The Business of Fashion
Agenda-setting intelligence, analysis and advice for the global fashion community.
Agenda-setting intelligence, analysis and advice for the global fashion community.
Before she enrolled in Levi’s AI bootcamp, Shelby Greeley, an analytics manager in the company’s wholesale e-commerce business, didn’t know how to code. It’s a foundational skill in computer science, the equivalent of knowing how to read and write.
“I spent a lot of extra hours outside of class catching up over the weekend to make sure that I was truly processing this,” Greeley said in an interview in October, about two weeks into the program. “It definitely was a steeper learning curve than I anticipated.”
Greeley was in the second group of Levi’s employees to go through the eight-week, full-time course, which opened to its first group in May and is designed to train people in roles across the company in AI. To date, 101 have completed it. Like Greeley, most had no coding experience beforehand.
Nearly six weeks in, though, Greeley was starting to write code without having to refer to any documents. The real turning point, she said, came when they started using real Levi’s data. Recently, she and a group of fellow bootcampers presented a functioning predictive model to determine the size distribution best for a specific style and colour combination. Levi’s is looking to put the model into production, she noted after finishing her training.
The bootcamp, which opened its first session in May, represents one of Levi’s most ambitious commitments to AI since enshrining it as part of the company’s business strategy in 2019. Far from the vision of AI as sentient computers, this version is focused on performing narrow tasks to provide a competitive business edge through smarter, data-based predictions. On earnings calls with investors, Levi’s executives have touted AI as delivering meaningful improvements in its margins and helping to better manage jobs as varied as determining prices and promotions, forecasting demand and fulfilling e-commerce orders from stores.
To keep expanding its use of AI, though, Levi’s needs more data scientists.
“There’s a very limited number in the world,” said Dr Katia Walsh, Levi’s senior vice president and chief strategy and artificial intelligence officer. “Everyone is fighting for them.”
Those that are interested in working for a fashion company instead of, say, a finance or tech firm often aren’t familiar with the intricacies of the industry. So Levi’s decided to create its own.
What Levi’s Employees Learn in AI Bootcamp
The bootcamp was Walsh’s idea. A data scientist with more than 20 years of experience in the field, she joined Levi’s in 2019 after time spent in finance and tech.
“I’ve been doing this longer than it has been sexy,” Walsh said.
At prior companies, Walsh had run AI bootcamps for employees with backgrounds in coding, engineering and statistics, but she wanted the Levi’s program to be open to a wider group. To get in, candidates must complete a demanding application process that tests their logic, problem-solving skills and perseverance.
In a session to kick off this round, an instructor explained the course would cover data collection, model building and prediction. The first step is to learn Python, an open-source coding language. A later session on Python covered elements such as objects, strings, arrays, lists and loops. For the uninitiated, it’s a foreign language: at one point a student asked how they know if their kernel is dead, referring to a mechanism that executes a piece of code.
Students then move into machine learning, the branch of AI focused on approximating human decision-making in computers, allowing them to perform tasks without explicit instructions. It’s useful for making predictions from large datasets, such as forecasting how many jackets a company should produce in a certain colour and automating labour-intensive processes. Eventually, they work with actual Levi’s data on a real-world use case.
The course concludes with them collaborating in groups to create a predictive model they present to the company’s executive management. Afterwards, graduates are paired with mentors so they can continue to learn and develop their skills.
To learn AI in a short time is not easy, though.
“Both coding and doing machine learning are new ways of thinking,” said Daniela Rus, who runs the Computer Science and Artificial Intelligence Laboratory at Massachusetts Institute of Technology. That said, it is possible in a short period, she emphasized.
“How to put it? Intense,” Ricardo Lemos, a Levi’s senior retail operations specialist, said of the bootcamp after having recently completed it. “We still have so much to learn, but it was already amazing to be able to do what we did.”
Wringing Real Value from AI
AI’s origins date back roughly 70 years, but the use of AI by businesses has only taken hold more recently. The past 18 months in particular have seen its adoption surge. In fashion, it often powers e-commerce chatbots and helps companies personalize recommendations to shoppers. Mytheresa is using AI to predict and foster customer loyalty.
For Levi’s to wring real value from the bootcamps will take more than getting graduates to whip up new ideas, however.
“The more common challenge that we find is scaling up that pilot impact into something that actually moves the needle in the business,” said Michael Chui, a partner at McKinsey Global Institute, the consultancy’s business and economics research arm.
“We start in one market and then we scale very quickly, but everything we do, we measure,” Walsh said of Levi’s existing AI projects.
A goal of the bootcamp is to produce results that will scale. One it’s looking into is the work of a graduate of the first bootcamp, Danisha Jefferson, a computer technician in the company’s Las Vegas distribution centre. By using data from work orders, she and a mentor devised a model to predict when equipment will fail, a problem Walsh says causes downtime that can cost millions over time.
Currently, the model is 60 percent accurate and Jefferson is continuously improving it. The aim is to eventually deploy it to other distribution centres and use it for preventive maintenance.
There are other hurdles that come with AI. Predictions are probabilities, not guarantees, and only as good as the data they’re based on. Most of the work of AI is collecting and cleaning data, something Walsh acknowledged is a challenge for any company using the technology. If companies aren’t careful, their AI can reproduce harmful biases. Levi’s recruits a mix of employees from different racial and ethnic backgrounds for the bootcamp to try to prevent the problem.
Shifts may be needed in a company’s culture, too.
“Culture, organization, process change — all of those things are necessary,” Chui said.
Predicting equipment failures, for instance, can affect the employees who handle repairs, and siloed AI departments don’t allow companies to reap the benefits possible, Chui noted. In his experience, bootcamps that broaden understanding of AI across the company are among the more successful ways to create change.
That’s another of Walsh’s goals. Though some graduates join the data-science team, most return to their previous jobs. It allows Levi’s to embed that newfound AI knowledge around the organization, gradually reshaping the company culture as more employees complete the bootcamp. It already has the next rounds planned for 2022.
Lemos is among those employees who returned to his regular tasks after the course. His day-to-day work entails producing numerous detailed reports on store performance. Early in the bootcamp, his expectation was that it would help improve this reporting, making it more effective and efficient. He also wanted to use AI to spot patterns in the data he wasn’t seeing on his own.
About six weeks in, after the course got into the basics of machine learning, AI felt “less scary,” in Lemos’ words. Once it was over, he seemed confident it would make him better at his role.
Greeley was similarly positive.
“I definitely think there are some great opportunities for us, both on the true AI side, so thinking about something that’s predictive, but also just in terms of process,” she said. She’s already speaking with her manager about how they can implement AI.