Product Recommender
Learn about what the Product Recommender Agent does and when to use it.
The Product Recommender Agent analyzes a customer's unified profile, interactions, relationships, and contributing sources in Reltio to recommend three products with transparent probability scores and evidence-based reasoning.
Who is it for?
Business User
Data Product Owner
Solution Architect
For more information, see About roles.
Why would I use it?
Use the Product Recommender agent to generate accurate, explainable product suggestions based on a customer's behavior, preferences, and context — all mapped to your data model.
When and where would I use it?
Use this agent during digital engagements, support follow-ups, or campaign preparation. It runs in Reltio AgentFlow and leverages the MCP Server to analyze one customer at a time.
How does it support business goals?
It helps improve conversion, retention, and upsell effectiveness by aligning product recommendations with real signals from the customer's data and activity.
Core capabilities
- Analyze unified customer profiles across sources
- Retrieve and evaluate customer interactions
- Incorporate graph-based influence from relationships
- Score candidate products using a weighted model
- Produce explainable, probability-ranked recommendations
- Support filters by product category, interaction window, or ownership
- Respond with detailed data counts and source attribution on request
Inputs and outputs
| Inputs | Outputs |
|---|---|
Customer ID – Provide a unique customer identifier or full URI (for example, Search Filter – Use a query to find a customer by status, location, segment, or activity window. Interaction Window – Optionally specify a time frame such as last 30, 60, or 90 days to analyze engagement recency. Category Filter – Limit recommendations to products in a specific category or type. Ownership Exclusion – Choose whether to exclude products the customer already owns. Request Detail – Ask for the number of data points analyzed or for a breakdown of attributes, relationships, and sources. |
Customer Snapshot – Key identifiers, tier, segment, and recent interaction summary. Top Signals Used – List of the strongest signals influencing recommendation (attributes, interactions, relationships). Product Recommendations – A table with rank, product name, probability score, and evidence-based reasoning. Notes and Assumptions – Clarifications about data availability, confidence levels, or skipped products. Next Step Prompt – Suggestions for refining, analyzing another customer, or comparing alternatives. Tools Used – A transparent list of tools and endpoints used for data collection and scoring. |
Safeguards, permissions, and governance
- Recommendations are read-only and do not modify customer or product data
- All scoring and enrichment are based strictly on retrievable and explainable tenant data
- Access to customer records follows existing role-based permissions configured in your tenant
Limitations and edge cases
- Only one customer can be analyzed at a time
- Recommendations depend on the availability of recent interactions and relationships
- Products with missing eligibility or confidence signals may score lower or be excluded