The secret to solving the fashion and data conundrum? Conversation (and tea).
When I joined EDITED as a data scientist, the challenge of solving new problems for the fashion retail industry was so exciting. It was also, I admit, a little scary.
I’d worked with data for years, but I really hadn’t spent much time thinking about fashion retail. Now I was going to work for a company whose entire goal was to help fashion retailers use data to do business better. I was worried that my work wasn’t going to help anyone if it didn’t connect with the real-world problems it was supposed to be solving.
To make things even more interesting, I would be working with a data manager whose skills were, for the first time, totally different than mine. Instead of a background in statistics, maths or computer science, she had spent years in fashion retail. At first that sounded like an odd pairing to me, but very quickly I realized it shouldn’t have.
From day one and every day after for the last two years, our contrasting skills have been at the center of a really interesting and unique collaboration. Recently, we sat down over tea and discussed the impact it’s had on EDITED.
But first let’s introduce ourselves:
Sophie (right): The data manager at EDITED. My background is in buying and fashion forecasting.
Paulo (left) : I’m a data scientist and engineer with a background in data, analytics, machine learning and computer vision.
So what have been some of the advantages of working together?
Sophie: Let me just start by stating the obvious: if we both came from fashion backgrounds, without the slightest understanding of data science, it wouldn’t be going so well.
Paulo: But on the other hand, without the fashion industry context, the analytics wouldn’t be as useful to the customers.
I can analyze data in so many ways, but not all of them would be interesting to the client. That’s why it’s so important to work closely with someone who has a retail background and understands the problems we’re trying to solve. It means we’re always working on something that’s really useful for our users.
Sophie: That’s one of the best parts of my job. I love that I can say, “There was something we did when I worked in retail that was really hard, let’s see what we can do to make that easier.” Then we just do that. It pushes everything forward so much more quickly. I imagine that if people didn’t work together like this no one would be able to come up with the best solution to our customer’s problems. And we’re always thinking from the point of view of, “what’s going to make their lives easier?”.
I think our ability to organize the data in a way that makes sense to our customers goes a long way towards establishing trust and showing that EDITED was built to make their lives easier. It’s always an interesting challenge to take an idea and develop it in a way that people can take onboard and integrate to their processes.
What have we learned from each other?
Paulo: One big thing I never understood about retail before is how important the concept of newness or freshness is in the fashion retail industry.
“I’m one of those people who basically just buys new jeans and check shirts when holes start appearing in my current jeans and check shirts.”
I’d never stopped to think about how important it is in this competitive market to constantly come up with new products, and how retailers are sometimes measured against each other based on that alone.
Sophie: It’s interesting that you bring that up. Retailers are always paying close attention to the mix of wardrobe staples like jeans and check shirts with more trend-led items.
Check shirts and jeans are perfect examples of continuity items, something you, as a retailer, might want to stock year-round. They may never be your marquee bestseller or anything exciting, but their consistency can make the business a lot of money over time. Newness can work around these continuity items offering people chance to jazz up their wardrobes.
It’s really useful that we’ve made it easy to distinguish the newness from the staples. That wasn’t really possible before we did it. For example, if someone was looking at a retailer’s website they wouldn’t be able to tell if something’s been available for one hour, one day, one month, one year or ten. They wouldn’t know how the prices have changed over time or how many times things have been restocked. Of course, now our customers are able to know all of that.
What else?
Paulo: My research mainly focuses on computer vision. Right now, a lot of computer vision research out there is really broad and general purpose, but in our problem space we need to go deeper with things. That’s something I didn’t initially expect, but it’s also been something really fun to work on.
For me, a bag used to be just a bag.
But now I can tell bucket bags from cross body bags, shopper bags, tote bags and so on. We’ve had to figure out how to train a computer model to understand those same visual differences, which has meant teaching the model to identify all the detailing and patterns on the bags as well. Things like houndstooth, shibori, damask…. all words that meant nothing to me until very recently. I guess it makes great party conversation.
Sophie: It’s funny you say that you can use that stuff to talk to people at parties because I feel like I’m the exact same way but with data science.
Honestly, I’m constantly looking at people’s clothes and thinking to myself, “Oh god it would be so cool to analyze that with our apparel classifier and see how it classifies it.”
Unfortunately, I can’t say anything because that would make no sense to anyone outside of work.
When you first explained to me that we could train a machine to classify garments by looking at a picture that even a person would find hard to process — like how different patterns are put together — I thought it was so cool. That’s something humans can’t do that easily, especially with the scale and speed that a machine can.
Paulo: Yeah, the fast scale is an important point. Our garment classifier is another great example of that since it’s doing something similar. It looks at every product that goes into EDITED and puts it into a category — tops, outerwear, bottoms...
If you think about it, that’s definitely something a human knows how to do, look at a picture and identify what it is. But there are 100 million products in our database. It would take an army of people ages to get through it all. With machine learning we can classify all the products in a few hours and have all that data (pricing, history…) ready for analysis right away.
Sophie:
And that’s something we wouldn’t have thought to ask for in retail because we didn’t know we could.
See I really like what we do because we’re teaching machines to think like a human, without making the human redundant. A machine can gather huge volumes of strategic data so quickly and condense it into insights that would be really, really laborious to attain manually, but our users still need the know-how to understand what the data is showing. The data is still there to support decisions, not automate them.
So what should we work on next?
Sophie: Well I was just telling you the other day how I’d like to find an easier way for people to get runway insights during fashion weeks without manually poring through images day and night trying to figure out which styles appeared most frequently.
Paulo: Like a trend classifier?
Sophie: Something like that. What do you think?
Paulo: That sounds exciting. I can think about it, but I’m going to need more tea first.
