How Advanced Recommender Works
The Advanced Recommender estimates customer body measurements to provide size recommendations. Here’s an overview of how it works:
Collecting Customer Data: Customers are asked for information such as age, weight, height, and bra size (for female tops).
If bra size is provided, chest size is directly calculated.
Estimating Body Measurements: The provided information is processed through a linear regression model to estimate body measurements.
Generating Fit Scores: The estimated measurements are compared to the product’s size chart. Using the recommender field configurations, the system calculates a "fit score" for each size and recommends the size with the highest score.
Common Recommender Issues
Including Incorrect Recommender Fields:
Only actual body measurements should be used for the recommender. Avoid including unrelated fields, such as “UK sizes.”
Typically, only 1-2 recommenders are needed.
Incorrect Canonical Measurement Linking:
Confirm whether the measurements in the size chart represent body measurements or garment measurements.
If using garment measurements, ensure proper linking, such as distinguishing between "Circumference" and "Length."
How to Debug Recommender Issues
Testing with User Match:
Use the User Match feature to input various recommender values and see how they align with the expected sizes.
If sizes aren’t recommended, verify the recommender field and canonical field settings.
Analyzing Size Recommendations:
If recommendations are consistently off, adjust the easing settings:
Lower easing nudges recommendations towards smaller sizes.
Higher easing indicates a looser fit and larger sizes.
Providing Body Measurements to Support:
When working with customer support, share actual body measurements for test cases. This enables accurate comparison with estimated measurements.
How to Tweak the Recommender Setup
Adjusting Length-Based Recommender Fields:
For fields like length, ensure the ideal length reference point is appropriate for the product type (e.g., shirts around the crotch, jackets 10-15 cm below).
Increase the acceptable range from the reference point to make length a less sensitive factor.
Easing Adjustments for Chest, Waist, and Hip:
Refer to user match statistics to fine-tune easing for the highest match rate.
Note: Easing adjustments are only applicable when the canonical measurement is set to product measurement.
How to use User match
The User Match feature allows merchants to input various potential data to see how the recommender’s size matches compare to the expected size. An experimental feature is available for automatic configuration adjustments, though it may not always be accurate. Merchants can also manually adjust settings and directly observe how these changes impact matching accuracy.
For detailed guidance, see the Adding Advanced Recommender User Match article.