►
Description
AMA agenda:
https://docs.google.com/document/d/1t7PjAGjEqMQn0bFPUYEKJrwH6sb85VxKi5rd0kuzKUA/edit
A
Awesome
so
I
will
give
a
quick
intro.
Today
we
had
the
pleasure
to
get
g
k,
chom
a
former
co-worker
and
a
good
friend
of
mine,
who
is
an
expert
in
data
science
to
come,
join
us
and
give
us
share
us
some
learnings
and
lessons
about
data
science
g.
Do
you
want
to
maybe
just
introduce
yourself
and
we'll
do
a
quick
round
of
intro.
B
Sure
happy
to
do
a
quick
intro.
I've
been
leading
data,
science
teams,
building
data
science
teams,
past
10
years
or
so
in
public
and
private
companies,
both
in
u.s
and
in
china,
and
it's
quite
rewarding
seeing
many
of
the
data
science
teams
continue
to
flourish
and
serve
the
companies
and
recently
I
started
writing
a
book
about
how
to
lead
in
data
science
and
publishing.
It
was
manning,
publishing,
co
and
I'm
going
to
share
a
little
bit
about
that
and
refer
to
some
of
that
later
on.
B
B
A
Completely
compliment
each
other
apart:
let's
keep
that
minimal,
but
thank
you
for
that
and
I
I
just
my
phone
memory
is
jika
always
buy
me
books
to
read
at
a
course.
I'm
like.
I
cannot
finish
the
last
one
and
there
is
a
new
one
about
data
science
about
designs,
so
he
he.
He
is
a
he's
a
learner.
So
I
learned
a
lot
from
him
and
I
want
to
bring
him
to
our
team
so
that
we
can
all
learn
from
him.
C
Cool
thanks
yeah.
Thank
you
for
for
doing
this,
they're
really
really
interested
to
hear
everything
you
have
to
share,
so
I'm
mike
krampus
principal
product
manager
on
the
growth
team.
I've
been
here
about
a
little
over
a
year
and
a
half,
and
I'm
really
just
curious
to
hear
more
about
data
science
and
how
we
can
apply
it
here,
sam.
Why
don't
you
go
next.
E
I
think
it
was
referring
to
the
wrong
thing
yeah,
so
my
name
is
nicole.
I
am
a
product
analyst
on
the
product
analytics
team.
I
joined
it
lab
about
three
months
ago
and
I
actually
also
came
from
acorns,
but
I
think
our
paths
because
I
started
in
october
of
2019,
so
I
think
our
paths
like
just
briefly
maybe
briefly.
A
F
G
H
I
Sure
yeah,
so
I'm
on
the
meltano
team.
Now
I
just
moved
over,
I
was
on
the
data
team.
I've
been
at
gitlab
for
over
three
years
now,
as
the
basically
the
founding
member
of
the
data
team
kind
of
helps,
build
it
up
yeah
just
I
love
these
growth
and
learning
opportunities.
So
you
know
thanks
for
coming.
Thanks
for
sharing
and
yeah,
let's
go
over
to
alper.
J
A
K
L
M
A
B
B
B
All
right
so
had
a
quick
conversation
with
gila
earlier
and
we
talked
about
what
can
data
science
do
for
me
right
and
I
think
that's
a
question
that
many
asks
when
they're
working
with
data
science
and
data
science
teams
and
I'd
love
to
dive
a
little
deeper
into
that.
But
first
want
to
do
some
quick
assumption
checks
so
that
I'm
on
the
right
track,
we're
on
the
right
track
in
this
session.
B
Well,
we
can
first
talk
about
what
data
science
can
do,
so
there
are
general
things
that
data
science
can
do
for
growth
team
and
also
other
functions,
so
I've
mainly
been
seeing
some
powerful
questions
being
asked
about.
