Direqt Analytics for AdTech¶

Overview¶

Direqt chatbots can be deployed on websites using traditional ad server technology. In this scenario, the chatbot is effectively an advertising creative, and like any creative it is useful to be able to measure the performance of the chatbot using standard adtech concepts in order to optimize for specific campaign objectives. For example, some campaigns may wish to optimize for initial clickthrough rate (CTR), whereas others may favor overall interactivity (session length).

Direqt Analytics for AdTech provides detailed information about viewable impressions, in-chat ad impressions, and various forms of user interactions that together provide valuable insight for optimizing.

Concepts¶

Analytics is based on a flow of events. Each event has a common set of metadata that is used to associate the events with a particular user, session, or chatbot.

In analytics, users are assigned an opaque and anonymous userId. Users on different websites will have different user ids. For a given user, we assign a unique session id that is shared with all events that are generated by the user during a period of activity.

Every chatbot is assigned a specific botid that is typically unique to a particular website. Every chatbot is also part of exactly one network, which is an organizational construct useful for grouping bots. These values are assigned by Direqt and are immutable.

Almost all of the events that are tracked will include these identifiers - userId, sessionId, botId, network - as metadata, alone with additional fields specific to their event type.

Similar to traditional display advertising, Direqt Analytics for AdTech supports events that track impressions, which fire when a chatbot first becomes visible to an end user, and interactions, which fire when a user interacts with the chatbot.

In addition, there are events that track in-chat impressions. These events are used to track the display of traditional display advertising units that are embedded within the chatbot session itself.

Metadata associated with these events can be used to distinguish between various types of user interactions.

Note

A single session can have multiple impressions (e.g., having multiple tabs open with Direqt loaded).

Note

A single impression can span multiple sessions if the user remains on the page but is inactive for 30+ minutes.

Event Types¶

Every event supported by the system as an event type that identifies its purpose.

Event type	Description
impression	The chatbot as visible to the user according to IAB definition.
inChatImpression	The chatbot has presented an embedded display advertisement to the user.
interaction	The user has interacted with the chatbot in a meaningful way.

The following sections give details about each of these events, when they are fired, and the metadata they contain.

Impression¶

On page load, a chatbot may be loaded into a host website by the ad server. If and when the chatbot becomes viewable according to IAB viewability standards, a viewable impression ('impression' event) is recorded. This is useful in conjunction with interaction events in order to determine what percentage of time a user chose to interact with the chatbot.

An impression is recorded when at least 50% of chatbot is in view for at least one second.

At most one impression is recorded each time a chatbot is loaded on a page. So for example, even if a chatbot is initially visible but then scrolled out of the viewport and then visible again will only record a single impression.

The metadata associated with each viewable impression includes:

field	definition
`impressionId`	Unique identifier for this impression
`botid`	Unique identifier for the bot
`network`	Unique identifier for the owning network
`userId`	Opaque site specific identifier for the user
`sessionId`	Unique identifier for the user's current session

A new impressionId is generated upon each load of the Direqt script.

Interaction¶

Definition:

tracks meaningful interactions initiated by the user to interact with a chatbot. This could include clicking link, composing a message, or tapping on a quick reply. Trivial interactions such as scrolling, clicking on non-interactive elements, or closing the window are not tracked.

Fields:

botid (string): Unique identifier for the bot
network (string): Unique identifier for the owning network
userId (string): Opaque site specific identifier for the user
sessionId (string): Unique identifier for the user's current session
impressionId (string): Unique identifier for the initial impression/view of the widget
interactionId (string): Unique identifier for the interaction
interactionType (action or message): Top-level classification
interactionSubType (click, compose, quickReply, prompt): Subclass of user interaction
inchatImpressionId (string, optional): Used to link to an in-chat ad view if applicable

Interaction Types:

Action: User clicks an element (e.g., opening a link from a chat message)
Message: User sends a pre-generated message (e.g., clicking a suggested reply or voting on a poll) or composed their own message via the chat/search bar.

Interaction Subtypes:

click: Associated with action type
compose: User-created message
quickReply: User responds with a pre-generated response (e.g., suggested reply, poll vote)
prompt: Sends a pre-generated message to start a new conversation/topic (e.g., promo prompt)

In Chat Impression¶

Definition: Viewable ad impressions inside a chat session. This includes both paid external links and organic links within the publisher's own domain.

Fields:

botid (string): Unique identifier for the bot
network (string): Unique identifier for the owning network
userId (string): Opaque site specific identifier for the user
sessionId (string): Unique identifier for the user's current session
impressionId (string): Unique identifier for the initial impression/view of the widget
inChatImpressionId (string): Unique identifier for the in-chat impression
inChatImpressionType (paid or organic): Ad type

Notes:

Story carousels are given a single impressionId.

In Chat Impression Types:

Paid: For paid external links
Organic: For links within the publisher's own domain (e.g., recommended stories)