How Levi's Is Using AI To Accelerate Its Design Process

Integrating data with speed generates more monetizable insights faster. JL 

Venture Beat interviews Levi's Design Coordinator Ron Pritipaul:

AI can automate tasks that humans shouldn’t be spending time doing (such as) color-matching threads to fabrics. An algorithm can help with in-store product recommendations by building an outfit based on past shopping history and what’s trending. We’ve been producing jeans for 150 years. Each piece of denim is data. We have an archive of thousands of jeans and how each one fades. Load that into a program, start unlocking insights. You can throw data programmers at a problem, but if you give non-technical people the tools and an understanding of what machine learning can do, there will be problems that are really easy to fix with artificial intelligence.

As ubiquitous as machine learning is in the enterprise, Levi’s might not be the first brand that comes to mind when you think of AI smarts. As a company that has been producing jeans and other denim apparel since 1853, Levi Strauss & Co. seemed to be doing just fine without the intervention of neural networks and machine learning algorithms. But like so many large companies, Levi’s has found plenty of uses for AI technology, from automating mundane tasks and analyzing denim-related data sets to helping its designers create new denim jacket designs. 

In 2019, Levi’s formalized its years-long flirtation with AI by hiring Katia Walsh as the company’s chief AI and strategy officer to lead its new global AI team. As part of her effort to integrate this bleeding edge tech into an established legacy brand, Walsh launched the company’s first-ever Machine Learning Bootcamp in early 2021. The intensive, eight-week program invited 40 Levi’s employees from across the organization to learn about machine learning, agile development methods and of course, how to write code.

Design coordinator Ron Pritipaul was one of the Levi’s employees who signed up for the inaugural AI-focused bootcamp, a decision that he says has had a major impact on the way he approaches design, collaborates across teams and thinks about the future. VentureBeat talked to Pritipaul to learn more.

VentureBeat: Tell us a bit about your role at Levi’s. How did a designer like you wind up working on machine learning algorithms? 

ADVERTISEMENT

Ron Pritipaul: I’ve been working at Levi’s for a little over three years. I’m a huge denim head. Growing up, I used to collect vintage jeans. In college, I created my own concentration about the history of denim. After I graduated, my dream was to work for the oldest and coolest denim company. So, I emailed everyone I could find at Levi’s asking for a job. I got a temp position doing manual labor around the office. Eventually, I worked my way into a position on the design team as a design coordinator. In that role, I assisted with helping the designers bring Trucker jackets and other products to life and did lots of little tasks to help out the design department. 

Just from being on that ground level, I knew that a lot of our processes were very manual and painful. Fashion is still a very analog industry. When I saw the Machine Learning Bootcamp opportunity posted on our internal website, I applied. That decision was completely transformative of how I saw myself and my career. I started developing algorithms and programs that automated work, allowed us to be more creative, and unlocked growth for the design department. Eventually, all of that work led me to coming into my new role as an associate data product manager for computer vision.

VentureBeat: Fashion and garment design may seem like an unexpected use case for AI. What did you wind up building and how has it impacted the design process? 

Pritipaul: A lot of people think AI is going to replace human beings or something. I don’t believe that to be the case. Fashion is a very intimate field. You’re clothing another person, which is probably one of the most intimate things you can do. It’s a very personal, human connection. Fundamentally, it’s always going to be humans at the heart of this industry, but AI can unlock growth and automate tasks that humans shouldn’t be spending time doing. For example, I spent a lot of time in my design role color-matching threads to fabrics. That’s a lot of hours that are wasted when we could just have a computer do that. So, I wrote a program to help automate our color-matching process to free up time, so designers can be more creative.

ADVERTISEMENT

I also did some neural network work that takes art and makes Trucker jackets. This unlocks a new form of creativity where people can be inspired by more than just what’s in Vogue or on Instagram, and instead take inspiration from art pieces or photographs and see how a computer might interpret it to help us create new designs. That’s something that can really shift the way you see fashion design and think about creative inspiration. I really see AI as a tool for creatives that lets them spend more time being creatives and develop more creative designs.

It was cool to work on these things, to think of new ideas, rapidly prototype them and just play around. There are so many cool projects that I worked on in the boot camp that now we’re turning into actual work processes.

