Practice Area
briefings
sponsored content graciously presented by McDermott Discovery
1 Predictive Coding at McDermott Discovery:
Real Experience, Real Results
By Alison Silverstein & Martha Louks
3 Why You Need Experienced Local Lawyers
for Your Managed Review in China
By Geoffrey Vance, Yodi Hailemariam
& Jared T. Nelson
Predictive Coding at McDermott Discovery:
Real Experience, Real Results
By Alison Silverstein & Martha Louks
There has been a lot of discussion
among e-discovery professionals
about how Technology Assisted Review (TAR) can be used to improve
the efficiency and reduce the cost of
document review. Typically, TAR is
thought of exclusively as “predictive
coding”. In reality, predictive coding
is just one way technology can assist
a document review project and make
that project more cost-effective and
efficient. Though each matter will require its own unique approach, there
are a few McDermott Discovery
techniques that are proven to reduce
the burden and risks while increasing the rate of review and accuracy.
Predictive Coding
In a typical McDermott Discov-
ery predictive coding workflow, a
relatively small number of “seed”
or sample documents are evaluated
for responsiveness by a core team
of attorney reviewers. The system
uses these decisions to code all
documents in the data set. Ad-
ditional seed documents are then
reviewed by the core team, and the
system is able to refine its under-
standing of the rules of responsive-
ness and relevance. A control set is
used to generate benchmark statis-
tics, including recall and precision,
which allow the users assess the
accuracy of the exercise. When an
acceptable percentage of respon-
sive documents have been found
by the system (recall) and the
accuracy of the coding is sufficient
(precision), the training process is
concluded. The threshold for recall
and precision varies from matter to
matter and is often determined by
the legal team or through agree-
ment with the opposing party or
government agency.
At the conclusion of the training
process, the system will attempt
to categorize each document as
responsive/relevant or non-respon-
sive/non-relevant. Depending on
the technology being used, some
documents will remain uncatego-
rized. At this point, the legal team
is able to decide whether to review
the remaining uncategorized docu-
ments. Often, the legal team will
also choose to review the responsive
set prior to production, as a final
means to ensure that irrelevant
information is being produced and
sometimes for no other reason than
to know what will be produced
before it goes out the door.
McDermott Discovery has suc-
cessfully used predictive coding
workflows in several cases to elimi-
nate large swaths of documents
from linear review. Making a small
up-front investment in the review
of seed documents yields enormous
cost savings because it segregates
and eliminates the need for attor-
ney review of the non-responsive
data. The number of seed docu-
ments reviewed usually consists
of a very small percentage of the