Discovery Search with Clarity
This article describes how Discovery Search integrates with Clarity's agentic architecture.
What is Discovery Search
Bloomreach Discovery Search is an AI-powered commerce-optimized engine that delivers highly personalized, fast, and relevant product search experiences. Discovery Search provides full merchandising control for complex strategies. Using Clarity with Discovery Search allows you to maintain your existing merchandising strategy while enabling conversational product discovery.
Note
Discovery Search is only available to customers who purchased Bloomreach Discovery.
How Discovery Search works with Clarity
Clarity uses Discovery Search with the agentic architecture, which involves the following steps:
1. Intent detection and query formulation
The agent identifies the search intent from the user's input and forms a precise query.
2. Filter extraction
Filtering extracts attributes to further refine the search query.
-
The agent extracts product attribute values for descriptive attributes (say, color and brand) and boolean attributes (such as on_sale). For example, from "Show me red sneakers on sale," the agent extracts color=red and on_sale=true.
-
For numeric attributes (such as price), the agent extracts filter and sorting conditions. For example, for "Show me dresses under $50," the agent extracts price < 50.
3. Discovery API call
The agent sends the refined query to the Discovery Search API, which retrieves results based on all configured Discovery rules and customizations.
4. Relevance filtering
The agent performs relevance filtering to ensure that the retrieved product results align with the conversation's context.
For example, if a shopper initially asks to see dresses and then requests "I want to see more results," the agent retrieves additional results from the Discovery API call. It then filters out any dresses already displayed in the conversation to prevent showing duplicate items.
5. Results display
The shopper finally sees filtered results within the Clarity chat interface.
Updated about 4 hours ago