Like many interns at Wayfair, Shima Dezfulian works on projects that have a high impact on the company’s operations. More specifically, Dezfulian is part of a Wayfair’s finance and algorithms team that leverages machine learning to make predictions about Wayfair’s profitability.
Machine learning models developed by the team that Dezfulian works with help Wayfair estimate the gross margin made in a short time interval after the order is placed.
Typical gross margin calculations involve fundamental costs and revenue involved with selling an item to the customer. In order to estimate the landed gross margin, Wayfair needs to consider the revenues and costs that are associated with incidents such as cancellations and returns.
“To calculate landed margins, my team develops models that take into account the revenue and cost of products,” says Dezfulian. “The predictions provide internal business teams at Wayfair with crucial and immediate feedback about our level of profitability.”
Furthermore, Dezfulian’s team also develops and trains additional models for each of Wayfair’s private brands, such as Perigold, that include hand-curated collections to align with a variety of tastes and price points.
Ordinarily, the team that Dezfulian works with would have been able to predict the landed margin using supervised machine learning methods, where algorithms are trained on annotated data to make accurate predictions.
However, smaller catalogs might not have as much training data as the broader Wayfair selection of products. This can make the model training a far more challenging proposition since the relation between the landed margins and the different product features is typically non-linear.
“We’ve found that nonlinearity can still be captured by fitting a simpler model with fewer variables. Another approach we are experimenting with is to sample data from the Wayfair US catalog for training smaller brand catalogs such as Perigold.”
After performing an exploratory data analysis, Dezfulian realized that the data distributions in Wayfair US and Perigold catalogs are inherently different.
“As a result, I came up with a sampling strategy that takes this difference into account,” she says.
Dezfulian is currently pursuing a doctorate at Northwestern University. Her journey to Wayfair is the latest stop in a career that has covered diverse facets of math and engineering.
A lifelong interest in Math
Dezfulian remembers being interested in Math from an early age.
“I am the first engineer in my family,” she says. “My mother was always encouraging me to pursue a degree that would help me become an engineer, which is a highly sought-after profession in Iran.”
Dezfulian earned a bachelor’s degree in mechanical engineering from Sharif University of Technology, where she found herself being drawn to the more mathematical aspects of her applied engineering work. Upon graduation, she applied for a master’s degree in Lehigh University. She decided to focus on applied mathematics and control theory.
When she was a graduate student, Dezfulian began to increasingly hear about machine learning. She was intrigued by how the field had the potential to have a transformative impact in a large number of areas such as natural language processing and image recognition. She completed a machine learning course hosted by Andrew Ng, the CEO and co-founder of Deelearning.AI.
“The course sparked an interest in machine learning that has continued to this day,” she says.
After her masters, Dezfulian applied to Northwestern University to pursue a doctorate in Industrial Engineering. Her research is focused on developing optimization algorithms, and is closely related to the work carried out by scientists in machine learning.
In her third year, Dezfulian decided to pursue an internship at a company.
“I had been thinking about where I wanted to pursue a longer-term career: in the industry or in academia. An internship seemed like a great next step to help me arrive at a decision.”
An impact at Wayfair
Recruiters from Wayfair sent Dezfulian a coding challenge within a few days of receiving her application. After she cleared the challenge, she went through a round of behavioral and technical interviews with leaders from Wayfair.
“Everyone here is really nice to you. All the people I have worked with are incredibly helpful. They just want to see you succeed.”
While she didn’t know what to expect from an internship, Dezfulian says that she has been surprised by the scale of real-world impact she has been able to have at the company.
“I had heard of interns at other companies being assigned smaller tasks,” says Dezfulian. “But that’s not the case at Wayfair. When I joined, my manager had developed an onboarding document that was made just for me. I joined the finance and algorithms team, and it was immediately made clear to me that I would be contributing to the goals of the broader team in a very real-way.”
Dezfulian says that managers at Wayfair set interns up for success: she has two meetings with her manager and a mentor every week, where she can ask questions, and conduct deep dives into different approaches to advance her work.
“Everyone here is really nice to you,” she says. “All the people I have worked with are incredibly helpful. They just want to see you succeed.”