Conversational agent event tracking

Event tracking helps you understand how users interact with Conversational agent. Each interaction provides insights into engagement, performance, and potential issues. This article outlines the events Conversational agent captures.

Overview

Few core mechanics influence how Conversational agent tracks events and attributes:

Two-level A/B testing: Conversational agent runs A/B tests at two levels. The global level determines whether a shopper sees Conversational agent or is assigned to a control group. The use-case level decides which specific use case a shopper sees.
Third-party A/B testing: Conversational agent supports third-party tools for variant assignment, including tests with 3 or more variants. For example, Conversational agent versus a test experience versus a control group.
Traffic throttling: Traffic throttling lets Conversational agent limit the amount of site traffic entering a Conversational agent experiment without changing the variant split. This supports gradual rollouts while keeping the test setup consistent.

Event types

Conversational agent events are grouped as:

A/B test events: Impression-level events that support variant analysis and measure Conversational agent's overall performance.
[Interaction events](doc:interaction-events]: Shopper behavior inside the chat widget, plus Conversational agent's responses. Analyze how shoppers engage with the Conversational agent interface.
Safety events: Errors and performance issues, so you can detect and resolve problems fast.

To build event reporting, see Conversational agent analytics and Conversational agent dashboard metrics.

Default event attributes

Every shopper event is also enriched with browser, device, operating system, and geolocation fields (browser, device, os, city, country, state, latitude, longitude, ip, and similar).

Attribute	Description	Values
`action`	The event action type. Identifies the event.	One of the action values documented on this page (for example, `use_case_session_start`, `user_respond`)
`agent_id`	The conversational agent configured for the account.	Example: `0199a817-49f6-73f3-8769-12e853567c30`
`banner_id`	Unique identifier for the widget.	Example: `68e02f017fe069386bb3073e`
`banner_name`	Display name of the widget.	Example: `Clarity-Embedded-Chat`
`banner_type`	Type classification of the widget.	Example: `medium_rectangle`
`chat_id`	Unique identifier for the chat session.	Example: `c123456789`
`global_variant_id`	Identifier for the global A/B test variant.	`0`, `1`
`global_variant_name`	Name of the global A/B test variant.	Example: `Clarity`
`variant_id`	Identifier for the use-case-level variant.	`0`, `1`
`variant_name`	Name of the use-case-level variant.	Example: `Variant A`
`variant_origin`	Source of the variant assignment.	`ABtest`, `contextual personalisation`
`experiment_id`	Identifier of the experiment when a third-party A/B testing tool provides the variant.	Example: `584sa5729971a4f992sj9`
`experiment_source`	Source of the third-party experiment.	Example: `bloomreach`
`global_rank`	A/B split rank used to determine control-group membership.	Integer 0–99
`throttle_rank`	The throttle bucket used to limit traffic exposure.	Integer 0–99
`clarity_session_state`	The shopper's current session state with respect to Conversational agent.	`unseen`, `seen`, `interacted`
`clarity_session_interaction_count`	Number of interactions in the current session.	Integer; `-1` before the first show
`path`	Page URL.	Example: `https://www.example.com/products/47748`
`interaction`	Whether the event resulted from a shopper interaction.	`true`, `false`

Non-Conversational agent events

To get a complete view of Conversational agent’s impact, track relevant external custom events:

Required events: cart_update, purchase, view_item
Recommended events: checkout, view_category