Search relevance is crucial for Algolia, and one way to ensure it is through their custom ranking feature. However, a common issue arises when using this feature with attributes that span a wide range of values, such as the number of views a photo might have. To overcome this, records are bucketed out based on the currently examined ranking factor until exhausted, which can lead to fine granularity and limited use of other attributes. To address this, custom ranking factors need to be considered, but their precision can also limit their effectiveness. To improve relevance, it's essential to reduce the precision of these attributes or convert continuous values into discrete ones using various methods such as creating tiers, reducing data precision, taking the log of values, or creating a custom score at indexing time. The right approach depends on how often the data changes and the number of records, with logarithmic systems potentially being more suitable for frequently changing data and tiering systems working better for records with clumped values.