Qubit’s attribution model

This is for:

Developer

Intro

Qubit’s attribution model for post-experience segmentation aligns closely with the logic in our stats engine:

  • We distinguish visitors and converters by their device. The same user on multiple devices is counted multiple times; multiple users on the same device are confounded. Also, we count visitors and converters as unique within an experience iteration–as opposed to counting conversions, or unique visitors / converters over all time. If a visitor converts more than once in that timespan, they only count as one converter

  • As a consequence of this unique counting, conversions are equally attributed to all previous views of an experience, within an iteration of that experience–as opposed to for example, only the latest view of the experience, or only views within the same session as the conversion. This equal attribution captures the continued influence of an experience on a visitor’s behavior, and is one of the standard attribution strategies in ecommerce

  • For persistent and mutually exclusive post-experience segmentation attributes like device type, or (typically) country, the above logic applies very simply. However, unlike other visitor attributes, it is relatively common for new visitors (first session) to become returning visitors (two or more sessions). Different attribution models for their conversions, may therefore give noticeably different (model-dependent) results

This attribution strategy, attribute a unique visitor’s conversion to all previous experience views within an iteration, has the following behavior for New / Returning visitors and when/whether they convert:

Scenario A

A view occurs during a visitor’s first session, they convert, and never return:

scenario-a

Attribution

Counts as a New visitor, a New converter, but not as a Returning visitor, nor as a Returning converter.

Scenario B

A view occurs during a visitor’s first session and they convert. They return in a subsequent session but never convert:

scenario-b

Attribution

Counts as a New visitor, a New converter, and a Returning visitor (if and only if they viewed the experience a second time), but not as a Returning converter.

Scenario C

A view occurs during a visitor’s first session and they convert. During any subsequent session, they convert again:

scenario-c

Attribution

Counts as a New and a Returning visitor and converter (irrespective of whether they viewed the experience a second time).

Scenario D

A view occurs during a visitors’s first session, they do not convert and they never return:

scenario-d

Attribution

Counts as a New visitor, but not as a New converter, a Returning visitor, or a Returning converter.

Scenario E

A view occurs during a visitor’s first session. In any subsequent session, they do not convert:

scenario-e

Attribution

Counts as a New visitor, but not as a New converter, and as a Returning visitor (if and only if they viewed the experience a second time), but not as a Returning converter.

Scenario F

A view occurs during a visitor’s first session. In any subsequent session, they convert:

scenario-f

Attribution

Counts as a New visitor, a New converter (because their conversion is attributed back to their previous session) and as both a Returning visitor and a Returning converter (irrespective of whether they viewed the experience a second time).

Scenario G

A view occurs for the first time, during or after a visitor’s second session:

scenario-g

Attribution

Will never count as a New visitor in this experience. Counts as a Returning visitor. May count as a Returning converter.

A focus on scenario F

Scenario F is particularly interesting worth discussing in greater detail.

It implies that a visitor’s first interaction with the experience has a continued influence on their subsequent conversion behavior. Indeed, their first view influenced their decision to return to the site, and allowed them to convert (irrespective of whether they viewed the experience a second time).

It is worth underlining that these segmented tests are separate tests, so we are not double-counting these conversions / converters.