This is The Age of AI Series, where we talk to the foremost entrepreneurs and innovators around the planet using ML to transform industries. (Join our special mailing list!)
If you’ve ever purchased clothes online, you’ve probably bought the wrong size at least once. There are two reasons for this: people are bad at knowing our own measurements, and most sizing charts online are unreliable!
This may seem like a daily trivial problem, but its impact on the fashion ecommerce business is horrific. Unreliable sizing means poor conversion rates, and bad sizing means high return rates.
But it seems like AI can help! I met with Leon Szeli, CEO and Co-Founder of Presize.AI, a software company that helps online fashion retailers help their own customers find the perfect size, through a white-labeled ML product.
We cover a lot of ground:
- 2:36 — The economics of online shopping returns, how they affect the business
- 7:02 — Getting the perfect size for clothes bought online: real challenges
- 11:00 — What Presize AI really does and how it works (in detail)
- 23:04 — Difficulties while finding the initial solution
- 27:24 — How to build ML-oriented product teams (mistakes made, lessons learned)
- 42:44 — Why a custom-made suits tech from Asia failed miserably in Europe
- 44:36 — Why Germany has very high return rates for clothes bought online
Aman’s 2-Minute Summary and Key Takeaways
Let’s first start with some numbers about the problem (at least from a European perspective):
- Customers don’t trust most sizing charts provided online: and the data shows they have a click rate of only 1%.
- Return rates for different brands due to poor size choices range anywhere from 10% to 50%.
- Very few online retailers have the scale and muscle to “absorb” the damage caused by such return rates. For most clothes, a single return means you barely break even, and a second return means you lose money.
So the value proposition of Presize to retailers is both higher conversions and fewer returns. Which explains why they’re now working with so many large customers now.
The challenge however, is in the implementation. How do you build an AI product that works better than human intuition? Here’s the key idea: if you can just measure people better, you can usually find the perfect size. And you can do it using computer vision.
The customer places their phone camera on the floor against the wall, and the app scans their body to estimate their measurements. They also use ML to read some metadata about the product — like whether it is a slim fit or a baggy fit etc, and then recommend the right size. It also creates a 3D model of the customer, so that they can reuse their measurements on other shopping sites where Presize is enabled.
Leon claims that their deep learning models have beaten all accuracy benchmarks across the industry, in terms of predicting the right size based on measurements. The data they used to train models was quite noisy and very hard to get, so it took them a long time to get the product working well!
We also talked a lot about the struggle of building a ML product when you have no proof of concept, no data, and no customers, and a looming question of whether it can be done at all.
Being in this situation required them to change their company’s engineering org structure in such a way that both R&D and product dev were always in sync and have the same north star. In fact you don’t really have “R&D” because you have to put things into production with real data, just to see if they work!
I think Presize is a great example of the “everyday” optimizations AI will make in our lives.
(Ethics Policy: These opinions are 100% my own as an independent observer and educator. I don’t own stock in guests’ companies or their competitors, nor do I get paid by them in any form for any reason at the time of publishing, unless specifically stated. Episodes are also not intended to be an automatic endorsement of any company or its products and services.)