You
know
how
to
think
about
business
impact
in
many
of
the
efforts
in
data
science
for
growth
as
well
as
you
know,
who
are
the
users
and
who
are
the
purchasing
decision
makers?
They
are
that
those
questions
are
relatively
easy
for
consumer
buyers
since
they're
the
same
person
but
for
enterprise
buyers.
B
Okay,
so
from
this
perspective,
what
we're
seeing?
Actually
you
know
some
of
the
efforts
we
have
at
linkedin
also
matches
this
description,
and
what
we're
seeing
is
that
there's
a
lot
of
effort
that
goes
into
looking
at
product
features
and
how
they
produce
product
engagement
and
on
the
other
side,
there
is
deciding
what
features
are
paid
and
what
features
are
free
and
then
that
feeds
into
the
customer
acquisition
and
customer
retention
process
in
this
customer
acquisition
retention
process,
depending
on
the
sales
model,
whether
it's
by
month,
by
quarter
by
year
or
multi-year
agreements.
B
The
customer
acquisition
and
retention
looks
quite
different
depending
on
what
business
model
it
is.
But
fundamentally
what
we're
seeing
is
that
there
is
actually
many
analysis.
That's
done
on
how
to
create
value
for
the
customer,
as
well
as
how
those
value
are
captured
in
terms
of
revenue
and
just
like
how
we're
familiar
with
the
google
search
right.
The
relevant
result
is
the
value
created
and
the
ads.
The
relevant
ads
are
the
ways
that
google
is
capturing,
the
engagement
and
when
we're
looking
at
many
of
the
growth.
B
M
C
A
B
All
right
yeah,
we
were
just
talking
about
the
various
opportunities
from
data
science
perspective
at
different
stages
of
technology,
adoption
cycle
for
a
particular
industry
and
we're
just
discussing
estimating
where
gitlab
is
in
its
industry,
and
I
think
the
consensus
there
is
in
the
early
majority
stage
and
depending
on
which
particular
product
we're
talking
about
it
could
be
slightly
earlier
or
later
in
the
early
majority
stage.
Does
that
sound
reasonable
to
you.
B
So,
looking
at
some
of
the
questions,
thank
you
gila
for
putting
together
that
google
doc
that's
really
helpful
in
making
sure
we're
focusing
in
the
limited
amount
of
time
we
have.
He
has
a
question
like
how
can
data
science
empower
growth,
product
team
differently
from
data
analytics
right,
looking
at
which
particular
stage
there
is,
we
can
have
different
interpretations,
but
you
know
I'd
love
to
first
understand
a
little
bit
better
about
what
are
some
of
the
questions
behind
the
question
like
what
are
in
the
context
about
that
question.
A
Yeah,
I
think
I
I
I
start
my
career
as
an
analyst
right.
Analysts
can
do
a
lot
of
things
in
terms
of
analyzing
data,
finding
some
opportunity
or
insights
to
inform
hypothesis
experiment
idea.
We
can
analyze
results.
We
can
do
dashboarding
a
lot
of
that
and
I'm
curious
for
data
science,
especially
for
a
b2b
product
like
gitlab
how
we,
how
data
science
can
enable
the
team
differently
like
beyond
the
the
basics
of
data
analytics.
B
Yes
great,
so
that's
a
great
question
and
I
think
the
field
is
learning
emerging
or
evolving,
and
many
of
the
data
analysts
are
also
taking
on
machine
learning
capabilities,
so
the
boundary
is
definitely
getting
further.
At
the
same
time,
in
very
large
teams,
the
roles
are
getting
more
clearly
fined,
as
people
are
taking
on
different
roles
in
this
larger
space.
B
B
Okay?
So
from
the
more
perspective
you're
looking
at
the
training,
so
data
analysts
are
very
good
at
interpreting
existing
data
and
producing
business
insight,
but
place
where
traditional
data
analysts
have
require
a
lot
of
support
using
the
data
wrangling
piece.
