By Cindy Courtney, Vice President and General Counsel

Always considered a strongly substantive conference as opposed to a trade show, the Georgetown eDiscovery Conference has not disappointed this year, particularly as to technology assisted review (TAR), which received top billing at the conference.  Of 20 sessions, three were devoted solely to TAR—that doesn’t include sessions devoted to such topics as preparing for a meet and confer session, which had a strong TAR component.  (Speakers made clear that their remarks included and were directed to computer assisted review, predictive coding, computer aided review, or any other term in use today, irrespective of the acronym they used.)  The TAR track also attracted significant talent, including Maura Grossman, Ralph Losey, Herb Roitblat, Gordon Cormack, and a host of corporate and service provider heavy hitters and law firm members.

The three TAR sessions covered whether and when to use TAR, definitions and meaning of key statistical terms that are critical to TAR, and a role-play example of negotiating a TAR protocol in a trade secret theft case.  Here is a sample of what the speakers said:

1. Any case is theoretically a candidate for TAR, although lawyers should take into consideration not only the cost of the tool and review, but the allocation of resources that will be required to get the system up and running—in other words, can litigants afford to take a subject matter expert out of commission while the TAR system is being put in place?
2. Don’t put all your eggs in one basket—familiarize yourself with more than one TAR platform and get to know them, because they do different things that can be valuable in different cases.  Do they provide a proportionality score?  Do they automatically supply the statistical metrics that may be important when negotiating the protocol? 
3. Know something about your documents in advance.  Some cases may not be candidates for TAR because the important documents don’t lend themselves to a textual analysis approach.  For example, does the case focus on engineering drawings or graphics?  If so, consider either pulling these documents out of the corpus and dealing with them separately or, if they constitute a preponderance of documents, don’t use TAR.  Drawings and graphics often don’t contain text, and TAR systems run on text.
4. Understand the statistical principles underlying TAR platforms.  Precision, recall, random, richness, confidence level, confidence interval, margin of error—hearing these terms may make your head spin, but gaining a basic understanding of them will help you recognize a draft protocol written by someone who does NOT understand them—trust me.
5. Use an expert.  This is not inconsistent with number 4 above.  It’s your service provider’s job to understand how the platform works, but you should still make an effort to understand the statistics.  And, when you are choosing a platform, and you have a choice to use the same platform with more than one service provider, grill each service provider about the number of TAR cases they have supported and their experience.
6. TAR is not a substitute for a sound review rubric.   TAR may greatly reduce the number of documents that will ultimately require eyes-on review, but you still need to take care with the initial universe—the appropriate number of custodians, the right sources, not over-collecting.  Your identification and collection of ESI can still be challenged, even with use of TAR.
7. TAR is best used for culling and mass classification, not determination of issues or attorney client privilege.  You may have suspected that, but the experts—some of them, at least—have now stated it openly.
8. The cases now generally support use of TAR if a party wants to use it, but are inconclusive about whether a party will be required to use TAR.  If you want to use TAR, approach your adversary with a plan that includes transparency and cooperation—that is, opportunities for opposing counsel to participate.
9. The battle ground in litigation will be around what details of the TAR project must be disclosed—the size of the seed set, selection method of seed set, content of the seed set, the expert who is selected to review the training documents, the method for analysis of the training documents to determine whether they are viewed as relevant, sharing of the training documents as well as the QC documents, levels of precision and recall—all of these are in play as disclosures that a party should consider making.  Decide what is non-negotiable and what your fallback position is.
10. TAR is here to stay.  It may that TAR will never see the light of day for most cases, but for cases of a certain size, the science demonstrates that neither linear review nor use of search terms yields a strong showing of relevant documents.  The Georgetown presenters predicted that a time is not far off in which those methods are not accepted for certain cases, which leaves TAR as the only possible choice for culling and mass determination of potential relevance.

 Stay tuned for more from the last day of the Georgetown Advanced eDiscovery Institute……..

Tags: e-discovery, ediscovery, Predictive Coding, Technology Assisted Review