Sizing is hard. Here’s the math. And the solution.
Sizing is hard. If you’re a modern-day retailer, you know this. These days, everyone’s trying to use “fancy” AI and machine learning and big data to solve this problem and that problem. Unfortunately, using much simpler math and some even simpler logic, we can prove that those techniques will solve our problems is wishful thinking.
Let’s begin by asking why a dress, which comes in typically 5 sizes, so hard to properly?
Sizing is multi-dimensional
Take a simpler base case: a pair of high-waisted shorts. To fit, they need to fit at the waist and at the hip, so we’re working with two dimensions here.
So if you have the right hip-waist ratio — that is to say, your customer’s body falls on the red line — then the shorts fit great. There may be some people who truly are “between” sizes, but for them, the fit should still be pretty reasonable because they’ll fall in the “neighborhood” of a size.
Covering all dimensions requires too many sizes
Now, what if we want to accommodate all the different hip-waist ratios in between? Well, we can add more sizes. How many more, you ask?
Approximately 52 = 25 more. You see, since we’re working with two dimensions here, complexity grows quadratically (y = x2) with our range. Hip and waist measurements have some independent variability, after all. Anyways, keeping around 25 uniques sizes might not out of the question for some brands who stock staples for years, but definitely an inventory nightmare for most stores. And that’s not even considering the amount of headache it takes a customer to consider 25 different sizes!
Now what happens when we want to fit a dress? The fit quality then depends on measurements taken from the waist, hip, shoulders, bust, and height. So five dimensions. Suppose we have five “straight” sizes now. That means that if we wanted to “complete” coverage, we would need approximately 55 = 3125 uniques sizes. Now that’s a lot of inventory!
Of course, some hip-waist-shoulder-bust-height ratios are very rare in practice, you can rule out about two thirds of these hypothetical sizes. Still, that means that your store would need over 1000 unique sizes to cover say, 80% of body types in the US. And machine learning and stylists aren’t magic either: these can only help a customer pick the best size for them, but even that “best” size might be an out-of-bounds, awful-fit!
Also through this exercise we learn why it is that it’s so much harder to fit women’s clothing. Notice that with a suit, we can have say, three dimensions for the torso, and two for the legs. That dramatically reduces the complexity because we can mix and match two less complicated pieces. With dresses, the opposite is true. The dress needs to fit the entire body. Thus, it’s not necessarily that women are pickier about fit; it’s that they’re operating in a higher dimensional space! After all, adding stretchy fabric can only go so far. Certainly, some cuts and fabrics are more forgiving than others, but the general rule is clear.
Evidence in business metrics: returns rates, sizes, inventory
Consider the table below:
Our partners have found that indeed, fit issues overwhelmingly follow the above trend. We rarely need to return socks — certainly not for fit issues anyways, and 2–3 sizes is usually plenty. On the other hand, dresses and jeans can be a nightmare for customers — even 10 sizes is far from enough.
Then remember how we’re leaving out plus-sizes and petites most of the time? In reality, most sizing lineups have coverage like this:
By our estimates, the typical sizing lineup of brands can serve between 20% to 40% of the shoppers who want to buy and wear the brand’s clothes. The rest have written off the brand entirely, return aggressively, or need to alter everything they buy. In a digital age of seemingly endless consumer choices, shoppers are increasingly gravitating to the former two, which leads to a long, slow death for many brands as they see their customers leave to adjacent brands or return to the point where the customer is no longer profitable.
My statistician friends will know this as the curse of dimensionality. At Fitted, we saw this as an opportunity.
The best solution has been around for a while
Back before the days of mass-produced ready-to-wear, most people would have clothes custom-made. But making clothes from scratch is slow and expensive, and today’s shoppers aren’t waiting for their clothes to be made from scratch. It’s why made-to-measure works great for a staple suit, but not for anything you need in less than three months. So let’s rule that out.
However, by analyzing sizing profiles and — most importantly — making some small but massively impactful customizations for customers, we help retailers go from serving only 20% of customers to serving 80–90% of them. You see, now that we can do the last-mile adjustments for each garment, each piece can serve a much wider range of customers, and even those that don’t fall on the “straight size” line: each item can “spread” to serve a wider range of bodies.
By not making things from scratch but by starting with something that’s “close enough”, we add as little as one day to getting the item in the hands of the customer. Even better? It’s much easier to exhaust inventory now, rather than getting stuck with a bunch of 00’s or XL’s that nobody’s buying and has been on clearance for weeks now. Funnier still? This isn’t anything new; fashion writers and models get everything tailored.
As for shoppers for whom your sizing profile already fits, well, they know that, and they don’t need the custom size. Fitted focuses purely on the massive market your inventory isn’t serving yet, without requiring your store to perform any complicated logistics or product overhauls or stock 1000+ sizes per item.
If you’re reading this and curious about what a Fitted option can do for your brand, reach me at yuan@tryfitted.com.
About the author
Yuan Wang is a mathematician. He has worked on using math and data to solve real-world problems since 2012, in aerospace, finance, music, marine conservation, and health insurance. He is a cofounder and CTO at Fitted, where he’s helping brands use math to deliver better products.