Database Reference
In-Depth Information
For us, Thursday is an unwritten deadline because we all want to make sure we
have something for show-and-tell. There's only 5 slots so it might be the case
that you don't get to show your work, but the goal is to have something, even
if it's a couple of figures. Transparency into one another's work—the progress
and the obstacles—is really helpful.
Immediately following show-and-tell we “retro,” short for retrospective, which
for us is basically just happy hour. At retro we talk about how the week went
in terms of productivity, what worked, what didn't, and how to be better next
week. Regardless of what we talk about, we see it as a time to get together as
a team to reflect, or just chat.
Gutierrez:
How do you come up with ideas for things to study or analyze?
Shellman:
We develop ideas internally or they come from our collaborators
in the company. One of the coolest things about our group is that we collabo-
rate with people from many different parts of business. For example another
data scientist on the team, Elissa Brown, has been working in the area of store
fulfillment. In addition to our warehouses, each Nordstrom store can operate
as a mini fulfillment center. Elissa's work involves forecasting expected fulfill-
ment requests so that they can be efficiently distributed across stores and
hopefully reduce the time it takes to get products to the customer.
We also develop our own projects. For example, we created a recommen-
dation strategy called My Color Trends that is an interactive visualization of
a customer's Nordstrom wardrobe through the lens of color. The recom-
mender is an interactive way to visualize your own palette and explore the
colors in the products you bought. My Color Trends is also a tool for custom-
ers to find new products that are a precise match with selected colors, or
items that complement. One of the features I built was a strategy that rec-
ommends products in complimentary colors. What is considered a comple-
mentary color is debatable, but I got around the issue all together by creating
associations rules from the most commonly co-occurring colors in the fabrics
themselves. The great thing about that strategy is that it's data-driven so I
don't have to assign rules that are inflexible and arbitrary. When seasons and
trends change, the association rules will adjust to reflect the shifts in taste.
Some projects that start out with collaborators have to evolve to find a
good fit. For example, the beauty replenishment work I mentioned previously
started out as tool for beauty stylists and evolved into a personalized e-mail
campaign It's been great because we have the freedom to work on that and
be able to change according to what we think is best. Fortunately, we've had
really good initial success, which has made it easy for us to go out on a limb
to create new things. We didn't mess it up on the projects we were asked to
do, so now we have the latitude to do some crazier work.
