The
legal
technology
company
Altorney
today
announced
the
general
availability
of
MARC,
a
generative
AI-powered
document
review
system
designed
to
automate
first-pass
review
decisions
before
documents
enter
traditional
review
platforms.
After
first
announcing
MARC
last
March
and
going
through
a
pilot
period
with
corporate
legal
departments,
the
company
is
now
releasing
the
product
for
general
availability
to
corporate
legal
teams,
litigation
service
providers
and
law
firms.
The
Problem
MARC
Addresses
The
product
tackles
a
core
inefficiency
in
e-discovery
workflows:
organizations
typically
load
entire
document
sets
into
expensive
review
platforms,
only
to
cull
large
portions
as
non-responsive.
Shimmy
Messing,
Altorney’s
CEO
and
co-founder,
says
this
approach
creates
unnecessary
costs
and
security
risks.
“If
you’re
loading
your
million
documents
into
a
review
platform,
as
an
example,
and
then
immediately
culling
out
800,000
of
them
for
not
hitting
keywords
or
not
being
part
of
TAR
or
whatever,
you
still
have
these
800,000
documents
sitting
there
in
your
database
that
you’re
paying
for
and
that
are
exposed
from
a
risk
factor
after
leaving
your
corporate
environment,”
Messing
said
during
a
demonstration
of
the
product
for
LawSites.
MARC’s
approach
is
to
automate
the
culling
and
initial
review
decisions
before
documents
reach
the
review
platform,
ideally
within
the
organization’s
own
environment.
This
means
only
relevant
documents
–
already
tagged
with
first-pass
decisions
on
issues
like
privilege,
confidentiality
and
responsiveness
–
are
loaded
into
expensive
hosting
platforms.
How
MARC
Works
MARC
operates
as
a
text
analytics
tool
that
sits
between
data
collection
and
the
review
platform.
The
system
is
agnostic
about
which
large
language
model
(LLM)
it
uses.
Organizations
can
deploy
MARC
with
Altorney’s
provided
Llama
model
installed
locally,
or
integrate
it
with
their
preferred
approved
models,
including
those
from
Azure
or
OpenAI.
MARC
can
operate
entirely
within
an
organization’s
firewall,
with
no
data
transmitted
externally.
“All
the
data
can
stay
there,”
Messing
said.
“Nothing
has
to
go
out
to
OpenAI
or
Azure
AI
–
it
can
all
be
contained
in
a
local
environment.”
This
approach
provides
security
while
also
reducing
costs,
as
local
LLMs
avoid
the
per-token
charges
associated
with
cloud-based
AI
services.
Rachi
Messing,
Altorney’s
co-founder,
said
that
installation
typically
requires
just
30-40
minutes
of
IT
time,
after
which
the
system
is
largely
self-managing.
Protocol
Analysis,
Not
Prompt
Engineering
Among
MARC’s
distinguishing
features
is
its
deliberate
avoidance
of
requiring
prompt
engineering
by
users.
Rather
than
requiring
users
to
craft
precise
prompts
–
a
skill
Rachi
Messing
described
as
“really
hard
to
master”
and
prone
to
inconsistency
–
MARC
uses
what
it
calls
a
“protocol
analysis”
approach.
Creating
the
MARC
relevancy
protocol
from
the
background
materials.
With
this
approach,
users
upload
background
materials
about
a
case
into
a
folder.
These
materials
might
include
complaints,
subpoenas,
counterclaims,
pleadings,
or
even
informal
documents
like
an
email
from
in-house
counsel
outlining
a
new
matter
or
an
HR
complaint
in
an
internal
investigation.
MARC
then
generates
a
comprehensive
protocol
document
in
Microsoft
Word
format.
This
protocol
includes:
-
Identification
of
all
parties
involved.
-
Relevant
date
ranges.
-
An
overview
of
the
matter.
-
Key
individuals
and
their
roles.
-
Relevant
technologies
and
products.
-
Different
themes
of
the
case.
-
Specific
issues
to
identify
within
the
dataset.
Attorneys
can
then
edit
this
Word
document
directly,
adding
missing
individuals,
removing
irrelevant
parties,
narrowing
overly
broad
themes,
or
adjusting
other
parameters.
Example
of
the
protocol
created
by
MARC,
which
the
attorney
can
edit
and
resubmit.
The
edited
protocol
is
uploaded
back
into
MARC,
which
then
uses
it
as
the
foundation
for
all
subsequent
analysis.
This
approach
keeps
the
workflow
in
familiar
territory
for
legal
professionals,
Rachi
Messing
said.
“There’s
no
reason
we
need
attorneys
to
become
prompt
engineers,
but
they
love
editing
Word
docs.”
Processing
and
Validation
Once
the
protocol
is
finalized,
MARC
can
ingest
data
from
multiple
sources:
text
files
on
a
file
system,
Microsoft
Purview
exports
from
M365,
or
directly
from
Relativity
databases.
