20.1 - Fredhopper Access Server Release Notes

Introduction

These release notes provide an overview of the resolved issues and improvements
of the most recent release of the FAS 20.1 line.

FAS 20.1 supports Oracle Java 8 VMs. For detailed setup instructions please
refer to the Fredhopper Learning Center.

Version 20.1.1

Release Date: February 28, 2020

FAS 20.1.1 is a maintenance release that improves the structure of breadcrumbs
in the FAS response, and the handling of missing conflated attributes.

Improvements & Resolved Issues

[Fixed] The internal structure of the breadcrumbs data in the response has
been updated to satisfy integrators requirements.

[Improved] In case FAS conflates to a missing attribute, then the default
value (for the particular strategy selected) is used as a conflation value.

Version 20.1.0

Release Date: January 22, 2020

FAS 20.1 is a major release. The Query and Data APIs in this version are
backward compatible with versions FAS 7.5.x, 8.x.y. and 19.x.y. The
Configuration API is backward compatible with FAS 8.x.y and 19.x.y.

The headline feature of the release is the Result modifications A/B testing
functionality.

For a complete list of features, improvements, and fixes, see below.

A/B testing of Result modifications

• From the A/B tests page in the Merchandising studio, you can create an A/B
  test for item campaigns, ranking rules or result modifications. You need
  to select the test of type as the first step after clicking the "New" button.

Note: to use A/B testing, your platform needs to integrate with Fredhopper
Insights.
Note: The A/B testing is not supported for Location replacements.

• You can create an A/B tests for two or more (recommend - up to four)
  alternative result modifications.You can allocate the traffic among the rankings being tested. Each alternative ranking can receive 10% to 90% of traffic.
• The modifications you set up in the A/B test will also appear on the result modifications page. However, you can only manage their content via the A/B test page. On the Result modification page, you can change the priority of an A/B tested modification, among other modifications, which would be shown on a page.
• It is also possible to A/B test one or more result modifications against not having that result modification at all. To do that you need to specify 'Useunderlying ranking' for the first case of an A/B test.

Note: In this situation, for all users that get the first variant in a response, no result modification related to this A/B test will be applied. However, other result modifications triggered for that request, will be
applied.

• Publishing of A/B tested result modifications works the same way as publishing of A/B tested rankings or item campaigns.
• After an A/B test is published, you will be able to analyse insights about the test on the Fredhopper Insights page. This analysis will enable you to pick a winning test. Please refer to the Fredhopper Insights release notes for details.
• Once a winning option has been identified, you can convert the test to a stand-alone result modification. Once selected, the A/B test along with all its modifications will be deleted, and the winning modification will appear as a new entry on the Ranking rules page.

Note: You will need to publish the deletion of the A/B test and creation of a new modification.
Note: If you have been testing result modifications versus an underlying ranking, and selected the underlying ranking as a winning case, the A/B test will be deleted. No additional rules will be created, because underlying rankings for that A/B test already exist in the system.

• Just like any other rule in the Merchandising Studio, you can edit, delete, or restore an A/B test. Please note that changing the content of campaigns and re-publishing an A/B test may lead to distortion of the results of the test.

Improvements & Resolved Issues

Merchandising Studio & Preview Pages

[New] You can name variants in an A/B test. By default, the first variant is
named 'Case A', the second - 'Case B', etc. However, you can specify any
other name. Test names can be up to 20 characters, and may only contain
Latin letters, numbers, spaces, and underscores. Variants within an A/B test
must be unique.

[Improvement] Alternative experimental algorithm for ordering of entities, e.g.
campaigns and facets, is available behind a feature toggle, off by default.

Known Issues & Limitations

On the pre-published environment, all cases from an A/B test result modification,
will be applied to a response simultaneously.