Product ranking

In Voyado Product Discovery, the product discovery engine uses several algorithms to determine the ranking of a product, including product life cycle analysis, visitor behavioral insights, exposure strategies, merchandising actions like boosting or burying, as well as business domain knowledge. In combination with exposure strategies and boost and bury settings, these algorithms provide a powerful ranking, giving highly relevant general product listings for any given context.

The product relevance ranking is a fundamental part of the different types of product listings, such as search results and category listings. It also plays a part in the different product recommendations.

Trend sensitivity

The relevance of a product changes over time. The engine uses various behavior-driven algorithms to calculate momentary relevance in order to show what is relevant right now. This includes an iteration of Voyado's own patented Top Seller algorithm (US Patent No. 10 528 577), which intelligently balances trending products with consistent sellers in product listings.

Exposure strategies

Exposure strategies enable merchandisers to optimize how the product discovery engine exposes products toward different business objectives.

There are three different exposure strategies available:

  • Expose higher priced products (Revenue focus)
  • Expose high margin products (Profit focus)
  • Expose products that sell more (Conversion focus)

Read more in the dedicated article Exposure strategies.

Product life cycle

The product ranking is not only based on customer behavior but also important events in the product life cycle. These events include when a product is new, is on sale, or is becoming out of stock. This affects multiple areas of ranking: search results, category listings, phrase suggestions, product suggestions, alternative recommendations, complementary recommendations, and facet relevance.

Product lifecycle

An example of life cycle understanding is that the engine knows that a high-performing product on sale doesn't necessarily keep selling well when the sale is over. Another example is the understanding that out-of-stock products might not be selling well because they can't be purchased, not because they are unpopular. It might even be that the high popularity caused some products to quickly become out of stock in the first place.

Newness

The newness of products is another part of a product's life cycle. New products need increased exposure to determine their popularity, but exploratory exposure comes at the cost of exploiting the insights of how they perform. This means that unpopular new products shouldn't keep taking up screen real estate that can be better used by popular products once they've had the chance to prove themselves. The engine caters to this through greater initial exposure of new products with especially reactive and clever ranking of this segment to avoid overexposure.

Product score smoothing

Products are almost always available in a varying number of sizes. For example, one blouse could be available in eight different sizes, while another blouse can have a coarser sizing scale and only be available in five sizes. Without an understanding of the data structure, a comparison of interactions on these products and SKUs can easily get skewed.

product listing relevance score smoothing

Products are fairly compared through product score smoothing, which is made possible through the well-defined data model. Furthermore, product score smoothing enables a fair and relevant ranking for partially out-of-stock products. This means that the more relative sizes a product is available in, the more likely it will sell, while simultaneously taking the individual popularity of each size into account.

Product property analysis

Some product properties are critical to provide a relevant result, for example in the fashion and lifestyle domain, such as imagery and stock levels. Through image availability and stock knowledge, the engine heavily buries products lacking imagery or being completely out of stock. This happens automatically and is revoked once the product is back in stock or gets accessible imagery.

In-session personalization

In-session personalization adapts the ranking to what a visitor does during their current visit. When a visitor clicks a product, the engine reads that click and finds similar and alternative products. A purchase-likelihood model then gives the most relevant of them a targeted boost, so the listing starts to reflect what the visitor is interested in right now.

It works for all visitors who've given consent, including anonymous ones, so no login is needed. If the visitor has a previous session, the engine also uses their latest click from that session.

In-session personalization is available in:

  • Search results
  • Category listings and Landing pages

The personalization boost competes fairly with the other parts of the ranking, such as slices, boost and bury, and retail media. It never overrides your merchandising rules. Exact-match and highly specific searches also stay unchanged, because a clear search intent always wins.

Boost and Bury

Boost and Bury can be used to either temporarily increase or decrease the strategy score of selected products in a market. Boosted products will gain positions, while buried products will suppress positions in search results, category pages, and other locations where applicable.

Read more in the dedicated articles Boost and Bury.

Was this article helpful?

/