The
system
includes
an
integration
that
allows
users
to
point
MARC
at
specific
saved
searches
within
Relativity
without
actually
moving
the
data.
Integration
with
Relativity
to
analyze
docs
based
on
saved
search.
MARC’s
results
can
be
verified
through
a
sampling
and
validation
workflow.
The
system
automatically
determines
the
statistically
valid
sample
size
needed,
analyzes
those
documents
according
to
the
protocol,
and
tags
them
as
relevant
or
not
relevant
at
a
low
per-document
cost.
Statistical
sampling
and
validation.
These
sampled
documents
can
be
pushed
to
Relativity
or
exported
via
load
file
for
attorney
review.
Once
attorneys
validate
the
sample,
their
decisions
are
compared
against
MARC’s
determinations.
If
discrepancies
exist,
the
system
can
regenerate
the
protocol,
analyzing
what
needs
to
change
to
correctly
classify
the
disputed
documents
without
affecting
already
correct
decisions.
Viewing
a
relevant
result
in
Relativity.
This
iterative
process
continues
until
the
legal
team
is
satisfied
with
MARC’s
performance.
Then
the
full
dataset
is
processed,
at
a
rate
of
over
one
million
documents
per
24
hours.
Deep
Analysis
Capabilities
Beyond
simple
relevance
determinations,
MARC
can
perform
multiple
types
of
analysis
in
a
single
pass,
all
included
in
a
single
additional
cost.
These
analyses
include:
Privilege
Review:
MARC
analyzes
documents
for
attorney-client
privilege
and
work
product
protection,
providing
reasoning
for
each
determination,
identifying
parties
involved,
noting
whether
privilege
was
potentially
waived
by
third-party
involvement,
assigning
confidence
levels,
and
automatically
generating
privilege
descriptions
suitable
for
privilege
logs.
PII
analysis.
PII
and
PHI
Detection:
The
system
identifies
personally
identifiable
information
and
protected
health
information
with
granular
control
over
what
types
to
flag.
Users
can
specify,
for
example,
that
they
only
want
to
identify
financial
information
and
health
information
while
ignoring
personal
email
addresses
or
phone
numbers.
MARC
performs
entity
analysis,
associating
information
across
a
document
even
when,
for
instance,
a
person’s
name
appears
on
page
two
and
their
Social
Security
number
on
page
seven.
Issue
Coding:
The
system
can
tag
documents
according
to
case-specific
issues
defined
in
the
protocol.
Confidentiality
Analysis:
MARC
evaluates
documents
for
confidentiality
designations,
including
trade
secrets
and
other
sensitive
business
information.
Hot
Document
Identification:
The
system
can
flag
potentially
significant
documents
requiring
priority
review.
Foreign
Language
Processing:
MARC
automatically
translates
and
summarizes
documents
in
foreign
languages,
allowing
English-language
protocols
to
analyze
non-English
documents
and
providing
summaries
in
English
for
reviewers.
Output
and
Transparency
For
every
document
it
processes,
MARC
provides
not
just
a
decision
but
also
its
reasoning.
In
the
demonstration,
one
example
showed
MARC
tagging
a
document
as
not
relevant.
Its
explanation
detailed
that,
although
the
document
mentioned
UV
protection
technology,
which
could
potentially
make
it
relevant,
it
concerned
exterior
paint
rather
than
interior
window
coatings,
making
it
irrelevant
to
the
specific
case.
This
transparency
serves
multiple
purposes.
It
allows
legal
teams
to
understand
and
validate
the
AI’s
decision-making
process,
provides
documentation
for
defensibility,
and
helps
identify
where
the
system
might
need
refinement
through
protocol
adjustments.
Export
using
Relativity
integration.
Documents
are
also
enriched
with
summaries
and,
for
relevant
documents,
snippets
highlighting
the
most
pertinent
portions.
All
this
information
can
be
exported
or
integrated
directly
back
into
Relativity.
Cost
Savings
and
Predictability
Altorney
says
that
in
the
pilot
program
testing
of
MARC,
users
saw
significant
efficiency
gains.
The
company
highlighted
one
Fortune
500
company
case
involving
more
than
200,000
documents
where
MARC
achieved
62%
review
cost
savings
and
78%
hosting
cost
savings.
The
company
claims
an
80%
reduction
in
documents
transferred
to
hosted
review
platforms
and
an
86%
reduction
in
cycle
time
compared
to
traditional
review.
Its
costs
are
also
predictable
with
a
high
degree
of
precision,
the
company
says.
In
one
proof-of-concept
with
30,000
documents,
Altorney
provided
the
customer
with
a
budget
estimate
of
$2,500.
The
actual
cost
came
in
at
$2,506
–
a
level
of
budget
predictability
the
customer’s
AI
team
said
they
had
never
before
had
with
an
AI-based
product.
Viewing
a
privilege
result
in
Relativity.
