Levi’s is best known for its denim, but the company’s master plan is to redefine itself with artificial intelligence, machine learning and data science. And the apparel company has 168 years of data to become more efficient, predict and create trends and improve the customer experience.
ZDNet caught up with Dr. Katia Walsh, Chief Strategy and AI Officer at Levi’s, to talk about implementing AI, machine learning and data science at a 168-year-old company during a pandemic. Here are a few highlights. The full conversation is in the video.
Where an AI and data science team sits in an organization. Walsh said that AI is a department that covers multiple units and is a horizontal function much like finance, technology and human resources.
This team is very new to the company. I barely got from London to San Francisco myself as the leader and founder of the capability, and then COVID happened. So we started the year with 12 people, including me and my assistant. We were just learning to educate the entire enterprise on what this combination of digital data and AI is and what it can do for the company, and then March 16th, the lockdown in San Francisco happened. And we all know the challenges that we have been encountering for the last year. But the last year was a great opportunity to really show what digital data and AI can do for a company.
The COVID-19 crash course. Like other technologies, the COVID-19 pandemic accelerated plans. Walsh said from March to August 2020, Levi’s saw a sprint where AI and data had to be used for “anything from improving the customer experience to delivering internal operating and operational efficiencies, and also possibly looking into new revenue models and business models for the company.”
E-commerce and shipping. Walsh said Levi’s saw a surge in e-commerce sales and the company moved to ship from the stores closest to the consumer. She said:
Using AI, we devised a machine learning engine that optimized a number of different variables, including what items each store had in its inventory, how far or close it was from the specific consumer placing the order, how much it would cost to ship, whether the item that was ordered was going to have to be discounted later if it went out today, et cetera…
We were able to use data on climate and weather and epidemiology models and financial and market outlooks. So you know how I talked about the three parts of this flywheel; digital, data, and AI. What makes this particularly useful is that it uses more data than before, which gives us more points, more perspectives, more variables, and then we’re able to apply machine learning, which then makes the model even smarter and to deliver even better.
Using 168 years of data. Levi’s has 168 years of data and the company considers itself one of San Francisco’s original startups. Walsh said that rich history and data set can inform what products will thrive in the future.
In the case of Levi’s all the products that the company has created and manufactured in the past 160 years are data. The Bing Crosby jacket that he wore in Canada is data. Once it’s photographed, and that photograph is digitized, that’s data. The Einstein jacket that he was photographed as man of the year by Times Magazine in 1939 that we created a replica of in the past year, that’s data. So we are now using images of products to predict demand for new products based on using computer vision that can tell us based on similarity between certain products that have been sold in the past new products that have never been sold, what the demand for new products would be. So the opportunities are absolutely endless when it comes to data and Levi’s.
We are absolutely predicting right now what demand for products will be like. The further you go into time, the less accurate the model will be because there are just so many unknowns that happen to accumulate as time goes by. And we are looking to predict demand in the next half of the year, in the next month, in the next three months.
The role of algorithms in product design. Walsh said:
Well, product design is a very creative process. I’ve worked in financial services, I’ve worked in telecommunications, I’ve worked in technology. This is my first time leading AI and helping drive digital transformation in the creative company, in a fashion company. It is incredibly creative. It is a highly imaginative process. What we are doing is partner with designers, partner with planners, planning is the original data science function in a company like Levi’s and retail and apparel, and bringing the latest tools and this combination, this flywheel of digital data and AI to be able to drive demand, to predict demand, to optimize costs, and to also really deepen the connection with consumers.
People, processes, privacy matter as much as the tech stack. Walsh said that building out AI capabilities has four building blocks.
I always start with the people. Yes, tech stacks are very important, but if we don’t have the right people and the right number, and I’m not talking about an army of people, but people who have the technical skillset, but are also entrepreneurial, good communicators, focused on business priorities, able to partner and think about the future. So people are very important. Processes are quite important to. Not to be bureaucratic, but we are incorporating agile ways of working in the company. We are driving a great deal of attention on privacy.
Privacy is always important. It’s particularly important when you’re dealing with data and AI. We talk about responsible and ethical AI. So we have a code of conduct when it comes to data and AI in the company. Data of course, itself, very important. We now have more data than ever before, certainly internal data from our own operational systems, but also external data from partnerships or from mobility patterns or from social media, always with permissions in place. And of course technology is the fourth building block, also important. Yes, we use open source tools. We also partner with cloud providers, from AWS to Google Cloud Platform to make sure that we the most advanced tools that we can find.
Build vs. buy. Walsh said Levi’s primarily builds its own algorithms and data science approaches. “It’s not that we don’t like to start with something that has been done, but when it comes to retail with the exception of the Amazons of the world, this is a very new field. We are now cultivating a new kind of professional data scientist or machine learning engineer that knows retail and apparel,” said Walsh. “So, for that reason, we’re starting from scratch and we’re creating our own custom algorithms.”