A/B tests

Module experiments distributes visitors between options, applies option to a public page and counts impressions and conversions. The appointment is stable: the visitor receives one experiment option for the entire cookie period.

Goals of the experiment

Goal	Target Key	Behavior
API or tag	Not required	The variant receives its own component or tag `[mod_experiment:code]`.
Regular block	Block ID	The module automatically wraps the block result.
System unit	System unit ID	The module automatically wraps the result of the system block.
CSS selector	For example, `#buy-button`	The option is applied to found DOM elements.
Price	CSS price selector	Only the displayed price text changes.

The Price mode does not change the price of the product, cart, order or payment. For the test calculation price requires a separate project processor experiments.assignment.resolved and server verification at all stages of the order.

Options

The experiment contains at least two options. Weight specifies the relative proportion, and coverage - the percentage of visitors participating in the test. Payload option supports:

control - leave the original content;
replace - replace the contents of the target;
before and after - insert HTML next to the target;
replacing text, adding a CSS class and links;
displayed price value.

You can leave the experiment as a draft, run it, pause it, or end it. The start and end period are optional.

Inserting into a template

[mod_experiment:promo_button]

A stable slot is created in place of the tag. The option is loaded upon receipt pages, so a common full-page cache does not mix variants of different visitors.

To automatically record a conversion, add an attribute to a button or link:

<button data-experiment-conversion="promo_button">Отправить</button>

An arbitrary event is specified by the second attribute:

<a data-experiment-conversion="promo_button"
   data-experiment-event="checkout_started">Оформить</a>

From JavaScript, the event can be sent manually:

window.AVEExperiments.convert('promo_button', 'form_sent');

Public API

GET  /api/v1/experiments/{code}
GET  /api/v1/experiments/assignments?codes=promo_button,hero_title
POST /api/v1/experiments/event

The last endpoint accepts code and event. The variant ID from the client is not accepted: the server recalculates the destination and writes the event only to this option. For one visitor, the uniqueness of an event is considered once a day, however, the total number of events is also maintained.

##Hooks

###experiments.assignment.resolved

Called after the server has selected an option. Available in context:

experiment - complete record of the experiment;
variant - selected option;
visitor_hash - anonymized hash of the visitor.

The result of the event is the destination array. The handler can return a modified array, for example for design integration with a service layer.

###experiments.event.recorded

Called after an event is recorded. The context contains experiment, variant, event_code and flag unique.

Cache and privacy

In HTML, only the same slot and active configuration are cached for all goals. The assignment is requested by the client via the API. Random seed is stored for a year in cookie ave_experiment_seed with HttpOnly, SameSite=Lax and Secure on HTTPS. IP and original seed are not saved in statistics.

Permissions and deletion

view_experiments - viewing experiments and results;
manage_experiments - creation, launch, statistics reset and deletion.

Uninstallation removes experiments, variants, statistics, deduplication visitors and module settings.

3.8 KiB Raw Blame History