For the complete documentation index, see llms.txt. This page is also available as Markdown.

A/B-Tests

A/B tests compare your base configuration against one test variant in live traffic.

They help you validate changes before rollout.

Resource usage and number of tests

  • Thesaurus

  • Stopwords

  • Preprocessor Entries

  • Trash Can Entries

  • Campaigns

  • Search Algorithm Settings

  • Result listing Settings

  • Personalization (except for (de)activating the module)

  • Semantic Enhancer

  • Facets & ASN: Filters

  • Single Word Search

  • AtlasAI

  • Ranking Rules, but with the following exceptions:

    • Ranking Rules are only A/B testable for the scopes 'Search' and 'Navigation', not for the scope 'Suggest'.

    • The ranking for the suggest module will always use the Ranking Rules from the base variant and never apply changes from the B variant.

Regular A/B tests require one additional in-memory copy of the configuration per test.

Ranking rule tests require one additional ranked product list.

You can create multiple A/B tests.

More parallel tests need more traffic.

That means each test usually needs longer to reach reliable results.

Support overview

Fully supported

These features support A/B testing without known module-specific limits:

  • Thesaurus

  • Stop Words

  • Preprocessor entries

  • Trash entries

  • Campaigns

  • Result Listing

  • Semantic Enhancer

  • Single Word Search

  • Atlas AI

  • Vector Search

  • Filters

Limited support

These features support A/B testing with restrictions:

  • Personalization

    • Personalization settings are supported

    • Activating or deactivating the module is not supported

  • Ranking rules

    • Supported for Search and Navigation

    • Not supported for Suggest

    • Suggest always uses the base variant

  • GPT Synonyms

    • You can activate or deactivate the module

    • You can choose whether to apply pending synonyms

    • You can add, modify, and delete synonyms

    • You can access synonym generation in the Cockpit

    • Other general settings on that page stay disabled

  • Recommendations

    • Manual recommendations are supported

    • Other recommendation types are not supported

    • Matrix edits are not supported

  • Search Algorithm

    • Standard search settings are supported

    • Exact Search support is still being verified

Not supported

These features do not support A/B testing:

  • ASO

  • Field settings

  • Geo

  • CSI

  • Suggest

  • Import and database settings

Important note

Send the sid parameter with every request.

sid contains the session ID that assigns each request to variant A or B.

Without sid, requests cannot be assigned reliably.

Last updated

Was this helpful?