A talk I gave at the Data Science London meetup on April 23, about how we at Causata use a combination of intelligent data structure, machine learning, and regulated experimentation to understand cause-and-effect in customer behavior
2. It’s
essen7al
to
connect
data
across
all
the
channels,
into
a
full
view
of
the
customer
3. browser
cookie
online
browsing
delivery
address
credit
online
purchase
card
loyalty
card
store
purchase
store
purchase
store
purchase
Connec7ng
this
data
is
hard
–
there’s
ambiguity
7. Retailers
know
from
surveys
that
customers
oJen
research
online
and
buy
offline.
But
unless
they
connect
individuals’
ac7vity,
it’s
hard
to
aNribute
marke7ng
influence
8. Arrow
of
>me
Website
Call
Center
Website
Loyalty
Card
Website
Loyalty
Card
Major
Product
Session
Ques7on
Session
Promo
Email
Session
Sign
Up
Purchase
in
Store
16. reinforcement
learning
Causata
must
choose
the
ac7ons
which
will
yield
the
greatest
reward,
where
reward
can
be
any
func7on
we
wish
to
op7mize,
and
the
rewards
may
be
deferred
to
some
7me
in
the
future
ac>ons
rewards
learning
agent
environment
18. Visit
website
from
Research
Request
Speak
to
advisor
Sign
Ongoing
online
banking
credit
cards
applica7on
form
in
branch
agreement
rela7onship
Current
state
of
the
art
is
to
locally
op7mize
each
interac7on,
in
terms
of
immediate
next
step
With
all
the
data,
can
op7mize
over
true
long
term
business
goals
21. We’re
working
in
very
exci>ng
>mes
Connected
datasets…
handle
ambiguity
Intelligent
decisions
with
machine
learning
Principled
experimenta>on
At
massive
scale
In
real
>me