The Postmodern Tailor: Size Personalization Beyond Labels

Zhou Yu and Brian Coffey
- San Francisco, CA

“What’s your size?”

“It depends could be XS, S or M.”

If this conversation sounds familiar, you know we have a serious problem with our sizing system.

Size labels are a relatively new thing in human history, especially for women’s clothing. A little over a hundred years ago, if you needed a piece of clothing, you would get it from your local tailor or your family members, using your personalized measurements. There was no such thing as size.

But today, we have all accepted size labels as a convenient and affordable way to produce and shop for clothes. It’s commonly assumed that the size labels are universal between different pieces of clothing and brands. Unfortunately, this assumption is not true and we’re going to do a deep dive on how it could be done better.

Problem

In our current sizing system, a few size labels or numbers (e.g., small, medium, large for tops, and 4, 6, 8 for bottoms) represent all measurements and attributes of both clothing and customers. Even worse, manufacturers often have fairly inconsistent sizing specs due to the lack of size standardization. Threadbase shared some intriguing results about the large variations of chest and body length measurement of a shirt across different brands.

At Stitch Fix, we have found similar patterns of variations, not only among clothing but also among our customers with a shared size label. In the chart below, we plot chest width and body length measurements of 180 Stitch Fix blouses (on the left), and 5,000 randomly generated body dimensions and size labels that represents our customer base.


chest width body length chest size height 20 26 32 32 40 48 30 40 50 55 65 75 Blouses Clients XS S M L XL (all measurements in inches) hover to highlighta single size


Selecting, for example, size S, one can see the huge variability of chest width and body length measurements, for both blouses and customers. Even within a single size, chest and body length measurements each have over 10 inches of variation! Additionally, by hovering your mouse from one size to another you can see how much the different sizes’ measurements overlap, which explains why one person can often fit across multiple sizes (e.g., from XS to S to M).

Due to such large variations, we have long realized that we cannot rely solely on the traditional labeling of size to make our customers happy. It is essentially a blind cross-matching process from inconsistently labeled clothes to a group of customers with very different body measurements. As a result, the current sizing system is far from satisfactory, let alone offering any personalization.

However, we have found ways to do much better.

Solution

At Stitch Fix, we collect size and fit feedback for every piece of clothing we send to our customers. Combined with the data above, this provides us with great insight into fit preferences. The chart below illustrates one such high-level insight. Note that in order not to give away too much of our process and feedback data, it uses fabricated sent-to-customer and feedback data that is meant to be conceptually illustrative but does not necessarily reflect our data with much accuracy.

chest width body length chest size height Blouses Clients too small just right too big 20 26 32 32 40 48 30 40 50 55 65 75 select a blouse to match to clients select a client tomatch to blouses personalized size match according to size label . . .


By selecting the “match according to size label” mode, you can see how blouses are matched to customers using traditional size labels. Due to the large variation of measurements within a single size, the clothes often fit poorly on customers and can cause low fit satisfaction.

However, if we match our blouses and customers based on their actual measurements, i.e., in the “personalized size” mode, we see significant improvements in fit feedback from customers. We can match busty customers with wide chest blouses and non-busty customers with narrow chest clothes, regardless of their size labels. Similarly, we can do the same for customers’ heights and blouses’ body lengths. We define a similarity metric between clothes and customers using the actual measurements. In most cases we still use a customer’s stated size label and select clothing that will best fit them within their preferences. Over time, we expect that generic size labels will vanish.

Takeaway

As many have noticed (e.g., Threadbase and the Washington Post), the current sizing system that cross-matches customers and clothes using size labels is a broken system. At Stitch Fix we help our customers look and feel their best, and a crucial component of that is shipping well-fitted clothes. We’re actively researching this domain and implementing the findings in our product.

Tweet this post! Post on LinkedIn
Multithreaded

Come Work with Us!

We’re a diverse team dedicated to building great products, and we’d love your help. Do you want to build amazing products with amazing peers? Join us!