Product Recommendations
Turn user–item interactions into ranked recommendations and campaign audiences. Items can be products, deals, donation campaigns, or other entities represented in your catalog.
What you can do
Recommend relevant items to each user.
Find items related to another item.
Find an audience for each item, vendor, or category.
Rank a user's vendor or category preferences.
Allocate one campaign audience between targets, with one assignment per customer and a separate recommendation ranking.
The distinction between audiences matters: independent lists per offer can overlap; allocation assigns each selected customer once. Neither an affinity score nor a revenue-based allocation weight is a conversion or revenue forecast.
How it works
An intelligence learns from your user, item, and interaction models. Recommendation scenarios reuse that trained model with their own scopes, objectives, delivery rules, and schedules. The interface calls these scenarios predictions.
Choose the business identity carefully: recommending a deal is not the same as recommending its vendor. Restrict sources to the population you want to serve and candidates to the items or users allowed in the result.
Results are published to your warehouse and exposed through DinMo models. Review publication, coverage, freshness, and evidence before activating them. Technical success, offline relevance, and campaign impact are separate checks.
Documentation guide
Technical prerequisites, required models, optional mappings, and data quality.
Output types, source and candidate segments, and item availability.
Overlap, unique assignments, weighted distribution, and companion rankings.
The main configuration choices, training, refresh, and delivery rules.
Coverage, evaluation metrics, score limitations, and commercial validation.
Output fields, activation checks, and keeping computation focused.
Last updated