By: Kevin L. Nichols
Earlier this year, I wrote an article on the Executive Counsel Institute in San Francisco and mentioned some key takeaways, including best practices and vendors that excelled in various phases of the EDRM. Surprisingly, I was contacted by one of the vendors referenced in the predictive coding section of the piece and was educated about its business model and counseled about its distinctions from its other predictive coding competitors. Logically, his arguments were compelling, however, I felt that I needed to spend a little more time researching this topic and share the results of this effort with the eDiscovery community. Thus, here is a capsulated philosophical glimpse of the technology assisted review landscape and how different technological approaches effect its purpose of producing responsive evidence during litigation that will most likely lead to a triable issues of fact at trial.
There are many schools of thought when it comes to philosophical approaches to predictive coding. For our purposes, most use it for culling massive amounts of data in order to conduct more manageable human document reviews. Using a “seed set” of documents that has been developed and tested by legal experts working on the case, review teams are able to train the computer to identify other relevant responsive documents based on this criteria. In theory, technology assisted review is not supposed to replace the need for conducting a second and third level review.
Predictive Coding has several criticisms and/or draw backs. It is often times considered ineffective as it relates to “low yield” documents. Moreover, there are circumstances when there is an “over training” problem with the machine such that it increases the risk of over-fitting nonresponsive documents into the responsive results. Lastly, since it is not inherently designed to alleviate the need for a second level review, it becomes only the first attempt to cull large document populations into a more finite human document review.
In contrast, technology assisted review uses complex queries to solve an information retrieval problem. Theoretically, technology assisted review can eliminate the need for the first and sometimes second level human document reviews. Essentially, lawyers, linguists, statisticians, and the like convene to create long (sometimes thousands) queries relating to relevant categories of issues in the case to run through the entire corpus of the document population. A human sample is taken of the results and tested against the machine for accuracy. After the final queries are run, review teams are said to be able to skip the first and sometimes second level of review with confidence.
Having over 14 years of experience in the industry and participated in countless manual document reviews containing hundreds of boxes, I am still warming up to using predictive coding and/or technology assisted review to review and produce documents. Nevertheless, there are proven studies that show the reliability of using technology to cull electronically stored information (ESI) so that review teams can conduct document reviews manageable.
Technology assisted review can go a step further, however, using it solely for first and second level reviews will depend on your own appetite and what your stomach can tolerate. Various companies and law firm have experienced success using both. The choice is yours.
Kevin L. Nichols is the Principal of KLN Consulting Group located in San Francisco, which specializes in
Litigation, Diversity and Business Development/Social Media consulting.
For more information, please visit http://www.klnconsultinggroup.com.