►
Description
Buenos Aires OSCG 2022 | Digitalización inteligente camino hacia la Banca 4.0
B
We
are
going
to
tell
you
a
little
about
the
experience
that
we
had
within
Banco
Galicia
and
in
conjunction
with
redhat
applying
or
what
has
to
do
with
artificial
intelligence
and
what
we
call
intelligent,
digitization,
digitization
4.0
here.
This
graph
illustrates
the
previous
situation.
Before
we
incorporated
these
technologies,
we
are
a
traditional
bank.
The
traditional
bank
had
an
operations
sector
that
operations
sector.
B
Received
bylaws,,
which
is
the
document
with
which
a
company
is
registered,
and
that
statute
has
several
entities,
and
each
of
these
entities
have
to
be
disaggregated
to
be
able
to
enter
the
system
to
pass
it
clean.
Imagine
the
social
capital
The
Partners
that
each
of
the
partners
can
do
is
a
very
sensitive
issue,
the
issue
of
what
each
one
can
do
of
the
partners.
There
are
partners
who
can
sign
as
a
company
request
loans
as
a
company
So.
B
The
legal
issue
is
critical
and
Artificial
Intelligence
has
to
work
well
for
the
bank
to
adopt
it
and
feel
that
it
can
trust
these
technologies,
but
the
problem
existed,
and
the
truth
is
that
we
need
to
speed
up
And
how
we
speed
up.
Basically,
we
begin
to
imagine
with
this
process
faster
and
faster,
and
the
truth
is
that,
since
the
statutes
are
images
and
are
properly
speaking
as
unstructured
data,
what
ends
up
happening
is
that
the
statutes
do
not
have
a
single
way
is
not
so
easy
to
structure
to
take
it
to
our
systems.
B
So
in
this
way,.
What
artificial
intelligence
is
going
to
do
is
read
that
semantic
content
and
it
is
going
to
try
to
take
each
of
these
entities
to
integrate
them
into
our
systems
And.
In
this
way,
In
order
to
speed
up
our
processes,.
We
are
not
doing
anything
different
because
it
was
done
with
the
people.
Each.
One
of
the
people
saw,
the
statute,
took
the
entities
and
added
them,.
B
There
are
machines
with
artificial
intelligence
that
can
simulate
that
human
intelligence,,
but
they
can
also
add
a
lot
of
speed,
well,
here
inside
the
solutions
we
are
going
to
be
emphasizing
the
part
of
the
benefits
that
benefits
brings
us.
On
the
one
hand,
the
benefit
of
efficiency,
because
efficiency,
because
we
make
much
fewer
mistakes
when
we
go
with
these
algorithms
on
the
other
side
of
time,
this
combination
makes
us
competitive,
With,
the
rest
of
the
banks
that
had
the
traditional
processes
prior
to
our
incorporation
of
Artificial
Intelligence,.
B
A
A
We
developed
the
architecture
of
a
natural
language
processing
platform
so
that
it
can
be
extensive,
starting
first
with
the
enough
area
or
for
the
automatic
registration
of
companies
and
then
being
able
to
start
incorporating
it
as
services
to
other
areas
of
the
bank.?
This
also
allows
making
some
architectural
decisions.
Now
I
am
going
to
comment
on
it
allowed
the
bank
to
start
taking
it
as
their
own
ability
to
start
to
beat
time,
in
the
sense
that
at
the
beginning,
the
Data
scientists
did
not
have
the
resources
within
the
bank.
A
This
allowed
them
to
start
giving
muscle
developing
the
project.
To
start.
One
of
the
peculiarities
of
this
project
is
that
it
came
from
a
few
attempts
and
that
it
was
important
to
be
able
to
work
with
the
business
team
How
to
be
able
to
work
with
the
infrastructure
area,
because
when
a
project
is
hit
the
describing
comes,
then
we
have
to
also
break
culturally
with
that,,
showing
that
it
was
feasible
that
technology
could
do
it.
And,
on
the
other,
hand,
create
team
dynamics
that
would
allow
this
later
in
the
future,,
as
is
the
case
today,.
A
The
bank
has
that
capacity
today,
they
develop
it,
continue
it.
So,
that's
one
of
the
points
that
make
these
implementations
successful.
Another
aspect
is
that
one
says:
ok
But:
how
did
they
begin?
As
Matías
said,?
There
are
documents,.
The
statutes
were
presented
in
person
before,,
the
branches
went
with
the
statute,.
They
scanned
it
to
the
statute.
