Subscribe
Share
Share
Embed
课代表立正的官方Podcast
深度访谈，有用干货，亲身验证的「真本事」
Superlinear Academy创始人，Maven Top AI Instructor
前Statsig布道师（OpenAI收购），腾讯副总监，Meta，Amazon；康奈尔经济学博士
社区：Superlinear.Academy
课程：ai-builders.com
个人：lizheng.ai
hey folks welcome to my channel today is
a special english episode
about why facebook is such a great place
for data scientists to work at
the content of this video is very
similar to my chinese version
so if you have seen that video feel free
to skip this one
but the reason i want to do it in
english is because
one the knowledge is pretty applicable
for english speakers as well
until because at the end of this video
i'm going to announce i'm leaving
facebook and i want to help
my org my team to hire my backfield so i
want to reach out to other english
speaking data scientists as well
because you might be interested after
hearing my video okay
let's get to it so first of all about
the conclusion facebook is a great place
for data scientists to work at
i know everyone's opinion is biased but
i try to collect a lot of information to
be
as unbiased as possible my personal
background is
i worked two years at amazon as an
economist
and then i moved to facebook when i came
to facebook
i was shocked by how important data is
in the products so all the products i
see at facebook
starts with data and ends as data and as
data scientists
you really have a lot of say in the
direction
strategy and outcome of these products
and people actually listen to you
by people i mean cross-functional
partners such as
engineers uh pms product managers
designers researchers and everyone like
people really respect data
and i really believe the work i do at
facebook as a data scientist
creates tremendous value for the product
after this discovery
i start talking to my industry friends
who work at different companies
especially the people who join facebook
from other companies
and people who leave facebook and went
to other companies
and to my surprise all the people
whether they came to facebook from other
companies or they left facebook and went
to other companies
they recognized the value of data at
facebook
and they all believe facebook is one of
the best places for data scientists to
work at
immediately we're thinking okay is it
because data scientists at facebook
are doing particularly different things
then we realized it's not the case
in most companies at least in most tech
companies
we all have good ideas of what data
centers should do
and what can we bring to the table but
reality is at different companies
are very different in some companies
people
say we want to have data driven uh
data-driven decisions
but they actually are just paying deep
service and
they are only using data to confirm
their opinions
instead of changing their opinions so i
went deeper into summarizing what's so
special
at facebook that makes data effective
and i summarized three key points one
is infrastructure two is the culture
and three is the organizational
structure and hopefully
this will help you establish a framework
of
how facebook did it okay without further
ado
let's start first infrastructure
speaking of infrastructure we have to
thank our data engineers
at facebook i think the ratio of data
scientists to data engineer is two to
one
and the data engineers made the life of
data scientists so easy
when i talk to data scientists at other
companies depending on the level of
infrastructure
they often spend fifty percent to eighty
percent of their time
trying to clean the data trying to get
the data they actually need
before they can do any analysis but at
facebook
we have specialized data engineers we
also have a lot of tools for software
engineers to build the logging
and then build the pipeline so the data
when we want to not analyze
they are right in front of us and in
next format in the right size
so all the time uh that other people
have to do
to clean the data is saved so we can
focus
our so data scientists can focus our
time in analyzing the important
questions
the other side of this infrastructure is
the automatic tools
are very advanced if i want to build a
dashboard i can directly do that
after running a query so if i run a
query i don't need to worry about what
machines i'm using
it's all automatic and elastic and after
i get the
results i can easily share my results or
i can do quick analysis or data
visualization and i can easily copy
paste into a dashboard
and it became automatic so so while we
can scale our
impact because uh it's there it's
automatic
it's visualized so more people can look
at it understand it and
change decisions based on the data or
based on the analysis
and two we get to automate a lot of our
work a lot of analytics in other
companies
uh their role is basically getting the
data and
support this ad hoc request at facebook
doing this ad hoc request is really a
data scientist's favorite to the cross
function teams
they should be able to do most of the ad
hoc queries themselves
it's not part of our core requirement
our core requirement
is to use scientific method
to analyze the problem uh think
strategies
drive the direction of the product and
prior prioritize our team's work
needless to say most of these work are
way more impactful than those ad-hoc
queries
okay so that is our infrastructure to
sum up it saves us a lot of time to
get the data and it also amplifies our
impact by making things automatic
the second thing is about culture the
reason that facebook
has a very strong data culture i believe
is because facebook has a very
strong age culture so engineering
culture if you look at uh
if you look at big tech companies at
silicon valley facebook and
google are probably the two most
engineer driven companies
but google moves kind of slowly and
facebook moves much faster
so at facebook a lot of products are
very scrappy
and they start almost like a hackathon
if
an engineer has a idea and wants to try
he or she already has the freedom to do
so
and uh they can just start the
experiment without
many permissions without jumping through
many loops
and then we use data to measure the
performance of these experiments and
decide which one is better for the users
so it's really this end-driven culture
that makes data powerful because this is
kind of a meritocracy right
and you have to have a standard to uh
to say which one autofill 100 projects
is more promising
and better for the user then you use
data to tell this also ties back to the
