Apptus aims to show the most relevant products to a visitor at any given time and in every context possible. Every product and variant is assigned a relevance score with each panel query. The relevance score, or relevancy, is then used to compare the products returned in the query.
The relevancy is based on context and context is based on four different components; the panel that is used, the business configuration of eSales (including exposure strategy settings, promotions, and demotions), the historic visitor behaviour of all previous visitors, and the historic visitor behaviour of the current visitor.
Panels that accept the
sort_by argument can sort products according to relevance with algorithms such as Apptus own Top Seller algorithm (patent pending, EP 16185313.0, US 15/680 44).
relevance sort order is Apptus own algorithmic sorting method for eSales, and the sort order that is recommended in most use cases. It selects a ranking algorithm based on what panel it is used with and is the sort order to use to fully benefit from exposure strategy settings, promotions, and demotions.
The relevancy of a product or variant is optimised to maximise its usability based on the current exposure strategy set in the Business app. The
relevance sort is unlike other sort orders a dynamic sort that is based on context and visitor interactions, and improves itself based on interactions such as searches and purchases.
Search panels, such as the Search hits panel, use an algorithm that matches products and variants into several classes. Each class is defined by a combination of features that can be related to match type, strategy score, and visitor behaviour history. Match type is defined as how well a search phrase matches, and in what attribute it matches.
Classes are sorted with the highest ranking class first, and products and variants are sorted internally within the classes based on strategy score and clicks. If a product or variant matches more than one class it will be used in the highest ranking class it matched.
Panels used for category listings, such as the Product list panel, use what can be considered a simpler search ranker where no attribute matching or search phrase related features are involved.
Active promotions and demotions can modify the position of products and variants altering their strategy score. The change in boost level (negative for demotions, positive for promotions) may move the products and variants either within their class or to a different class.
Visitor behaviour history is based on the actions of not only the current visitor, but all previous visitors. Notifications such as Click Notification and Secure Payment Notification must be correctly used to gather behavioural data.
Relevancy with ads and recommendations¶
Not all panels support arguments for product sorting, such as panels used for recommendations or ads. In these panels the relevancy of a product or variant is determined by different algorithms, including Apptus' own Synergy algorithm; a self-learning algorithm for recommendations driven by visitor behaviour. Each panel will use an algorithm that is optimised for use with that specific panel and what algorithms these panels are using can not be altered.
Recommendations are ranked using algorithms that like the
relevance sorting also take promotions and demotions into account. Products in an active promotion are more likely to appear in a result as long as they are relevant to the query. Products in an active demotion are removed from recommendations. Personal recommendations also utilise historic visitor behaviour of the current visitor.
Ads are primarily optimised based on order value and number of displays. Based on the type of ad panel used, additional ranking parameters are used such as search queries, product keys or context in general including historic visitor behaviour of the current visitor.
Relevancy of a product or variant changes over time. Apptus eSales uses various algorithms to calculate momentary relevance in order to show what is relevant right now. All strategy score based relevance ages data using our own Top Seller algorithm (patent pending, EP 16185313.0, US 15/680 44) which intelligently balances trending products with consistent sellers.