And
from
that
it
happened.
The
area
where
they
began
to
say
Ok.
This
company
has
the
capacity
And.
A
These
attorneys
are
really
the
representatives,
so
they
begin
to
run
all
the
issues
of
the
orcas
boyfriend,
taking
techniques
to
detect
themselves
and
money
laundering,
that
is,
all
that
makes
them
able
to
say
Ok
My
bank
is
going
to
operate
with
this
company
or
not,.
What
happens
on
the
company
side
is
that
a
company
generally
starts
to
register
an
account,
chooses
a
bank,
another,
and
the
one
that
responds
to
it.
First
is
the
one
that
stays
with
afterwards.
So
the
response
time.
A
It
was
one
of
the
great
challenges
we
started
with
the
documents
that
we
transferred
to
PDF
the
scanned
images.
We
generated
a
document
sanitizer
to
be
able
to
detect.
If
the
document
was
Ok
or
not
What
does
it
mean
if
it
had
to
be
scanned
again
because
capable
that
it
had
a
lot
of
noise
so
extracting
it
could
generate
some
feedback
slowdown
and
problems
later
in
action.
The
images
were
detected,
we
said
Ok,
it
is
or
not
If.
A
A
They
were
using
an
ocr.
We
saw
that
what
is
called
semi-arid
text,,
which
is
what
that
came
out
after
the
detection
of
that
text,
did
not
have
a
good
quality
So.
Nothing
good
was
going
to
come
out
of
there.
So
we
started
with
what
was
the
adaptation
of
1cr
libre
from
those.
This
is
a
statute.
They
see
that
they
have
stamps.
They
began
to
apply
different
types
of
techniques.
To
remove
that
noise,
that
are
the
signature,
stamps
that
may
be
above,
different
contrast.
Contract
techniques
began
to
be
applied.
A
And
that
allows
us
to
begin
to
delimit
character
by
character
that
are
still
characters.
They
have
not
taken
the
meaning
of
words,
then
we
can
begin
to
detect
those
lines
and
see
which
characters
then
have
continuity
with
another
line
and
from
that
text
begins
to
make
sense.
After
doing
that,
extraction
of
what
we
mentioned
previously
line
by
line
of
what
is
called
semi-arid
text,
one
begins
to
apply
different
types
of
techniques,
to
be
able
to
complement
the
loss
of
characters
that
there
was
a
loss
of
8%
Then.
A
Complementary
techniques
were
applied
to
be
able
to
start
applying
each
of
the
algorithms
later.
Those
algorithms,
an
algorithm,
is
developed
to
detect
ID
cards,
an
algorithm
to
detect
The
Partners
And,
so
on
successively
with
each
of
those
entities
that
had
to
be
recognized
and
subsequently
impacted
in
the
bank's
systems,.
An
important
point
is
that
each
of
these
algorithms
was
amplified.
Then
it
was
made
available
in
a
container
And.
A
That
is
what
gives
it
the
elasticity
so
that
it
can
be
trained
later,
as
Mati
said,
so
that
you
can
train
hot
and
also
so
that
you
can
make
inferences
in
almost
real
time.
All
of
this
were
the
architectural
bases
and
the
good
decisions
that
allowed
them
to
work
with
any
type
of
this
technology,
with
the
more
well-
known,
open
ones
than
what
I
was
telling
you
about,
with
the
Open
reference
architecture
or
with
others.
B
Well,,
let's
see,
as
Victoria
said,,
it
was
very
important
for
us
to
know
what
was
inside
the
architecture,
because
in
the
future,
what
happens
is
that
some
component
may
become
outdated
and
what
you
have
to
do
is
replace
it,
and
you
replace
only
that
component
and
all
the
architecture
are
giving
as
they
were.
Giving
and
another
issue
that
is
also
very
important
for
us
to
have
in
the
Nou.
House
is
that,
for
example,
as
an
architecture.
We
make
the
first
case
of
use
available,
but
then
we
are
the
reference.
B
If
another
component
has
to
be
changed,
there
is
another
sector
that
is
evolving
this
technology
and
that
technology
evolves,
for
example,
with
balances
and
trades
And.
If
that
knowledge
were
in
a
black
box,,
it
would
not
be
possible
to
adopt
other
teams,
or
it
is
also
part
of
our
being
credible-
is
the
part
of
having
open
source
and
that
you
can
see
how
it
works
inside.
