I recently caught up with Mary Cameron, Data Scientist at Tophatter, to get her compelling insights into how Tophatter uses the principles of data science. She also delves into her life as a data scientist at a dynamic and growing company. Mary received her PhD from Stanford University and leads data science at Tophatter, Inc. At work, she uses math, computers, and data to solve problems. Tophatter is a pioneer in the mobile discovery shopping category. With more than 12 million users around the world, the company is on track to grow over 100% in 2017, selling $1M of product a day.
Daniel D. Gutierrez – Managing Editor, insideBIGDATA
insideBIGDATA: Can you tell us about how Tophatter uses data science, for example understanding how AI and machine learning can help personalize a mobile shopping experience?
Mary Cameron: Tophatter is a discovery shopping platform where users can view and bid on jewelry, electronics, apparel, home, beauty and other items in online auctions in real time. Each auction is fast–about 90 seconds–creating the engagement of a game. We have about 12 million users around the world and the company is on track to grow over 100% in 2017, hitting about $350 million of sales this year.
Ultimately, we use data to describe our users, describe our products, and model the interactions between those two, to make the process more enjoyable for both our buyers and our partner sellers. For our sellers, it means carefully selecting items for auction that encourage competitive bidding, but still allow shoppers to get a great deal. For our shoppers, it includes predicting which products they may be interested in, based on items they’ve previously liked or what similar shoppers have liked in the past.
It’s also important that we maintain product variety, so that our customers are delighted to see new items each time they come back to the app or website. We’re doing a good job if shoppers think “I didn’t know I wanted this!” There are a lot of moving parts, but ultimately, the better we can understand and predict the patterns in our marketplace, the more enjoyable the experience can be. Here, data science can be a very powerful tool.
insideBIGDATA: Please tell us a little about your background and how you came to practice data science at Tophatter. What’s your typical day like?
Mary Cameron: I was a math major in college because I didn’t know what I wanted to do for a career, but I liked calculus and I loved puzzles. The longer I was in it, the more I realized the power of being able to solve big problems with computers, rather than a pencil and paper. I went to Stanford for a MS in computational math, and stayed to apply those skills to climate, pollution, and renewable energy modeling for my PhD. After graduation, I wanted to gain hands-on experience so that I could continue to grow as a modeler, and build upon the skills that will help me solve big problems. I was drawn to Tophatter because I met so many bright people with backgrounds and skills that complemented my own.
As a data scientist, it’s important for me to be able to translate business problems into practical solutions that have meaningful impact. A typical day for me includes working with people throughout the company to define the scope of a problem, and put that into actionable metrics. Are we trying to gain insight, predictive power, or both? Does an algorithm need to run quickly on large amounts of data, or can it run updates weekly? A model and its implementation need to fit the problem, and good communication avoids having to struggle through tasks or redo them later. I’ll then spend time coding and testing different models, and researching new methods and tools that help me keep up with the field. Finally, I love data visualization and have a lot of fun coming up with intuitive ways to present complex issues. Right now, I’m creating an interactive visualization that tracks the number of bugs our engineering team fixes, using a lot of colors, sound effects, and javascript. It’s not mathematically sophisticated, but it’s great being able to play with data to make everyone’s job a little more fun.
insideBIGDATA: As a woman, can you give us any observations about the importance of diversity in the profession?
Mary Cameron: [Laughs] There are two ways to answer this. As a data scientist, I can say that any time you reduce a person down to a set of attributes, you risk pigeonholing them into a category that doesn’t fit. As a shopper, I can say that there are a lot of reasons I shop, sometimes for myself, and sometimes for others. Part of my job includes identifying and avoiding the dangers that can arise when we say “Who is this shopper, and what do we think they’d like?” A product selection too closely tied to who an algorithm thinks our shopper is has the potential to alienate our customers and turn away revenue. It’s bad for business.
The second answer stems from my personal experience of sometimes feeling like an outsider in my field. Fewer than 20% of the software engineering positions in Silicon Valley are filled by women (https://github.com/triketora/women-in-software-eng), which is silly when we’re half the population. Still, being one of few women in the room is something I’m fairly sensitive to, and it has influenced a lot of my career decisions. I’ve benefited a lot from open discussions about diversity, unconscious bias, and the imposter syndrome through good mentoring and organizations that support women in STEM. It’s important to me to stay involved in those communities, both for my own growth and to help young women who may be struggling with the same issues.
To attract top talent, a company needs to appeal to the full workforce, and a different background and viewpoint is just as valuable an asset as those listed on a resume. Supporting diversity isn’t just good for society; it’s good for business.
insideBIGDATA: What’s out there on the horizon with respect to the use of data science at Tophatter? Any plans for the future?
Mary Cameron: Mobile commerce generates a lot more data from lightweight actions than other shopping formats like catalog shopping, TV, and even desktop website usage. This data provides businesses like Tophatter the power to personalize unique shopping experiences for each user.
Right now, we’re working on different ways to better describe and compare our products, which is what drives our recommendations and auction selections. We have ~2 million unique product listings, so being able to group like items means we gain that much more insight about what our users like to see and what performs well on our site. Additionally, we’re working on methods to describe our user-marketplace interactions so that we can increase the product variety on our site while maintaining the competitive bidding that makes Tophatter fun and affordable.
The secret to retail is all about facilitating discovery and giving people what they want. We’ll continue working and building a team of diverse data scientists and engineers on making that vision a reality.
Sign up for the free insideBIGDATA newsletter.
I currently sell on tophatter and finding it more increasingly frustrating once this new Algorithm was implemented. Now my items are grouped unfairly which is loss of sales. We are unique sellers now pooled into a group cluster of sellers from China. Tophatter is using a dot system intending to rate products but does not recognize new products and will automatically classify them as low sale items.. What the system does not factor is the time of day buyers are there, nor the number of buyers that might by present. Items can sell great in the morning and not so well during the late night hours or early morning.. Human input needs to be once again a part of this operation.