...
Product properties
the item brand and categories are indicated in GA4 view_item events, but Boxalino has a lot more product attribute from the e-shop, including (the properties would be provided as an associate array with property names and values: ARRAY<STRUCT<name, ARRAY<string> values>>):brand
category (several levels)
price (before or after discount)
number of stars
state of discount (in discount and how much)
pictures (number of pictures, urls, …)
description (to see if the length or other aspects could play a role)
delivery state (in stock, …)
…
in example after: product_properties : have string properties for the recommended products
GA4 data
view_item indicating that it is a PDP
traffic source
device
geo
…
in the data : all the ga4 view_item event are therepage_views (to see what page view happens before / after, because it is maybe very different if it is a landing page)
traffic source
device
geo
…in the data: array_length of prior_pages
fyi- the sample data after are computed (except for the product_properties) fully on the basis of GA4 data.
Performance metrics:
Engagement rate / Bounce rate
in the data: next_pages would be emptyNumber of page views
in the data: array_length of next_pagessession conversion rate
in scope conversion rate (not only there is a conversion, but there is a conversion with this product)session the data: count(distinct session_id) / count(distinct next_purchases.ecommerce.transaction_Id)in scope conversion rate (not only there is a conversion, but there is a conversion with this product)
in the data: compare the main item with the items of the next purchase itemssession $ value (average value per session)
in the data: next_purchases.ecommerce.purchase_revenue and session_idAOV (averge order value)
Factor analysis of parameters affecting the performance of product recommendations on the PDP
Identify automatically what are the factors of a good performance or a bad performance of a product recommendation
Example: recommending products which have bad ratings might be a bad idea
Example: recommending products of the same brand than the one of the currently viewed product might be a good idea for some brands and a bad idea for other brands
These parameters can be related to the product data of either the PDP or of the recommended products, as well as the session data (device, …) or the combination of them (e.g.: on products of the brand X recommending products of the brand Y doesn’t work)
Data Available:
Product properties (for both the PDP and the recommended products)
the item brand and categories are indicated in GA4 view_item events, but Boxalino has a lot more product attribute from the e-shop, including (the properties would be provided as an associate array with property names and values: ARRAY<STRUCT<name, ARRAY<string> values>>):number of stars
state of discount (in discount and how much)
pictures (number of pictures, urls, …)
description (to see if the length or other aspects could play a role)
delivery state (in stock, …)
…
Boxalino tracking data
the display of each recommended product
the click on each recommended product
the add-to-basket following a click on the product
the purchase following a click and an add-to-basket on the product
test variant (it is often the case that there are several algorithms tested as different test variant, we could use the test variant as a factor to see if it makes a different (positive or negative) for specific products)
GA4 data
view_item indicating that it is a PDP
page_views (to see what page view happens before / after, because it is maybe very different if it is a landing page)
traffic source
device
geo
…
Performance metrics:
Click-Through-Rate (the number of times the recommended product is clicked compared to the number of times it is shown)
Add-To-Basket-RAte (the number of times the recommended product is clicked and then added to the basket compared to the number of times it is shown)
Buy-Through-Rate (the number of times the recommended product is clicked and then added to the basket and then bought compared to the number of times it is shown)
Display $ Value (the average value bought by the Buy-Throughs compared to the number of times the product is shown, so the average value of showing the product as a recommendation)
Engagement rate / Bounce rate
Number of page views
session conversion rate
session $ value (average value per session)
AOV (averge order value)in the data: next_purchases.ecommerce.purchase_revenue / next_purchases.ecommerce.transaction_id
Expand | |||||
---|---|---|---|---|---|
| |||||
|
Factor analysis of parameters affecting the performance of product recommendations on the PDP
Identify automatically what are the factors of a good performance or a bad performance of a product recommendation
Example: recommending products which have bad ratings might be a bad idea
Example: recommending products of the same brand than the one of the currently viewed product might be a good idea for some brands and a bad idea for other brands
These parameters can be related to the product data of either the PDP or of the recommended products, as well as the session data (device, …) or the combination of them (e.g.: on products of the brand X recommending products of the brand Y doesn’t work)
Data Available:
Product properties (for both the PDP and the recommended products)
the item brand and categories are indicated in GA4 view_item events, but Boxalino has a lot more product attribute from the e-shop, including (the properties would be provided as an associate array with property names and values: ARRAY<STRUCT<name, ARRAY<string> values>>):number of stars
state of discount (in discount and how much)
pictures (number of pictures, urls, …)
description (to see if the length or other aspects could play a role)
delivery state (in stock, …)
…
in example data :
string_properties : have string properties for the recommended products
numeric_properties : have numeric properties for the recommended products
source_string_properties : have string properties for the current PDP
source_numeric_properties : have numeric properties for the current PDP
Other data
test variant (it is often the case that there are several algorithms tested as different test variant, we could use the test variant as a factor to see if it makes a different (positive or negative) for specific products)
in example data : variant_id
Performance metrics:
Click-Through-Rate (the number of times the recommended product is clicked compared to the number of times it is shown)
in example data : KPI_defined.click / KPI_defined.mainImpressionAdd-To-Basket-RAte (the number of times the recommended product is clicked and then added to the basket compared to the number of times it is shown)
in example data : KPI_defined.click_n_atb/ KPI_defined.mainImpression
Buy-Through-Rate (the number of times the recommended product is clicked and then added to the basket and then bought compared to the number of times it is shown)
in example data : KPI_defined.click_n_buy / KPI_defined.mainImpression
Display $ Value (the average value bought by the Buy-Throughs compared to the number of times the product is shown, so the average value of showing the product as a recommendation)
in example data : KPI_defined.click_n_revenue / KPI_defined.mainImpression
Expand | |||||
---|---|---|---|---|---|
| |||||
|