Database Reference
In-Depth Information
Gutierrez:
Do you see yourself working in data science in your 40s?
Shellman:
Well, the thing about data science is that it's almost a catchall
to the point that it's meaningless. The reason that almost everybody starts a
data science talk with a slide discussing “What does it even
mean
?” is that it
almost means nothing. A data scientist to me is a person with a certain set
of quantitative and computational skills that are applicable across different
domains. So as a data scientist, even if I don't have the domain expertise I can
learn it, and can work on any problem that can be quantitatively described.
I can almost guarantee that I won't be in fashion retail in my forties, but I'm
sure I'll be working on something that relies on data and using similar tech-
niques and methodologies.
Gutierrez:
How would you describe your work to a data scientist?
Shellman:
I build the recommendation engines like the ones you're used
to seeing all over the web, and sometimes I do it with really unique data,
like transactions involving personal stylists in our brick-and-mortar stores or
color trends from fabrics.
Gutierrez:
What have you been working on recently?
Shellman:
Over the last year and a half I've mostly worked on Recommendo,
building new algorithms and the real-time scorer. For the last couple months
we've been working on a follow-up to Recommendo that will offer customer
segmentation as a service. We're calling it Segmento and we've already
conducted some initial tests with the email marketing team.
Segmento is an internal tool for creating custom audiences partly inspired by the
Facebook audience builder tool. For our initial test we constructed customer
affinity scores for certain brands and categories of products from past purchase
activity and product page views. I've been working on a classifier to assign similar
customers to pre-defined groups based on demographics and purchase behavior.
Gutierrez:
Why are recommendations important to Nordstrom?
Shellman:
For starters, recommendations play to Nordstrom's strengths,
because we have data other retailers don't. We've got those lovely brick-
and-mortar stores! The first recommendation engine I worked on is called
Our Stylists Suggest, and it works by analyzing transactions that occurred in
our full-line department stores and were facilitated by a personal stylist.
The idea is that our personal stylists are the best recommenders we could
ever hope to find and are fantastic at building cohesive outfits for our customers.
Our goal was to emulate those stylists online and build a recommendation
strategy to help our customers coordinate their wardrobe.
Our Stylists Suggest is a great example of a recommendation algorithm that's
unique to fashion data and takes advantage of the deep expertise of our stylists.
It's a competitive edge against strictly online retailers because it's a customer
experience they can't easily replicate.
