Understand ranking with Variant slicing

This guide explains how Bloomreach calculates ranking scores for Variant slicing in search results.

How ranking works with Variant slicing

With Variant slicing, the Discovery search ranking algorithm determines how product variants appear in search results and which variant becomes the "hero" (the primary variant displayed). 

The algorithm distributes a product's overall ranking score among its individual variants using a three-step process that ensures the most relevant and best-performing variants appear prominently in search results.

Step 1: Distribute the product ranking score

Every product starts with a total product ranking score (also called the performance score). This score gets divided among all the product's variants based on how well each variant converts. Variants that convert better receive a larger portion of the total score.

Example

Product A has a total score of 100 for "red shirt" with five color variants. Note these are imaginary values:

• Variant 1 (red): 50% of conversions → 50 points.

• Variant 2 (blue): 25% of conversions → 25 points.

• Variant 3 (red): 15% of conversions → 15 points.

• Variant 4 (blue): 7% of conversions → 7 points.

• Variant 5 (blue): 3% of conversions → 3 points.

The scores are distributed based on actual performance and sum up to 100.

Step 2: Add relevance scores

Each variant inherits the parent product's relevance score. Variant-level attributes are then scored against the query. 

Example

Consider a query for "red shirt" where a "SKU color" attribute is configured at the variant level. The parent product has a relevance score of 20.

• All variants inherit this base relevance score of 20.

• Variants 1 and 3, which are red, receive an additional score on top of the base product score.

• Variants 2 and 4, which are black, don't receive this score boost.

Read on the importance of sending SKU events (link). 

📘

Note

Category ranking uses all signals except the relevance score, since all items in a category have equal relevance.

Step 3: Select hero variants and form listings

Once each variant has a combined score (split product score + relevance score), the algorithm groups variants into listings. The variant with the highest combined score becomes the hero variant that is the primary option displayed to shoppers. 

Ranking calculation example

Suppose a shopper searches for "red dress garden party" and finds a product called "Summer dress" with a total product ranking score of 100.

Product setup

Product: Summer dress

• Total product ranking score: 100

• Number of variants: 5

  • Variant 1: Red, sizes XS/S/L

  • Variant 2: Blue, size S/L

  • Variant 3: Red, size XXL

  • Variant 4: Blue, size 12

  • Variant 5: Blue, size 8/10

Score distribution based on conversion

The algorithm splits the product ranking score among the 5 variants according to their conversion performance:

• Variant 1: 50 points (highest conversion rate)

• Variant 2: 25 points

• Variant 3: 15 points

• Variant 4: 7 points

• Variant 5: 3 points

Adding relevance scores

All variants inherit the base product relevance score of 20.

The query "red dress garden party" contains "red dress," so red variants receive higher relevance scores:

• Variant 1 (red): 20 (base) + 30 relevance points = 50

• Variant 2 (blue): 20 (base) + 5 relevance points = 25

• Variant 3 (red): 20 (base) + 30 relevance points = 50

• Variant 4 (blue): 20 (base) + 5 relevance points = 25

• Variant 5 (blue): 20 (base) + 5 relevance points = 25

Final listing formation

After combining split product scores and relevance scores:

• Listing 1 groups Variant 1 (100 total points) and Variant 3 (65 total points). Variant 1 is the hero variant. 

• Listing 2 groups Variant 2 (50 total points), Variant 4 (32 total points), and Variant 5 (28 total points). Variant 2 is the hero variant. 

Listing 1 appears higher in search results because its hero variant (Variant 1) has the highest combined score of 80 points.

Requirements for accurate ranking

To ensure Variant slicing ranking works correctly, you must send SKU IDs with your conversion events. Without SKU-level conversion data, the algorithm can't determine how individual variants perform.

When SKU IDs are missing

If conversion events lack SKU IDs, the algorithm distributes the product ranking score equally among all variants. Using the previous example, all 5 variants would receive 20 points each (100 ÷ 5) instead of the performance-based distribution of 50, 15, 10, 7, and 3.

This equal distribution means the algorithm can't identify which variants actually drive conversions, resulting in less relevant hero variants and potentially lower search performance.

Implementation requirement

Configure your conversion tracking to include SKU IDs for all variant-level events like add-to-cart, purchase, and other conversion signals. This data enables the algorithm to accurately split product ranking scores based on real performance metrics.