What are some examples of boot camp projects that are now being utilized by the company?

Pritipaul: The neural network stuff that I’ve done turning paintings into jackets is definitely one. Another example is a Levi’s store manager who used her learnings to create an algorithm to help with in-store product recommendations. It can help customers by building an outfit based on their past shopping history, and also what’s trending generally.

With my design background, I’ve looked at data collection differently than how a data scientist might. We’ve been producing jeans for almost 150 years now. Each piece of denim is a piece of data. We have an archive of thousands of jeans and data on how each one fades over time. If we load that into a program, we can start unlocking insights about denim that no human could ever see. 

ADVERTISEMENT

A company like Levi’s must be sitting on other untapped troves of data. Are there any other historical data sets that you’ve considered running algorithms on? 

Pritipaul: We’ve been discussing things as a team like, how would we take our fabric library of all the fabrics that we’ve ever used and extract data from that to understand more about them? For example, how they react to different processes, washing techniques and things like that. If we’re talking about data sources, we have fabrics, we have sundries–things like buttons, zippers, etcetera — we have all these things. There are definitely insights that we can glean if we take that data and work with it, as opposed to just storing it.

AI can unlock so much insight in so many different areas, like sustainability. If we know our fabrics really well, and we know what happens when we wash them, we can start thinking about how we can optimize things, so our products are aesthetically beautiful, but also good for the environment. We can also use this technology to predict how things might look in the future when we’re producing them, so we don’t have to produce as many samples.

VentureBeat: What specific technologies, tools and platforms are being used here?

Pritipaul: In the bootcamp, we learned everything in Python. So there was a one-week introduction to Python. Then we went straight into machine learning. I’ve personally been using Pytorch a lot, which is for neural networks. My thread color-matching stuff was originally a neural network that I trained.

The [Trucker jacket] style transfer stuff was also a neural network, and it’s done by freezing different layers. The backend of that has been around since 2017, but what I did was just segment the images so that we are style-transferring only on the trucker jacket and not the background, small things like that.

ADVERTISEMENT

Most of what we’re doing is just writing Python codes that connect things. We’re using home-built tools and building out platforms ourselves. We’ve found that off-the-shelf tools are helpful to get us to the first stage, but after that, we need to build our own tools. That unlocks a lot of growth for us because we can build tools for exactly what we need. When we need image processing, instead of opening up Photoshop, we can write a Python script that processes the images exactly how we want as efficiently as possible and push through hundreds of images. 

VentureBeat: Beyond your own role, how do you think a program like this AI boot camp impacts the organization more broadly? 

Pritipaul: During the boot camp application process, employees can choose whether after the graduation if they’d like to be considered for a new role under the AI department or if they’d like to go back to their previous role with these learnings and figure out what they can do with ongoing mentorship and tools to take what they’ve learned and implement it. 

Since we all did the boot camp and are all on the same email lists, there’s a lot more cross department collaboration. It’s creating this big network of people who all speak the same language. I can work with IT, with merchandising, or with product developers, who have all been through the boot camp. So, we’re not as siloed as we were before.

VentureBeat: What do you think other companies can learn from Levi’s experience here? Any takeaways for tech decision-makers and other leaders in the enterprise? 

ADVERTISEMENT

Pritipaul: Not every company can do a boot camp, but every company can bring non-technical people into those technical conversations. The level of domain expertise that they can unlock is so much more than they can get from just putting a bunch of programmers on a problem. Someone from the boot camp who works at our distribution facility wrote a program for predicting when machines would break in the distribution center. That comes from the data, but it also comes from the insight of working there hands-on. If you have our data programmer, who’s based in San Francisco, trying to solve that problem in our plant, they’re not there with the machines. They’re not seeing what actually happens.

You can throw the best data programmers at a problem, but sometimes if you just give non-technical people the tools and the language to speak about these problems and give them an understanding of what machine learning can do, there will be so many things that you can accomplish. I’ve been sort of acting as a liaison for the data science department to design, merchandising, and product development. People have been coming up to me and telling me all about these problems that are actually really easy to fix with artificial intelligence, but someone with a strict data science background would never even think about these problems. Tapping into your domain field experts and really looping them into the design of things can really unlock so much potential for companies.