SKU Implementation

Overview

This page offers a high-level overview of how Algonomy currently works with SKU data to render the correct SKU for a recommended product.

SKU data helps Algonomy deliver highly personalized recommendations. Current SKU functionality does not support specific SKU recommendations. Also, a report specific to SKU is not available. All attributions (sales, clicks, etc.) are at the product level, not at the SKU level.

Follow the below requirements to use SKU attributes:

  • You must send Algonomy SKU specific information to feeds AND must also send SKU information through instrumentation (both are required for science to be able to generate the user preferences).

  • Algonomy should ensure to understand which SKU attributes are primary/considered most important to the client to set preferences. For example, is Size more important than Color? This provides the science team with preferences to load (if any) for the customers.

Rendering SKUs

Currently, Algonomy renders SKUs using the following process.

On the Backend

The science team generates some implicit user preferences based on SKU attributes for a user (currently size, color, or age).

The science team loads these preferences into UPS

Enable "Use SKU Inferences" site configuration (under the Customer Preference Center (CPC))

Graphical user interface, text Description automatically generated

On the Website

  1. Customer browses on the website.

  2. All the regular strategies that are enabled for that site and applicable to the customer behave exactly as before in KOTH.

  3. KOTH comes up with the winning strategy and product recommendations.

  4. If the customer has stored implicit preferences in UPS, and the products being recommended have a SKU that matches the implicit preferences for that customer, then Algonomy changes the product URL to the appropriate SKU URL.

Things to Note

  • We still do product level recommendations.

  • We only renders the appropriate SKU for the product recommendation if a preference is loaded into UPS.

  • This is global which applies to all APIs.

  • All attributions (sales, clicks, etc.) are at the product level, not at the SKU level.

Merchandising Rules

Merchandising rules can be set up for specific SKU attributes. If the seed product matches the defined SKU attribute, then the action defined by the rule is executed.

Enable Current SKU Functionality for Clients

  • Create an Engineering Ticket and assign it to Jeremy York for the science team to generate implicit SKU preferences.

  • Confirm that the client is enabled for using UPS and the site configuration is set to "UPS enabled and data used".

  • After the preferences are loaded in UPS, Recommend automatically considers them in future.

SKU On/Off MVT Testing

The use of SKU inferences can be MVT tested. It is an ON/OFF test, to use or not to use the SKU inferences.

How the science team builds the customer's SKU preferences

For the current prototype, let’s use Carters as an example:

  • We only attempt to determine size preferences for a user who has purchased at least 5 products.

  • We ignore shoe purchases for the moment, size is determined only for clothes.

  • We infer the age of the child the product was purchased for, based on the Carter’s sizing guide.

  • We determine what the current age of that child would be today (that is, if an item for a 3-month-old was purchased on October 1st, and today is February 1st, we assume that the child is now 7 months old).

  • We determine the age that is shopped for most frequently, preferring to rely on recent purchases in the event of a tie.

  • We map that age back to a size, based on the Carter’s sizing guide.

  • If we have a preferred size for the customer, we will also attempt to determine a preferred gender and preferred color using similar logic. If we cannot determine a size, we do not try to specify gender or color.

Examples

If a customer looks at 3 pink items,1 yellow and 2 purple items – how will Algonomy determine affinity

  • If we have also been able to determine a preferred size, we will try to show a pink SKU in the desired size. If no pink SKU is available, we will try to show a purple SKU and then a yellow SKU, in that order. If none of those colors are available we will show an arbitrary SKU with the desired size. If we do not know a desired size, we will not link to a specific SKU.

Customer is shopping for 2+ kids and all the behavior is stored in UPS –how is determination made on affinity?

  • If the customer purchases for two or more children, in most cases we will show items in the size for the child that the customer has bought for most often. If there is a tie, we will prefer the size implied by recent purchases.