Sorting Rules within Merchandising uses a combination of Ranking Score and Weighting Factors which relies on separating products by splitting them into percentiles. In this article we will discover these concepts and explain how items are Sequenced into a certain order by using Product Attributes or Performance Metrics. 

The ranking is also context-dependent, involving the Nosto product, its features, the shopper, and the specific use case. However, the concept and the weighting mechanism as a calculation remain consistent.

Within categories and category merchandising, the logic is arguably the simplest: for instance, ranking based on price simply orders products in either descending or ascending order based on their price.

In Search, ranking is influenced by a relevancy factor calculated as a score, which is derived from the search query or term used and reflecting the relevancy of each product. This means that using the example of ranking by price is influenced by both relevancy factors of the search in relation to products, and the merchandising applied: price. In the case of search, merchandising never completely overrides relevancy; it merely helps to adjust the ranking. A practical merchandising and search guide is available here.

In Recommendations, the ranking and sequence of products depend on the type of recommendation as the baseline and the degree of relation between the products. For example, 'best sellers by purchases' is straightforward, as the baseline ranking follows commercial popularity. However, for cross-sellers, the relationship between products is determined by a concept of 'similarity score' calculated by Nosto's ML and AI models; in essence, some products are more closely related than others. The applied merchandising rule is then considered as part of the score, which, similar to the search relevancy score, influences the order of products.

Lastly, any personalization factor, whether as a merchandising rule or as an inherent component in recommendations, is accounted for and can affect order of products for each individual shopper separately.

In summary, except in categories, the merchandising rule never solely determines the order; it is also influenced by the search term, recommendation type and settings, and the shopper's personal preferences (personalization). Merchandising should be seen as a powerful tool to adjust the order to meet strategic and preferred commercial objectives, but it is not the sole determinant of product ranking or sorting.

The next chapter breaks down how ranking itself is calculated.

Every product within a catalogue is ranked in a specific order for every single Product Attribute and Performance Metric indexed by Nosto. This means that one product might rank 1st on price, whereas it can rank 17th on conversion rate. 

For example, using high price as a factor, the most expensive product within the product catalogue would have a Ranking Score of 100% attached to it, whereas a product somewhere in the middle could have a Ranking Score of 57%. 

​When adding Product Attributes and Performance Metrics into a Sorting rule, you will notice an individual slider for every row added. This defines the weight and allows you to prioritize flexibly based on what you want to achieve.

If an item were to rank 1st on price, whereas it ranked only 17th on conversion rate, we could use the weights to try and normalize the scale to favour conversion rate when calculating the final results. 

Example: The product ranks 100% on price, and let's assume 17th means that the product ranks 87% on the scale considering conversion rate. This is relative to the conversion rate of other items so the real percentage can be 87% or 15% in this example. If the weighting for price is set at 30%, and the weight for conversion rate is set at 70%, we can easily do the math to figure out the Ranking Score for this product. 

This is done by calculating the percentages with the weighting factors, and adding up the total sum.
​
(1.0 * 0.3) + (0.87 * 0.7) = 0.909 = 91%

This item has a relatively high score when considering in what order to Sequence the products. This weighting allows for items with a poorer rank on price, but a higher rank on conversion rate to surpass the score of 91%, which is the expected behavior when using a weighting of 70/30 as we have done here in our example. 

Category Merchandising applies a percentile rank for every Product Attribute or Performance Metric to split products into percentiles. Using this method it does not matter how wide or narrow the scale, how many or few items observed, or how clustered items are within the center of the scale. 

By using traditional value-based ranking, most items would be clustered in the middle, with some outliers on the high-end and the low-end of the spectrum. By using percentile rank as the underlying methodology there is no bell-shaped curve when looking at a collection of items. Instead products are evaluated on a scale of percentiles for every Product Attribute or Performance Metric. 

​