Skip to content
Toposort Logo
Knowledge Base

Sponsored Listings Associations

As part of the integration process, Topsort has a one-time offline data requirement. This data is used by the bidding and pacing algorithms to bootstrap its predictions, as it allows us to have them up and running from day one without having to wait for the collection of data through our systems. While this is not a hard requirement, it increases the chances of having a smooth launch of the product for the initially onboarded vendors.

There are two kinds of data requirements that will help us enhance the initial performance of the autobidder:

  1. Behavior data,
  2. Catalog data.

The Behavior data consists of daily aggregations of listings organic interactions with the users, including impressions, clicks and purchases. In the case the marketplace already has ads running in their systems, the promoted interaction aggregations are of higher value for our purposes. The suggested data to be sent is the following:

  • day_date
  • listing_id
  • seller_id
  • audiences (if available)
  • placement_id (page of the app)
  • geo_region_id
  • organic_impressions_count
  • organic_clicks_count
  • organic_orders_count
  • organic_orders_amount_total
  • promoted_impressions_count (if available)
  • promoted_clicks_count (if available)
  • promoted_orders_count (if available)
  • promoted_orders_amount_total (if available)
  • ad_revenue_from_clicks (if available)
  • ad_revenue_from_impressions (if available)

The ideal range of these events is of a month, but may depend on the traffic of the marketplace.

The Catalog data consists of the metadata associated with the products listed in the marketplace. In particular, relational data such as categories or other fields that relate products to each other is highly relevant for us. This allows us to transfer information from one product to another related product via AI models tailored to our autobidder. The suggested data to be sent is the following:

  • listing_id
  • seller_id
  • categories
  • price
  • geo location (if relevant)

If the products of the marketplace don’t have a notion of category, it can be replaced by other kinds of hierarchy or taxonomy fields, such as department, division, brand, or any other field that relates products to each other will be useful for us to bootstrap the algorithms.