.svg){kind=icon}
.svg)
Introduction
At Virtusize, our in-house data science team continuously conducts evaluations and regularly refines the accuracy of our recommendation services. The information we collect from our Virtusize users consists of only four size details (height, weight, gender, and age). By combining this with some basic item information provided by our clients, we can adjust our how our system makes recommendations. Today, we will introduce the important metric for adjusting our recommendation logic, called the "Size Match Rate," and share a case study where we applied logic adjustments.
Size Match Rate
As we work to improve recommendation accuracy, we rely on a key metric called the "Size Match Rate." This term refers to the percentage of purchases, where customers bought the size that Virtusize recommended first.
This rate is calculated by combining our recommendation data with order information from our clients. We can delve into various categories and groups, such as brand + product type and product type + style, to score the match rate. By identifying where our recommendations align with customer expectations and where they do not, we can make more precise logic adjustments.
Previously, we manually verified and evaluated the data. However, since developing an automated "size match rate" calculation tool- we can now calculate a year's worth of "size match rates" within an hour. This allows us to more specifically identify areas where we can improve, for each client, and significantly reduce the time required for the software adjustments, leading to expected improvements in the quality of our service.
Size Recommendations Based on Both Customer and Item Size Information
This approach is closely related to the core philosophy of our size match algorithm. At Virtusize, we believe that size recommendations should be based on both the customer's size information and the item's size details. Common approaches that ignore item measurements and only make suggestions based on customer attributes, can mislead customers in the apparel industry. An example of this: "Users similar to you (age, gender) also purchased this."
Simply recommending items based on what other people bought, could result in customers purchasing sizes that do not fit their unique body measurements.
Therefore, by taking extra steps to verify the "size match rate," we can determine how accurately Virtusize recommends sizes and specifically how well the purchased sizes actually match the sizes recommended by Virtusize. A high "size match rate" is not always good, nor is a low rate always problematic. For example, if a specific product has unusual sizing and most people who usually buy M would need an L, and if this information isn't well communicated, some customers might still purchase M based on their usual size… leading to a lower "size match rate." A low "size match rate" doesn't necessarily indicate poor recommendation accuracy; it serves as a starting point for identifying causes and finding solutions in collaboration with our clients.
Case Studies of Improving Size Match Rate
When we identify groups with "size match rates" below our standard, our data science team begins evaluating whether adjustments to Virtusize's software are necessary or if there are potential improvement points on the client’s site.
We have collaborated with many clients to make various improvements, successfully increasing "size match rates" by up to 20% on numerous sites.
A simple example is a client site we'll call Brand A. Brand A's performance at the beginning of 2023 was relatively stable and not bad. We had sufficient data to measure the "size match rate" for each item and test new software adjustments. Through repeated adjustments and measurements over a year, we improved Brand A's "size match rate" by 18.6% compared to the same period the previous year!
Conclusion
Our data science team's mission is to continuously explore core methods for improving recommendation accuracy from both precision and performance perspectives, with the "size match rate" playing a significant role in this. In 2024, we will continue our efforts to enhance the quality of our service!
Introduction
At Virtusize, our in-house data science team continuously conducts evaluations and regularly refines the accuracy of our recommendation services. The information we collect from our Virtusize users consists of only four size details (height, weight, gender, and age). By combining this with some basic item information provided by our clients, we can adjust our how our system makes recommendations. Today, we will introduce the important metric for adjusting our recommendation logic, called the "Size Match Rate," and share a case study where we applied logic adjustments.
Size Match Rate
As we work to improve recommendation accuracy, we rely on a key metric called the "Size Match Rate." This term refers to the percentage of purchases, where customers bought the size that Virtusize recommended first.
This rate is calculated by combining our recommendation data with order information from our clients. We can delve into various categories and groups, such as brand + product type and product type + style, to score the match rate. By identifying where our recommendations align with customer expectations and where they do not, we can make more precise logic adjustments.
Previously, we manually verified and evaluated the data. However, since developing an automated "size match rate" calculation tool- we can now calculate a year's worth of "size match rates" within an hour. This allows us to more specifically identify areas where we can improve, for each client, and significantly reduce the time required for the software adjustments, leading to expected improvements in the quality of our service.
Size Recommendations Based on Both Customer and Item Size Information
This approach is closely related to the core philosophy of our size match algorithm. At Virtusize, we believe that size recommendations should be based on both the customer's size information and the item's size details. Common approaches that ignore item measurements and only make suggestions based on customer attributes, can mislead customers in the apparel industry. An example of this: "Users similar to you (age, gender) also purchased this."
