Database Reference
In-Depth Information
Gutierrez:
How does the Discover Similar Subscribers tool work?
Foreman:
When you think about the billions of email addresses that we have,
you can think of them as a globally connected graph where they're all con-
nected to each other based on mutual subscriptions or interests. With all of
these connections, I can then figure out who “lives” (in an interest sense) next
or near to each other based on proximity calculations. Basically, you can think
of it using the concept of neighborhoods—that is, which emails are neighbors
to each other. So a user with a list of people, who have already subscribed
to their content, can pull different segments based on the globally connected
graph. They can then create and name these specific segments of their list.
Going back to Mandrill, that's what we did—we created a segment, based on
software development as an interest. We then sent our product announce-
ment email specifically to that segment saying, “Hey, just so you know, Mandrill
now exists. It's really cool if you want to use it.” We got amazing engagement
from that email. And not only did we get great engagement from it, we also
didn't bother all of the other people who wouldn't have wanted to know
about the new product. That's where I see the power in having this kind of
data and providing this kind of segmenting capability to users.
Gutierrez:
How do you describe your work to someone who's not familiar
with the math behind it?
Foreman:
For those who aren't familiar with the math, I talk about how
I serve three roles. First is that I build data products, second is that I am a
translator between customers and our data science team, and third is that I
am an ambassador for MailChimp. In regards to building data products, I do
this for two types of customers. The first type of customer is the MailChimp
user. The data products for this customer tend to be external-facing products
or artifacts like reports or blog posts. We do research on our global data set
and provide artifacts or products, like Discover Similar Subscribers or Send
Time Optimization, to external customers.
The other set of customers are actually internal customers. There's a huge
need at MailChimp across teams to understand the data we have, especially
around understanding our customers and their unique characteristics. Those
are needs the data science team can facilitate, so we also build tools for these
internal customers. These tools range from things like anti-abuse models to
support scheduling tools to likelihood to pay models.
The second piece is being a translator between internal/external customers
and the data science team. Being a translator helps me to figure out what to
build. I want other people's input; I don't just make our workload up. That's
actually scary when some analytics teams do that. They just kind of show up
and say, “Well, we should build this,” and they haven't talked to anybody. They
just thought it would be cool to use some tools. So a large piece of what I do
is talking to people, understanding their needs and their problems, and trans-
lating that back to technical folks within MailChimp.
