio to reset all offer summaries generated by the Similarity Ingestion Assistant, enabling summary regeneration and re-ingestion through the scheduled job. This will update or insert the current Vector DB index.

To reset summaries for all offers, run the Similarity Cleanup Assistant with isDelete parameter set to false and idCollection set to []. Use caution, as this action will require re-ingesting all offers.

Offer Integrity Advisor

Offer Design Integrity is an AI-powered assistant that helps you create offers right the first time.

While you design or update an offer, the system automatically checks for:
  • Missing or incorrect configurations
  • Conflicting products or services
  • Risks that may cause order failures, billing errors, or provisioning issues later

The goal is to prevent problems before the offer goes live, instead of fixing them later. Below is an overview of how this agent works:

As you design or revise an offer, the AI:
  1. Reviews the offer in real time
  2. Compares it against compatibility rules and existing offers
  3. Flags issues based on severity
  4. Suggests fixes or improvements
  5. Lets you know about similar offers that may be reused
Each issue is clearly explained with:
  • What the issue is
  • Why it matters
  • How to fix it
All findings are grouped into three easy-to-understand categories:
  • Critical Issues
    • Must be fixed before you can save or confirm the offer. (This validation can be turned off using the steps mentioned later.)
    • Can cause order failures or system errors

    Example: Incompatible products in a bundle

  • Warnings
    • Important issues that may cause problems later
    • You can proceed, but a review is strongly recommended
    Examples:

    • Offer start date is before the start date of its components

    • Offer end date is after the end date of its components

  • Recommendations
    • Best practices and improvement suggestions
    Examples:
    • Bundle starts with a zero price
    • Offer description mentions a device, but no device is included

How to use the agent

You can configure the agent (through business configurations) to run automatically before saving or confirming an offer. If automatic checks are enabled, a successful integrity check is required to save or confirm the offer. You can also invoke the agent manually by clicking the Check Integrity button.

Rule Integrity Advisor

Rule Design Integrity is an AI-powered assistant that helps you create product rules right the first time. While you design or update a product rule, the system automatically checks for:
  • Missing or incorrect configurations
  • Conflicting rules
  • Risks that may cause order failures, billing errors, or provisioning issues later

The goal is to prevent problems before the rule goes live, instead of fixing them later. Below is an overview of how this agent works:

As you design or revise an offer, the AI:
  1. Reviews the rule in real time
  2. Compares it with other product rules in the catalog
  3. Flags issues by severity
  4. Suggests fixes and improvements
Each issue is clearly explained with:
  • What the issue is
  • Why it matters
  • How to fix it
All findings are organized into three easy-to-understand categories:
  • Critical Issues
    • Must be fixed before you can save or confirm the rule.

      Note:

      This validation can be turned off using the steps described later.
    • Can cause order failures or system errors.
    Examples:
    • A Compatibility rule states that Product A and Product B exclude each other, but a Recommendation rule already marks them as cross-sell recommendations.
    • A Recommendation rule marks Product A and Product B as cross-sell recommendations, but a Compatibility rule already states they exclude each other.
  • Recommendations
    • Best practices and improvement suggestions.

How to use the agent

You can configure the agent (through business configurations) to run automatically before saving or confirming a product rule. If automatic checks are enabled, a successful integrity check is required to save or confirm the rule. You can also invoke the agent manually by clicking the Check Integrity button.

Business Configurations For Integrity Agents

Both the Offer Integrity Agent and the Rule Integrity Agent can be configured (through business configurations) to run automatically before saving or confirming. This is disabled by default. You can enable this in the business configurations. For more information, see Set Up Business Configuration Values.

Creating Your Own AI Agents

Fusion AI Agent Studio enables you to create custom AI agents. For more information, see How do I use AI Agent Studio?

An additional subscription may be required to create custom agents. For more information, see When do I need a Custom AI Agent Subscription for AI Agent Studio?