infrastructure at facebook ap testing
infrastructure is very advanced and very
efficient
actually there is a former facebook vp
vj he started uh he started a company
called static
to try to make facebook a b testing
infrastructure available to the public
uh i'm actually going to interview him
and his company
you know in the following weeks so stay
tuned if you want to know more about
this
but to sum up my previous point facebook
has a very strong
data culture because it has a very
strong edge culture
and to be able to make this meritocracy
work
be able to hold people comfortable we
really need to find ways to make data
speak
and the data scientists be the objective
voice
so that helps us to build a very strong
data culture
and because data scientists has a lot of
time to think about products
we also have the ability to do a
scientific understanding
and scientific analysis we'll also deal
with data all the time
so we probably know which part of the
product is more impactful
and easier to drive so naturally data
scientists contribute to strategies
most products starts with data
which means we do some analysis pointing
out a potential opportunity
or something that is uh a gap and we
need to fix
and then we start working on those and
ends with data
uh we use all sorts of measurements
including a b testing
and including other like such as
retention analysis
to say okay this is when we reached our
potential opportunity and this
is uh a product that this work is
actually helping our users
so the culture or data scientist has a
lot of say
and a lot of influence in products
then the third thing is the org
structure organizational structure
at facebook the organizational structure
is this
centralized but embedded structure
centralized means
data scientists report to data
scientists until very high
so for example at amazon engineers pms
designers
data scientists they'll report they'll
roll up their reporting chain
to a senior manager or to a director so
they report to the same director
but at facebook uh your data scientists
reported data scientist manager who
report to data scientist
director and data scientist director
data science
data scientist vp uh my reporting chain
uh doesn't merge with an engineer
reporting chain
and tell fiji who is the boss of
facebook app
and engineers also have their own
reporting chain pms have their
own reporting chain designers have their
own reporting chain so this is the
centralized part of it
this allows us to remain independent
right our credibility rather comes from
us being neutral
like if i report to a senior manager
who is also the single third owner of
this product
he or she may not want me to say certain
things if their career
is tied to the success of failure of
this product
i cannot speak ill of this product but
if my reporting chain is independent
i'm free to say whatever i i feel like
is objectively true
and also best for the org overall
instead of
just best for the for the career of this
particular pm
or this particular engineer manager and
when we have disagreement
i'm afraid to escalate to my boss and
then like
the bosses can have their discussions so
i don't have to be
strong-handed by my cross-functional
partners
i have my own people to back me up
meanwhile i'm
also embedded embedded means like all
the
cross-functional team right uh like uh
get together to work on one product so
for example i'm working on the seller
team
at marketplace and we have engineers who
work on this
we have pms work on this i work
solely on this product on this area of
the product
so i develop my domain expertise
of this product and i think about this
product all the time
so it gives me this product sense and
businesses
so my analysis is just not just pure
analysis
based on math it's based on my
understanding of the product and
my desire to solve problems for this
product
so that's the arc structure the
centralized plus embedding
allows allows us to remain independent
but
also be invested in the products and
have
impact on the product all right so those
are the three reasons i summarize of
why i think data is particularly
effective at facebook
to balance my opinion i also try to
think about
why facebook is not a good place for
data centers to work at
and i found two reasons why
especially for junior data scientists
because this
centralized and embedded model right you
are
the data owner of this product and
uh your manager cannot help you as much
as
other companies in other companies it's
usually like data
is a consultant role right and your
manager gets some big projects and
he or she can handle you a small piece
of it and uh
your manager would know how to how to
how to do this
but at facebook uh data centers have a
lot of
autonomy to do things that they believe
are best for the product but
because of that it also lacks directions
and guidance
you need to figure out what you need to
do that may not be very productive for
junior data scientists
the second thing is facebook have been
able to attract a lot of
great data scientists a lot of smart
talents
it works very well when the product is
growing very fast
but in certain areas of the product it
is might be mature
or then you will have more people than
the projects you have
and then you will have very smart people
compete with you for those high impact
projects
uh that can create stress it's not my
personal experience because the product
i'm working on is the hyper growth stage
we always have more things to do but i
certainly
have heard about this from other people
who work on other products
all right end of the video this is a
recruiting call
if you are interested in facebook.org
i'm working in is marketplace
like if you use facebook you probably
know marketplace or use marketplace
if not uh it is a great place for you to
sell
your used stuff and uh it is becoming
a great place for small businesses to
start their uh commerce activities
because it's very easy to set up like
amazon we don't charge a lot of seller
fees
and there are a lot of buyers are coming
to marketplace every day
trying to find products to purchase so
it's a very promising area
and because i'm going back to china we
need to hire a backfield
for my position and the org has a lot of
open headcounts
so i'm going to post the linking of my
manager
you feel free to reach out to her and
send a resume
she'll be more than happy to refer you
to a position all right
this is end of the video hope you
enjoyed see you next time
bye