So
many
data
sources
are
not
as
clean
as
those
that
are
in
business
production
tables
or
in
cleaned
data,
warehouses
and
or
was
clean
schema.
B
That
part
tends
to
be
different
also
for
scientists.
They
usually
have
more
cs
background
than
data
analysts.
This
is
usually
but
not
always.
We
now
have
a
lot
of
data
scientists
analysts
with
cs
background.
What's
that
cs
background
can
do
is
to
provide
more
precise
tracking
specifications
so
that
the
signals
can
be
captured
more
precisely
from
the
implementation
perspective
and
then
in
terms
of
operation
like
operationalized
iteratively.
A
B
N
N
B
B
The
app
frameworks,
but
don't
have
a
good
appreciation
of
the
entire
stack
and
what
it
takes
to
be
able
to
produce
those
so,
especially
in
early
stage
companies
where
the
infrastructure
is
hacked
together,
they
produce
a
lot
of
tech
debt
that
shows
up
as
early
prog
a
lack
of
progress.
After
some
early
progress.
N
I
love
that.
That's
been
my
experience.
Yes,
I
love
that
we,
you
know,
companies
think
they
need
a
data
scientist,
so
they
hire
one,
but
the
infrastructure
isn't
ready.
The
data's
not
ready.
The
data
scientist
spends
a
lot
of
their
early
time
as
data
engineering,
but
they
want
to
deliver
quick,
wins
right,
and
so
they
do
quick
wins,
but
then
they
accumulate
a
massive
amount
of
tech
that
you
have
to
uncover
before
you
can
get
your
next.
Even
more
important
wins.
B
Absolutely
yes,
I
think,
that's
also
where
the
different
roles
coming
right.
The
data
analysts
data
scientists,
the
data
engineers
they're
all
they
all
have
their
place
to
put
together
the
entire
infrastructure,
to
make
sure
that
there
is
the
output,
that's
relevant
for
business
and
at
certain
scale
and
certain
impact
in
the
industry.
B
There
is
also
the
piece
about
data
science
and
as
agile
development
process,
so
in
general
data
science
really
values
stability,
if
you
think
about
the
foundation
of
it,
you're
looking
at
past
experience
and
predicting
future
behavior.
If
your
past
experience
is
based
on
a
certain
set
of
data
like
how
data
is
collected,
how
data
is
processed
and
that
changed,
then
all
your
prediction
is
off.
B
B
A
lot
of
the
data
engineering
work
needs
to
be
there.
So
how
do
you
bridge
that?
Well,
people
have
found
ways
to
be
able
to
bridge
this
stability
versus
agility
gap,
namely
through
good
road
mapping
at
the
business
line
properties,
so
that
you
can
provide
middleware
such
that
many
of
the
functions
can
be
provided
by
adding
filters,
adding
parameters
to
the
middleware
capabilities
so
that,
while
the
predictions
and
the
data
infrastructure
is
stable,
you
can
still
provide
the
agility.
B
B
So
those
capabilities
are
appearing
and
also
you
have,
for
example,
amplitude
that
provides
different
kinds
of
analysis,
like
behavior
analysis
and
pipeline,
funnel
analysis
out
of
the
box
with
standard
metrics
that
you
can
collect
from
data
pipelines
as
well
as
other
capabilities
like
tableau.
So
all
those
capabilities
are
appearing.
What
many
people
are
getting
wrong?
Is
they
need
to
wring
on
the
wheel
and
implement
the
entire
stack
off
from
scratch?.
N
Yeah
and
not
realizing
the
advances
that
have
been
made
in
these
areas
right
because
you
know
my
experience,
mimics
yours
around
tools
like
amplitude
or
mixed
panel
that
are
now
incorporating
data
science.
What
used
to
be
really
the
realm
of
a
of
a
bespoke
data
science,
team,
they're,
incorporating
those
in
the
product
itself,
so
for
you
know,
for
a
company
to
try
to
try
to
catch
up
to
that
it
even
gets
more
and
more
difficult.
