Technology Assisted Review (TAR) has dominated our recent briefing sessions with providers and consumers alike. Consumers want eDJ to clarify the terminology, technology and market hype surrounding recent cases. Providers have expressed their frustration with the portrayal of TAR as some kind of ‘Easy Button’ that will magically reduce your review expense by 95%. Really. We are hearing second hand stories like, “But VendorX says his system only needs to train with 5%.” Early TAR innovators like DiscoverReady’s CEO Jim Wagner long ago understood that, “It’s not the technology. It’s the people and process.”
That can be a complicated message for a relatively unsophisticated consumer who reads blogger headlines instead of the actual transcripts. Discovery and review cease to be easy or routine as the volume and composition of potential collections exceed the ability of a single reviewer to manually code every item. Beyond simple relevance the additional complexities of privilege in TAR keep coming up in our briefings.
The comingled business and legal roles of many corporate counsel and experts challenges the old assumption that including an attorney on communications automatically conveys privilege protection. This is old news and most of you are familiar with the process of 2nd and 3rd pass review to adjudicate these complicated issues. But how do TAR technologies and workflows deal with embedded privilege in email threads, attorney comments in spread sheets or secondary work product reports? Most clustering, propagated or predictive review systems are optimized to train for one ‘category’ at a time, that being relevance. Training for issues or privilege can be done with separate seed sets of known documents and iterative sample sets, but the stakes for missing a single privileged document buried in the final production set can be serious, just ask Google.
Planet Data’s CTO Michael Wade
laid out some of the challenges with TAR that they see with every project.
1. Custodial based preservation result in ever broader collections
2. Broad collections of unrelated ESI create very low recall and relevancy rates
3. Most TAR systems have a minimal level of relevant documents to be effectively trained (eDJ heard this from end users while researching TAR usage)
4. Clustering, categorization and other linguistic grouping technologies can be overwhelmed by non-relevant similarities in raw collections (eDJ – they can’t see the trees for the forest)
5. The comingling of business and legal roles in corporate communications mean that potential privilege contaminates a large portion of ‘decision’ documents (eDJ – the same ones that TAR is being trained for)
6. The threshold for error in relevancy review is usually much higher than that in privilege review
7. TAR is just one tool to leverage within a structured discovery process
Every TAR conversation I have with experts, providers and consumers reinforces the need for adaptive, integrated solutions that combine search, sampling, profiling and TAR for optimized, defensible results. This is why Planet Data’s ExegoV3 technology has prioritized upstream processing and feedback over developing a downstream review platform. There is definite consumer interest and excitement around all the new TAR offerings. It is important to keep them in perspective with the true needs of your specific case and to integrate them into your discovery lifecycle to maximize their benefits while minimizing risk and cost. Remember to think outside of the TAR ‘box’ for uses in shaping collection criteria, creating inclusion/exclusion processing filters and quality control throughout the process.
Have you found any advantages or challenges with privilege in TAR reviews? If so, we would like to hear about them. Please comment below or send me a note at Greg@eDJGroupInc.com.