B
Something
that
we
were
discouraged
by
this
type
of
architecture
was
exactly
what
we
could
see
and
we
could
replace
and
we
were
happy
with
the
bases,
with
the
foundations
of
the
architecture
here
within
the
challenges.
This
It
is
more
like
advice
than
what
we
live
when
we
implement
this
architecture,.
As
I
said
at
the
beginning,,
it
seems
that
it
is
a
simple
process
that
only
affects
one
sector,,
which
is
the
operations
sector,,
which
has
all
the
operational
load.
But.
B
So
there
is
a
clear
survey
of
the
times
and
each
person
who
needs
to
participate
is
key
for
the
project
to
be
useful..
There
are
no
shortcuts.
If
you
get
a
person
who
is
key
capable
to
register
the
company,
the
circuit
cannot
be
tested
end
to
end
and
the
solution
is
not
complete.
The
solution
is
complete.
When
it
is
finished,
then
another
of
the
points
is
the
infrastructure.
We
started
doing
this
solution
I
bought
and
we
had
to
be
very
meticulous
with
the
types
of
equipment
we
need.
There
are
shortcuts,
yes,.
B
There
are
some
shortcuts
that
can
be
taken,
for
example,.
What
we
could
do
was
that
the
documents
arrive
on
the
platform,
they
are
processed
at
night
and
the
next
day
the
operations
person
who
checks
all
the
documents,.
All
he
had
to
do
was
validate
himself.
a
bad
field
from
a
good
field,
but
it's
not
that
it
didn't
have
the
cost
of
waiting
for
the
document
to
load
Obviously,
they
processed
all
at
night.
B
So
you
had
a
lag
of
24
hours,
but
we
can
solve
that
every
time
with
more
and
more
infrastructures,
as
Victoria
said,
we
they
had
to
believe
at
the
beginning
and
once
they
believed
us,
they
decided
to
bet
more
and
more,
and
then
the
coordination
of
teams
is
key.
Each
person
has
to
know
what
they
have
to
do
in
different
parts
of
the
project,
but
they
also
have
to
have
a
global
vision.
B
For
example,
I
want
to
give
feedback
instead
of
one
field,
I
want
to
give
it
to
all
the
fields.
At
the
same
time,
I
want
to
start
doing
this
I
want
start
doing
the
other
thing,
And
that
generates
more
logic
and
many
more
tests.
So
what
you
have
to
do
is
mark
a
definition
of
dan
of
being
good
to
see
what
we
want
to
do.
B
B
When
one
makes
that
first
interaction,
it
begins
to
having
a
real
product
that
is
used
and
the
feedback
from
the
operations
people
and
the
feedback
from
the
illegal
people
starts
coming
in
and
critical
improvements
that
are
sometimes
more
important
than
the
details
in
the
forms
start
to
fit
into
the
roadmap,
and
you
begin
to
do
what
is
important
and
not
what
is
desirable
So.
It
is
also
very
important
to
understand
that
it
is
a
finished
document.
B
For
what
is
image
to
text
later,
The
second
step
was
good
to
see.
Now
it
is
very
focused
on
statutes
how
it
works
with
other
documents.
Some
tweaks
had
to
be
done,
Yes
in
the
first
document
Yes
in
the
next
one.
It
was
almost
inherited
naturally,
after
another
challenge
that
we
are
doing
It
is
that
for
at
least
what
I
experience
as
an
architect
is
that
the
future
is
within
the
structured
data
Because.
The
structured
data
has
already
been
exploited
a
lot
and
there
is
a
lot
of
information
hidden
in
audio
and
images.
B
So
one
of
what
unites
us,
One
of
the
challenges
that
we
are
considering,
is
to
go
find
the
audio
from
the
call
center,
transcribe
it
into
text.
And
if
you
look
closely,,
it
is
quite
similar,.
It
is
to
exploit
text
and
give
analytics
to
that
text,,
but
the
source
data
is
quite
complicated..
When
one
goes
through
an
unstructured
data.
B
No
and
added
to
this
also
another
of
the
challenges
we
have
is
to
find
the
resources
and
that
those
resources
feel
comfortable,
because
what
usually
happens
is
that,
since
everything
is
new
after
new
after
new
after
new
the
turnover
is
very
much
and
and
the
Resources
that
are,
specialists
are
difficult
to
find
and
maintain.
So.
The
truth
is
that
having
state-of-the-art
technology
and
the
support
of
the
Had
network
was
also
key
to
retaining
the
loyalty
of
many
resources
that
found
within
the
bank,,
which
is
not
a
technology
company,.
A
challenge
for
them.