B
B
A
K
K
Wondering
with
you
know
your
wealth
of
experience
or
companies
that
you've
been
involved
with
you
know.
Has
there
been
one
that
has
kind
of
found
that
sweet
spot
between
avoiding
these
pitfalls?
That
you've
mentioned
about
you
that
had
a
good
understanding
of
what
data
science
can
do
and
incorporate
it?
As
a
you
know,
an
active
part
of
the
overall
product
strategy.
B
Yes
got
it,
yes,
so
looking
at
the
different
companies,
I
think
there
are
so
working
companies
that
has
a
more
traditional
business
model
like
the
lending
business
model
in
like
air
and
digital
in
china,
as
well
as
companies
with
more
modern
business
models.
Like
acorns
many
of
the
data
science
work
are
started
at
least
started
with
a
particular
function,
or
a
particular
business
line
having
a
business
challenge
that
needs
to
be
resolved
with
a
data-driven
approach.
B
But
if
you're
talking
about
one
company,
that's
best
integrated
data
science
into
their
product
strategy
like
right
from
the
beginning.
There's
one
company
that
very
found
off
and
talked
about
in
the
book
is
levango
and
I
think
lavongo
was
recently
acquired
for
18.
Some
billion
dollars
was
the
teledoc
and
it
has
this
strategy
behind
it.
B
That's
called
the
aiai,
so
I
talked
about
it
in
chapter
eight,
but
briefly
describe
it
here,
so
they
have
a
they're
accompanying
the
healthcare
space
looking
at
chronic
care
and
they
are
working
with
insurance
companies
and
employers
to
lower
the
cost
of
chronic
care.
So,
with
the
emergence
of
a
lot
of
health,
the
health,
personal
health
devices,
if
you
would,
there
is
a
lot
of
capabilities
to
capture
and
aggregate
data
from
people's
daily
blood
pressure,
measurements
or
daily
blood,
glucose
level
measurement
and
then
using
not
just
ai.
B
B
A
A
Those
are
the
basic
examples
I
can
think
of
as
data
science
like
how
to
utilize
data
science
in
gilap
are,
there
are
other
things
or
other
areas
you
can
think
of,
like
that's
being
used
by
other
sas
type
of
company
like
even
linkedin
right.
They
have
the
probably
free
user,
payday
user,
the
enterprise
user.
Like
what
kind
of
application
do
you
see?
B
Yes,
so
there's
definitely
a
lot
of
different
places
where
data
science
could
be
applied.
I
think
that
the
ones
you
talk
about
are
specific
ones
that
are
closely
related
to
sales
leads
and
sales
nurturing
process
right.
There
are
also
other
business
other
business
functions
that
could
make
use
of
this,
for
example,
for
features
right.
You
may
have
features
that
are
early
that
are
in
beta
that
could
be
offered
for
free
and
depending
on
people's
engagement,
you
can
decide
where
the
pay
will
start
or
ends.
B
So
there
are.
There
is
the
value
creation
process
where
you
are
launching
a
product
or
a
product
feature
and
refining.
That
product
feature
until
is
to
a
certain
engagement
level,
and
then
that
could
be
turned
into
the
premium
version
of
it
could
be
turned
into
a
paid
feature
to
make
sure
that
there
is
a
willingness
to
to
renew
at
the
end
of
whatever
contract
cycle.
B
So
that
is
from
the
product
side
and
then
from
the
marketing
side,
there
could
be
the
different
pipeline
stages
so
like
in
every
business
function.
There
could
be
a
similar
kind
of
metric,
like
the
ones
that
you're
talking
about
for
sales
and
potentially
customer
service
could
have
also
have
that
right.
You
could
be
looking
at
the
customer
service.
