Database Reference
In-Depth Information
Through looking at the data and seeking to understand these questions, we
have developed a better understanding of the different groupings of people.
We have women who come to the web site often and never rent, and then
we have other women who rent every weekend, as well as some groups who
display behavior in between. So it's very helpful in understanding each group's
dynamics. By understanding their behavior, we can tailor their experience to
the right platform at the right time.
Gutierrez:
What have you found to be an interesting behavior that people
display on different platforms?
Smith:
As a data scientist, there's two responses. On one level, I really care
about the data, so I just want to see what they're doing, and I'm less concerned
if they buy. Obviously, on the second level, I do care if they buy and get to rent
their dream dress, so I find it really interesting how the data describes the
decisions and how they get made. For instance, I think it's interesting that, no
matter what we show our users, they're going to find their dress. They have
their own way of getting to whatever they're renting.
I still don't quite understand the thought process. We can change one page
and do tests, but it's not going to really do anything in an aggregate way. So I
think that's really interesting. My assumption is that they're just doing a lot of
filtering or using other pages than the main ones to get to their dress. There's
not even much of a conversion difference between whether they land on a
particular page that we think is most important or any other page. So I think
that it's really neat to be able to look at the data in an aggregate form from
disparate sources and work on understanding the behavior.
Gutierrez:
What specifically excites you about understanding the behavior?
Smith:
One of the things I was—and am still—very excited about when I
came here was the idea of how to represent someone's taste in dresses. After
all, the physicality of the dress involves so many individual tastes in fabric,
color, shape, fit, style, and other attributes. If we are able to represent some-
one's taste, just think of how cool it would be to understand. I think we're just
getting to the point where we can expose that really easily.
We're not quite at the point where we can break it down by color or any
discernible physical attribute of the dress. However, we have found that there's
some type of latent variable or variables that is able to represent someone's
taste in a dress. We just don't know what those variables are explicitly yet.
At this point, we can't just say, “Oh, it's because they're pink and flirty-looking
with ruffles.” That's a bit farther away from where we are because what does
“flirty-looking” or “trendy” actually mean, and how do we go about categoriz-
ing them in a form that is stable over time.
