The Gap’s Data Science Director Has Tailored the Retailer’s Operations

Shoppers probably don’t realize how large a role data science plays in retail. The discipline provides information about consumer habits to help predict demand for products. It’s also used to set prices, determine the number of items to be manufactured, and figure out more efficient ways to transport goods.

Those are just some of the insights that data scientist Vivek Anand extracts to inform decision makers at the Gap, a clothing company headquartered in San Francisco. As director of data science, Anand—who is based in Austin, Texas—manages a team that includes statisticians and operations research professionals. The team collects, analyzes, and interprets the data, then suggests ways to improve the company’s operations.

Vivak Anand

Employer:

The Gap, headquartered in San Francisco

Title:

Director of data science

Member grade:

Senior member

Alma maters:

Indian Institute of Science Education and Research in Pune; Columbia

“Data science is trying to effectively solve problems that were previously unsolvable,” Anand says. “The technology is used to group similar transactions that look different on the surface. But underneath they are similar.”

Anand is an IEEE senior member who has spent his career using data science, artificial intelligence, and mathematical and statistical modeling to help businesses solve problems and make smarter decisions.

Last year AIM Research honored Anand’s efforts to transform the retail industry with its AI100 award, which recognizes the 100 most influential AI leaders in the United States.

A data scientist at heart

Growing up in Gopalganj, India, he set his sights on becoming a physician. In 2006 he enrolled in the Indian Institute of Science Education and Research (IISER) in Pune with every intention of earning a medical degree. During his first semester, however, he enjoyed the introductory mathematics classes much more than his biology courses. A project to design a statistics program to determine the best way to vaccinate people (pre-COVID-19) helped him realize math was a better fit.

“That was my first introduction to optimization techniques,” he says, adding that he found he really liked determining whether a system was working as efficiently as possible.

The vaccine project also got him interested in learning more about industrial engineering and operations research, which uses mathematical modeling and analytical techniques to help complex systems run smoothly.

He graduated in 2011 from IISER’s five-year dual science degree program with bachelor’s and master’s degrees, with a concentration in mathematics. He then earned a master’s degree in operations research in 2012 from Columbia.

One of the courses at Columbia that intrigued him most, he says, was improving the process of identifying a person’s risk tolerance when making investment choices. That training and an internship at an investment firm helped him land his first job at Markit, now part of S&P Global, a credit-rating agency in New York City. He created AI and mathematical models for financial transactions such as pricing cash and credit instruments, including credit default swaps. A CDS is a financial instrument that lets investors swap or offset their credit risk with those from another investor.

Anand, who began as an analyst in 2013, was promoted to assistant vice president in 2015.

Later that year, he was recruited by Citigroup, an investment bank and financial services company in New York City. As an assistant vice president, he developed data science and machine learning models to price bonds more accurately. He also led a team of quantitative analysts responsible for modeling, pricing, and determining the valuation of credit derivatives such as CDSs in emerging markets.

He left Citi in 2018 to join Zilliant, a price and revenue optimization consultancy firm in Austin. As a senior data scientist and later as lead data scientist and director of science, he led a team that built and serviced custom price optimization models for customers in the automotive, electronics, retail, and food and beverage industries.

“We used to estimate elasticities, which is a key component for pricing products,” he says. Price elasticity shows how much demand for a product would change when its cost changes. “The existing algorithms were not efficient. In a number of instances, it used to take days to compute elasticities, and we were able to bring down that process to a few hours.”

He was director of science at Zilliant when he left to join the Gap, where he oversees three data science subteams: price optimization, inventory management, and fulfillment optimization.

“In the fashion industry a vast majority of product assortments are continuously refreshed,” he says, “so the objective is to sell them as profitably and as quickly as possible.” Clothing tends to be season-specific, and stores make space on their shelves for new items to avoid excess inventory and markdowns.

“It’s a balance between being productive and profitable,” Anand says. “Pricing is basically a three-prong approach. You want to hold onto inventory to sell it more profitably, clear the shelves if there is excessive unproductive inventory, and acquire new customers through strategic promotions.”

Managing inventory can be challenging because the majority of fashion merchandise sold in the United States is made in Asia. Anand says it means long lead times for delivery to the Gap’s distribution centers to ensure items are available in time for the appropriate season. Unexpected shipping delays happen for many reasons.

The key to managing inventory is not to be overstocked or understocked, Anand says. Data science not only can help estimate the average expected delivery times from different countries and factor in shipping delays but also can inform the optimal quantities bought. Given the long lead times, correcting an underbuy error is hard, he says, whereas overbuys result in unsold inventory.

Until recently, he says, experts estimated transit time based on average delivery times, and they made educated guesses about how much inventory for a certain item would be needed. In most cases, there is no definitive right or wrong answer, he says.

“Based on my observations in my current role, as well as my previous experience at Zilliant where I collaborated with a range of organizations—including Fortune 500 companies across various industries—data science models frequently outperform subject matter experts,” he says.

Building a professional network

Anand joined IEEE last year at the urging of his wife, computer engineer Richa Deo, a member.

Because data science is a relatively new field, he says, it has been difficult to find a professional organization of like-minded people. Deo encouraged him to contact IEEE members on her LinkedIn account.

After many productive conversations with several members, he says, he felt that IEEE is where he belongs.

“IEEE has helped me build that professional network that I was looking for,” he says.

Career advice for budding data scientists

Data science is a growing field that needs more workers, Anand says. For people who are considering it as a career, he has some advice.

First, he says, recognize that not all data scientists are the same; the job description differs from company to company.

“It’s important to network with people to understand what kind of data science they are doing, what the role entails, and what skills are needed to make sure that it’s a good fit for you,” he says.

There are eight of what he calls spokes in data science. Each one represents a specific skill: exploratory data analysis and visualization, data storytelling, statistics, programming, experimentation, modeling, machine learning operations, and data engineering.

Continuing education is important, Anand says.

“Just because you’ve earned a degree doesn’t mean that learning stops,” he says. “Don’t just scratch the surface of a topic; go deeper. It’s also important to read fundamental textbooks about the field. A lot of people just skip that part.”

In data science, he notes, there are premade libraries such as SciKit Learn. If you don’t understand what the library is doing in the background or what’s under the surface, though, then you are just a user, not a developer, he says, and that means you aren’t building things.

“Data science needs a lot of developers,” he says. “The people who know things can build programs from scratch. We have a ton of users but fewer developers. And this field is going to be here for a long time.”

Reference: https://ift.tt/RVU8jK1