A
Interesting,
so
the
the
product
side
example
you
mentioned,
is
there
a
case
study
or
is
there
an
example
we
can
look
into
which,
like
which
companies
or
or
some
some
companies
are
doing,
that
using
data
science
to
evaluate
future
adoption
and
to
decide
where
to
pay
put
paywall
or
things
like
that
you
don't
have
to.
We
don't
have
to
get
that
today.
But
that's
just
sounds
fascinating.
I'll
have
to
look.
B
B
Like
what
would
be
your
ideal
case
so
or
ideal
scenario
for
value
created
for
the
customer,
so,
for
example,
at
linkedin,
there
is
the
customer
that
is
engaged
in
job
search
process.
There
is
a
customer
engaged
in
the
linkedin
learning
process
and
there
may
be
metrics
that
you
define
to
segment
the
user
in
terms
of
those
that
are
more
engaged
and
less
engaged
more
likely
to
be
able
to
turn
into
a
paid
user
and
less
likely
to
be
turned
into
a
paid
user.
What
are
the,
where
are
the
boundaries?
B
C
C
B
Yes,
so
for
the
cross
stage,
usually
we're
talking
about
having
ad
hoc
capabilities
put
together
so
there's
definitely
the
craw
walk
and
run
type
of
thing,
and
people
have
split
into
various
stages.
I
happen
to
have
split
into
five
stages,
but
there
are
some
implications
in
each
different
stage,
right
so
for
the
ad
hoc
area
really
you're
starting
out
you're.
B
B
So,
in
order
for
those
algorithms
to
grow
for
those
benefits
to
continue
to
perform,
you
need
to
do
a
lot
of
manual
things
to
maintain
and
keep
it
up,
and
that's
the
tech
debt
that
rob
was
also
mentioning
earlier,
and
then
there
is
the
functional
stage
where
you
have
identified
those
use
cases
and
launched
and
be
able
to
really
maintain
some
of
them
by
automating
some
of
the
process
and
then
later
on
for
the
integrated
and
governance
and
culture.
But
those
are
the
different
stages
later
on
to
focus
at
the
beginning.
C
B
B
B
Those
are
general
area
cases
where
you
know
strategically
that
you
want
to
go
down
a
particular
path
and
you're
working
on
putting
together
a
set
of
infrastructure
with
some
infrastructure
risks.
You
want
to
implement
the
piece
with
highest
infrastructure
risk,
first,
to
make
sure
that
the
whole
project
works
now
in
order
to
combat
these
emerging
problems
like
spammers
and
crypto
miners.
B
What
you
want
to
do
is
put
in
an
initial
solution.
First,
so
for
an
initial
solution,
it
should
be
a
well
defined
problem
with
a
set
of
proven
solution
that
can
get.
Eighty
percent
of
the
business
benefit
was
twenty
percent
of
the
work,
so
in
those
cases,
for
example,
we
encountered
this
with
a
fraud
financial
fraud
as
well.
B
The
priority
is
just
first
come
up
with
a
generative
model,
with
domain
expertise
about
what
are
some
of
the
basic
rules
that
we've
seen
that
could
filter
out
a
large
portion
of
these
kind
of
customers,
either
through
ip
through
their
behavior
and
like
patterns
in
the
username.
For
example.
You
know
the
same.
B
After
a
while
right,
I
would
see
so
you
can
collect
those
labels
and
then
reverse
engineer
or,
like
other,
in
other
words,
use
machine
learning
to
detect
what
patterns
those
people
exhibited
the
sign
up
stage
that
had
those
kind
of
bad
undesired
behavior
later
later
on.
So
those
look
at
the
phase
two
and
then
you
work
further
with
the
team
to
understand.
J
A
J
C
B
A
B
Two
reasons,
so
that's
the
style
that
making
books
choose,
and
actually
there
is
a
bit
of
interesting
tree
behind
many
of
these
drawings
drawings.