Rachi
Messing
emphasized
that
beyond
cost
savings,
the
technology
addresses
human
inconsistency
in
review.
“You
give
the
same
document
to
four
different
attorneys
and
you’ll
come
out
with
four
different
decisions.”
In
tests
comparing
MARC’s
decisions
to
completed
human
reviews,
customers
found
that
discrepancies
often
revealed
human
reviewers
had
been
either
over
broad
or
over
narrow,
allowing
them
to
tune
MARC
to
find
what
they
actually
needed.
An
Expanding
Market
When
Altorney
initially
launched
MARC
in
March,
it
focused
exclusively
on
corporate
legal
departments
for
behind-the-firewall
deployment.
The
reasoning
for
that
limited
focus
was
both
technical
and
strategic.
The
company
believed
that
culling
should
happen
within
the
corporate
environment
before
data
leaves
for
external
review
platforms,
reducing
both
costs
and
security
risks.
However,
the
market
quickly
pushed
the
company
to
expand
its
approach.
Some
corporate
customers
expressed
strong
interest
in
using
the
product
but
indicated
that
internal
security
and
IT
approval
processes
could
take
up
to
two
years.
These
customers
asked
to
host
MARC
at
their
preferred
litigation
service
providers,
which
would
enable
them
to
accelerate
deployment
while
still
achieving
cost
savings
from
reduced
data
volumes.
Once
the
LSPs
were
on
board
and
began
using
the
product,
they
wanted
to
also
be
able
to
use
it
with
their
law
firm
customers.
That
led
Altorney
to
open
the
platform
to
law
firms.
“We’ve
now
opened
it
up
and
a
lot
of
LSPs
and
law
firms
are
hopping
on
board
and
have
it
installed
in
their
environments
as
well,”
Shimmy
Messing
said.
Pricing
Model
MARC
uses
volume-based
pricing
with
two
tiers.
The
initial
relevance
determination
costs
just
pennies
per
document
or
less.
Additional
analysis
–
including
privilege,
confidentiality,
issue
coding,
PII,
PHI
and
other
determinations
–
is
also
priced
at
a
single
per-document
rate
of
just
a
few
cents,
depending
on
volume.
Notably,
organizations
can
rerun
analyses
without
additional
charges
if
requirements
change,
such
as
modifications
to
a
confidentiality
order.
Humans
in
the
Loop
Despite
the
automation,
Altorney
emphasizes
that
MARC
is
designed
to
keep
humans
involved
in
the
review
process.
“GenAI
doesn’t
eliminate
the
need
for
human
oversight
–
but
it
enables
the
right
human
to
be
in
the
right
place
at
the
right
time
to
optimize
their
value,”
said
Stephen
Goldstein,
the
company’s
chief
product
officer.
Rather
than
replacing
human
reviewers
entirely,
Altorney’s
vision
for
MARC
is
to
transform
first-pass
review
into
quality
control
review,
allowing
reviewers
to
then
work
two
to
three
times
faster
on
a
smaller
set
of
more
important
documents.
Shimmy
Messing
acknowledged
that
while
some
users
might
eventually
feel
comfortable
producing
documents
straight
from
MARC
without
human
review,
most
currently
prefer
having
“eyes
on
everything,”
using
MARC’s
determinations
to
accelerate
rather
than
replace
human
judgment.
‘The
Ultimate
Truth
Seeker’
Altorney
was
founded
by
brothers
Shimmy
and
Rachi
Messing
in
late
2021.
The
company
initially
focused
on
its
Altorney
platform,
a
marketplace
for
document
reviewers
and
legal
talent,
which
launched
at
Legalweek
in
2022.
MARC
emerged
from
a
collaboration
with
Goldstein,
now
the
chief
product
officer
and
formerly
global
director
of
practice
support
at
Squire
Patton
Boggs.
Last
year,
he
approached
the
Messings
with
work
he’d
been
doing
on
using
gen
AI
for
first-pass
review.
After
evaluating
his
technology,
they
decided
to
productize
it,
spending
the
latter
half
of
2024
and
early
2025
developing
MARC
into
a
commercial
product.
The
product
name
honors
the
founders’
late
father,
Marc
Messing,
an
attorney,
rabbi
and
educator
who
died
of
pancreatic
cancer
in
2021.
Shimmy
Messing
described
him
as
“the
ultimate
truth
seeker,”
making
the
name
appropriate
for
a
tool
designed
to
find
truth
in
document
sets.
Both
founders
have
extensive
backgrounds
in
the
e-discovery
industry,
having
both
started
their
careers
at
Merrill
Corporation
in
the
early
2000s.
With
MARC
now
generally
available,
Shimmy
Messing
told
me,
Altorney
positions
itself
as
a
“boutique
coding
shop”
creating
“elegant,
unconventional
legal
software”
that
addresses
persistent
pain
points
in
legal
work
–
first
with
legal
talent
sourcing
through
its
Altorney
platform,
and
now
with
AI-powered
document
review
through
MARC.
