In Part 1, we defined and looked at how automated document review has entered the eDiscovery market. Attenex and Stratify both encountered the same slow adoption and educational sales cycles when they brought concept clustering analytics to the hosted review market. Being on or over the cutting edge can be rough when you have a relatively conservative customer base. Counsel want strategic advantages without corresponding risks while corporations push for cost containment. In the midst of this pressure cooker, DiscoverReady has launched a new automated first pass review system called i-Decision™.
Equivio’s Relevance system showed us how training samples could be used to extrapolate relevance criteria and to provide cost/proportion estimations to support Meet-and-Confer negotiations. DiscoverReady pushes the envelope with the expectation that parties will conduct their first pass review within the i-Decision™ system and produce some portion of items that have not been individually reviewed. Before you reject the entire concept, remember that we have been using search criteria to include or exclude data prior to review for decades. The difference is the assumption that counsel does not have to check every item before it is produced, which also happens to be the largest cost driver in the discovery process.
So the real question is ‘How does it work?’ As one of the most innovative document review providers, DiscoverReady has extensive experience in using multiple non-linear review platforms with per document pricing. I was impressed with CEO James Dolan’s technology agnostic service vision as expressed at Duff & Phelps Legal Technology CEO Summit in New York. All of this background had me expecting a service offering that managed processing and review prioritization to optimize similar or related documents for a higher review rate. Basically an up-front investment of expertise and structured project management.
Instead, i-Decision™ is a hosted software platform that takes the culled and processed collection and uses an un-named technology to group 70% or more of the documents into ‘stacks’ based on multiple points of similarity. The system selects a number of ‘Prime’ exemplar documents from each stack for review. As long as the relevance designations within the prime documents are consistent, the system applies those decisions down the stack. This methodology differs from the blind, random training set methods used by other applications. They assume a minimum of 10-20% of the collection cannot be stacked and must be fully reviewed. The prime documents are usually 10-15% of the stacked collection, which means that the ‘Prime Review™’ still requires eyes on about 30% of the original collection. What is interesting about this approach is that the ‘stacks’ are not dependent upon or related to the case. This reminds me of the Attenex DocMapper clusters with a process to consistently review only the top items.
The customer can perform the ‘Prime Review™’ or outsource some portion to DiscoverReady’s reviewer team. Because the Prime documents are all relatively unique, the actual review rate on the 30% can be slower than a full pass linear or clustered review which includes a lot of junk and items that can be bulk marked. As expected, there is an emphasis on quality control sampling checks throughout the system. DiscoverReady engaged a professor to do a statistical validation model comparing the accuracy and error rates of i-Decision™ reviews to previously reviewed collections. Just hearing providers talk about third party validation makes me happy. I would be happier if I could crawl through the technology behind the ‘stacks’, but that information is only available to actual clients under a Non-Disclosure Agreement.
In answer to my original question, ‘Is the market ready for automated review?’ We have to acknowledge that automated document review is already part of the market and is being used on live cases. Whether parties are disclosing this to opposing counsel or the courts is a better question. Our young industry suffers from a lack of guidelines and rulings. The pace of technology will always outstrip legal standards, so we should just get used to moderating our risks by applying basic principles, good faith efforts and documenting the decision process to demonstrate the reasonableness of our process. Automated review deserves careful consideration and may be the right answer for some matters. Only time and outspoken judges will tell the full tale.