Are
you
know,
traditional
clothing
for
people
of
different
professions,
back
in
the
1800s
in
france
and
the
particular
one
that
we
chose
for
our
book
is
actually
an
artisan
someone
who
is
a
practitioner
I
think
for
making
wine
they're
actually
many
times.
These
artisans
are
not
just
men,
they're
women
as
well,
and.
D
B
Yeah,
that's
a
good
question
so
with
the
different
tools
and
proliferation
or
democratization
of
data
access.
There
is
definitely
that
challenge,
but
I
think
I
have
a
very
strong
opinion
that
there
should
be
a
single
source
of
truth,
mainly
because,
in
order
for
people
to
use
data,
they
need
to
trust
it.
And
if
there
is
two
different
version
of
the
truth,
then
people
need
to
start
understanding.
Why
they're
different
in
order
to
be
able
to
trust.
M
D
Well,
I
think,
there's
a
there's
a
balancing
act
between,
I
think
there's
in
this
my
personal
opinion,
I
think,
there's
like
a
balancing
act
of
what
data
should
be
self-service
in
the
sense
of
enabling
non-technical
users
to
have
quick
learnings
and
then
what
data
is
needed
for
in-depth
analysis
for
important
business
decisions,
and
I
don't
I
don't
know
where
how
you
marry
those
two
things.
I
think
it's
a
tough
thing
to
solve
for
so
that
that
was
kind
of
the
intense
for
the
question.
B
Got
it
so
there
are
ways
to
solve,
for
that.
One
is
to
understand
the
daily
lineage
of
these,
so
usually,
when
you
have
two
sources
of
truths:
they
branch
out
somewhere
in
the
data
production
process
and
yeah.
Right
now,
if
you
have
like
say
two
source
of
truth,
imagine
like
when
the
company
grows,
you
have
ten
thirds
of
truths.
Then
nobody
understand
what
to
look
at
right.
B
So
it's
really
important
to
be
aware
of
those
concerns
and
really
look
at
data
lineages
through
the
use
of
data
catalogs
and
then
metadata
management
to
make
sure
that
those
differences
are
actually
necessary
right.
What
you
can
do
is
to
prune
these
kind
of
divert
trees
by
consolidating
the
source
of
truth
to
one
branch
or
to
make
sure
that
those
branches
don't
evolve
uncontrollably,
because
it's
very
typical
for
very
large
companies
to
have
these
data
processing
capabilities
as
three
or
four
layers.
I
B
I'll
take
the
last
minute
to
really
quickly
talk
about
some
things
that
that
could
be
helpful
for
the
set
of
questions
so
for
the
particular
book
we're
really
talking
about.
You
know
how
to
look
at
people
at
different
look
at
data
scientists
at
different
stage
of
the
career,
and
I
think,
for
a
pillar
function
working
with
data
science
data
scientists.
It
would
be
great
to
understand
what
to
expect
from
each
of
the
levels
and
if
you
look
at
on
the
left,
we
actually
have
a
lot
of
different
things.
It's
almost
like
a
menu.
B
You
can
order
off
if
you
hire
a
database,
lead
or
manager
or
director
or
executive
right,
so
you
can
figure
out
what
you
can
be
requesting
a
data
scientist,
for
example.
It
would
be
pretty
harsh
to
ask
senior
data
scientists
to
put
together
an
entire
roadmap,
but
it's
very
reasonable
to
ask
the
director
to
do
so.
B
A
Thank
you
so
much.
I
think
it's
a
it's
a
very
eye-opening
like
we
all
learned
a
lot
and
in
areas
I
think
many
of
us
are
not
super
familiar
and,
as
I
mentioned,
jiko
is
a
is
a
great
friend
and
mentor.
If
you
are
interested
in
learning
more
about
data
science,
maybe
check
out
his
book,
I
I
might
just
buy
those
books
to
collect
those
book
covers
because
it's
just
very
unique
and
beautiful
cool.
Thank
you
so
much
bye.
Everyone.