MXP Platform

Serving Configuration

Named configurations that define how search, browse, and recommendation results are served in Lightweight mode

Serving Configuration is available in Lightweight mode only. In the full platform, the equivalent settings live in Global Configuration.

What it solves

Search and recommendation behavior has to be described somewhere: which facets are generated dynamically, whether results are personalized, which recommendation model serves a placement, how much price and variety influence the order of results.

In the full platform that description is spread across Global Configuration records and Target Groups. In Lightweight mode it lives in a single object — a Serving Configuration — and every other feature points at it. Rules, linguistic overrides, and evaluation environments are all scoped to a serving configuration, so one screen answers the question "how are results served here?"

When to use it

  • Setting up a new surface — create a configuration for a search and browse experience, or for a recommendation placement
  • Changing how results are ordered — adjust price reranking or diversification without touching individual rules
  • Switching the model behind a placement — point a recommendation configuration at a different model
  • Scoping a rule or an override — create the configuration first, then reference it from discovery rules and linguistic overrides
  • Enabling or disabling personalization or dynamic facets for a search and browse surface

Key concepts

Serving Configuration — a named record describing how one class of results is served. A deployment can hold several, each identified by its Config ID.

Product — whether the configuration serves Search and Browse results or Recommendation results. This choice determines which of the remaining fields apply.

Config ID — the identifier other features use to reference the configuration. Discovery rule lists, linguistic override lists, and evaluation environments can all be filtered or scoped by Serving Config ID.

How it works

Open Serving Configuration from the navigation menu. The page lists the existing configurations and lets you create, edit, and delete them.

The form fields are:

FieldApplies toWhat it sets
NameAllThe display name of the configuration
Config IDAllThe identifier other features use to reference this configuration
ProductAllSearch and Browse, or Recommendation
Dynamic FacetsSearch and BrowseThe mode used for dynamically generated facets
PersonalizationSearch and BrowseThe personalization mode applied to results
Recommendation ModelRecommendationWhich recommendation model serves this configuration
Price RerankingRecommendationHow strongly price influences result order: None, Low, Medium, or High
Diversification TypeRecommendationHow variety is introduced into the result set
Diversification LevelRecommendationHow strongly diversification is applied

The Product selection drives the rest of the form. Choosing Search and Browse shows the Dynamic Facets and Personalization fields. Choosing Recommendation shows the Recommendation Model selector and the preferences that shape recommendation output — price reranking, diversification type, and diversification level.

Target Group assignment is not part of the Serving Configuration form in Lightweight mode. Audience segmentation is a full-platform capability — see Configurations.

Quick example

A retailer wants a "customers also viewed" carousel on product detail pages, showing varied results rather than several near-identical items, and without letting price dominate the order.

  1. Open Serving Configuration and create a new configuration.
  2. Set Name to pdp_also_viewed and give it a Config ID the storefront will request.
  3. Set Product to Recommendation.
  4. Select the Customers Also Viewed model in Recommendation Model.
  5. Set Price Reranking to Low so price nudges the order without driving it.
  6. Set Diversification Type and Diversification Level so the carousel spreads across products rather than repeating variants.
  7. Save the configuration.

From this point, any recommendation rule or evaluation environment can be scoped to pdp_also_viewed, and the carousel is served according to these settings without further changes.