Simply recommending items based on what other people bought, could result in customers purchasing sizes that do not fit their unique body measurements.
Therefore, by taking extra steps to verify the "size match rate," we can determine how accurately Virtusize recommends sizes and specifically how well the purchased sizes actually match the sizes recommended by Virtusize. A high "size match rate" is not always good, nor is a low rate always problematic. For example, if a specific product has unusual sizing and most people who usually buy M would need an L, and if this information isn't well communicated, some customers might still purchase M based on their usual size… leading to a lower "size match rate." A low "size match rate" doesn't necessarily indicate poor recommendation accuracy; it serves as a starting point for identifying causes and finding solutions in collaboration with our clients.
Case Studies of Improving Size Match Rate
When we identify groups with "size match rates" below our standard, our data science team begins evaluating whether adjustments to Virtusize's software are necessary or if there are potential improvement points on the client’s site.
We have collaborated with many clients to make various improvements, successfully increasing "size match rates" by up to 20% on numerous sites.
A simple example is a client site we'll call Brand A. Brand A's performance at the beginning of 2023 was relatively stable and not bad. We had sufficient data to measure the "size match rate" for each item and test new software adjustments. Through repeated adjustments and measurements over a year, we improved Brand A's "size match rate" by 18.6% compared to the same period the previous year!
Conclusion
Our data science team's mission is to continuously explore core methods for improving recommendation accuracy from both precision and performance perspectives, with the "size match rate" playing a significant role in this. In 2024, we will continue our efforts to enhance the quality of our service!
Introduction
At Virtusize, our in-house data science team continuously conducts evaluations and regularly refines the accuracy of our recommendation services. The information we collect from our Virtusize users consists of only four size details (height, weight, gender, and age). By combining this with some basic item information provided by our clients, we can adjust our how our system makes recommendations. Today, we will introduce the important metric for adjusting our recommendation logic, called the "Size Match Rate," and share a case study where we applied logic adjustments.
Size Match Rate
As we work to improve recommendation accuracy, we rely on a key metric called the "Size Match Rate." This term refers to the percentage of purchases, where customers bought the size that Virtusize recommended first.
This rate is calculated by combining our recommendation data with order information from our clients. We can delve into various categories and groups, such as brand + product type and product type + style, to score the match rate. By identifying where our recommendations align with customer expectations and where they do not, we can make more precise logic adjustments.
Previously, we manually verified and evaluated the data. However, since developing an automated "size match rate" calculation tool- we can now calculate a year's worth of "size match rates" within an hour. This allows us to more specifically identify areas where we can improve, for each client, and significantly reduce the time required for the software adjustments, leading to expected improvements in the quality of our service.
Size Recommendations Based on Both Customer and Item Size Information
This approach is closely related to the core philosophy of our size match algorithm. At Virtusize, we believe that size recommendations should be based on both the customer's size information and the item's size details. Common approaches that ignore item measurements and only make suggestions based on customer attributes, can mislead customers in the apparel industry. An example of this: "Users similar to you (age, gender) also purchased this."
Simply recommending items based on what other people bought, could result in customers purchasing sizes that do not fit their unique body measurements.
Therefore, by taking extra steps to verify the "size match rate," we can determine how accurately Virtusize recommends sizes and specifically how well the purchased sizes actually match the sizes recommended by Virtusize. A high "size match rate" is not always good, nor is a low rate always problematic. For example, if a specific product has unusual sizing and most people who usually buy M would need an L, and if this information isn't well communicated, some customers might still purchase M based on their usual size… leading to a lower "size match rate." A low "size match rate" doesn't necessarily indicate poor recommendation accuracy; it serves as a starting point for identifying causes and finding solutions in collaboration with our clients.
Case Studies of Improving Size Match Rate
When we identify groups with "size match rates" below our standard, our data science team begins evaluating whether adjustments to Virtusize's software are necessary or if there are potential improvement points on the client’s site.
We have collaborated with many clients to make various improvements, successfully increasing "size match rates" by up to 20% on numerous sites.
A simple example is a client site we'll call Brand A. Brand A's performance at the beginning of 2023 was relatively stable and not bad. We had sufficient data to measure the "size match rate" for each item and test new software adjustments. Through repeated adjustments and measurements over a year, we improved Brand A's "size match rate" by 18.6% compared to the same period the previous year!
Conclusion
Our data science team's mission is to continuously explore core methods for improving recommendation accuracy from both precision and performance perspectives, with the "size match rate" playing a significant role in this. In 2024, we will continue our efforts to enhance the quality of our service!
