Skip to content

Agentic Search

Preview feature

Agentic search is currently a Preview feature. It requires feature enablement and may change in future releases.

Agentic search is an AI-powered fallback mechanism that uses the Bizzkit AI Assistant (BAIA) to intelligently reason about user search intent. When standard keyword-based search returns zero results, agentic search engages an AI agent that interprets the user's natural language query and identifies relevant products through reasoning — not just keyword matching.

While both features use AI to improve search results, they serve different purposes:

Semantic Search Agentic Search
Approach Embeds search phrases into vector space to find semantically similar products Uses an AI agent to reason about intent and suggest specific products
When it activates For top phrases or zero-result phrases (configurable per experience) As a fallback when standard search returns zero results
Output Boosts semantically similar products in results Pins specific AI-suggested product IDs into results
Scope Can find semantically similar products beyond keyword matches Reasons about complex intent to identify specific relevant products

Agentic search complements semantic search — both can be enabled simultaneously to provide multiple layers of intelligent search enhancement.

When agentic search activates

Agentic search acts as a fallback and is triggered when all of the following conditions are met:

  1. The search request has agenticSearch set to true
  2. The search phrase is non-empty
  3. No automatic Did You Mean correction was applied

If these conditions are met and the initial search returns zero results, the system engages BAIA to find relevant products through AI reasoning.

How it works

The agentic search flow operates as follows:

  1. User searches — A search request arrives with agenticSearch: true
  2. Standard search executes — ECS performs the normal keyword-based search
  3. Fallback evaluation — If results are zero, the agentic search fallback is triggered
  4. AI reasoning — ECS sends the query to BAIA, which reasons about the user's intent using configured context and product data
  5. Product suggestions — BAIA returns a set of recommended product IDs
  6. Pinned results — The suggested products are pinned into the search results and returned to the user
Hold "Ctrl" to enable pan & zoom 
flowchart LR
    A[User Search] --> B[Standard Search]
    B --> C{Zero results?}
    C -->|No| D[Return results]
    C -->|Yes| E[BAIA AI Reasoning]
    E --> F[Suggested Products]
    F --> G[Pin into results]
    G --> D

Product exclusions

Product exclusions allow you to prevent specific products from being suggested by agentic search. This is useful for:

  • Products that should not appear in AI-suggested results (e.g., discontinued items)
  • Products that are frequently suggested incorrectly
  • Sensitive or restricted products

Exclusions are managed per segment through the Admin Preview API.

Field exclusions

Field exclusions allow you to exclude specific field values from the context that BAIA uses when reasoning about products. This helps you:

  • Remove irrelevant or noisy data from AI reasoning
  • Prevent certain attribute values from influencing suggestions
  • Fine-tune what information BAIA considers when matching products to queries

Like product exclusions, field exclusions are managed per segment through the Admin Preview API.

Relationship to other AI features

Agentic search is part of the broader AI capabilities in Ecommerce Search:

  • Semantic Search — Enhances results by understanding meaning; agentic search goes further by actively reasoning about intent
  • AI Related Tags — Generates context-aware suggestion tags; complementary to agentic search
  • Conversational Filtering — Guides users through interactive questions; operates independently of agentic search

All these features use BAIA as the underlying AI service and can be enabled together for a comprehensive AI-enhanced search experience.