Ediscovery: US judge approves predictive coding results
For the first time in the US, a court has approved the result of an ediscovery production in which predictive coding was both ordered and used. When it was demonstrated that the predictive coding’s accuracy in identifying responsive documents exceeded the court’s agreed requirement, opposing counsel raised no further objections. Therefore, the defendants’ 173,000 document production in the Global Aerospace case will stand.
The plaintiffs had originally objected to the use of any computer-assisted review (CAR) to cull the initial 1.3 million document set; the judge overruled that objection.
Thomas C. Gricks III, chair of the e-Discovery Practice Group at Philadelphia-based Schnader Harrison Segal and Lewis, was the attorney who persuaded Judge James Chamblin in the 20th Judicial Circuit of Virginia’s Loudoun Circuit Court first to allow use of predictive coding, and then that his client had done its duty in using the technology (which also saved the client hundreds of thousands of dollars in attorney fees).
Using the OrcaTec Document Decisioning Suite for predictive coding, an experienced Schnader litigator was able to review and code 5000 random documents from the 1.3 million document collection within a matter of days. There were no keywords or seed sets used, such as have been ordered in other cases like Da Silva Moore, because OrcaPredict does not require either, thus saving even more time and money.
Once the coding was completed to the attorney’s satisfaction, OrcaPredict trimmed the document set 83% in just a few hours. By significantly reducing the number of documents in this way, Schnader was able to efficiently conduct second-pass review with a team of only five attorneys over a much shorter time period than they would have needed using traditional review. Roughly 173,000 documents of the 1.3 million ultimately were produced in discovery.
“Using a strict manual review process, just culling the irrelevant documents would likely have taken as much as 20,000 man-hours to review, at an anticipated cost of $1.5 million,” said Gricks. “We were able to generate results that measurably exceed typical manual review, at a fraction of the cost, and we were prepared to produce the vast majority of the responsive documents within a tight timeframe.”
Karl Schieneman, president of Review Less and the predictive coding consultant Schnader used on Global Aerospace, believes using CAR is going to be a significant trend, especially in light of this decision. “With the ever-increasing amount of data that corporations are storing, pre-trial e-Discovery costs are becoming absolutely prohibitive without the use of technologies like OrcaTec’s predictive coding.”
To obtain the April order from the Virginia court that, for the first time, allowed a party to use predictive coding for ediscovery over another party’s objection, Gricks had proffered the testimony of predictive coding experts. Schieneman, OrcaTec’s CTO & Chief Scientist Herbert L. Roitblat and Timothy Opsitnick, president of JurInnov in Cleveland, helped Gricks convince the court that predictive coding would create better, defensible document culling, while providing significant time and cost savings. JurInnov collected and processed an initial 8TB of electronically stored information for Schnader before delivering the 27GB, or 1.3 million documents, to OrcaTec – an initial reduction of more than 99% before predictive coding even came into play.
Gricks also proposed, and the court agreed, that there should be a floor for the quality of the predictive coding in order for it to pass muster. Gricks’ client had to show that predictive coding had found at least 75% of all responsive documents in the document set (75% Recall). Human review, also called linear review, is normally 50% or less on Recall.
Upon post-predictive coding testing, OrcaPredict was found to have achieved 81% Recall plus 80% Precision. Precision means that 80% of the documents OrcaPredict predicted would be relevant were found upon review actually to be relevant.
“OrcaTec’s predictive coding has been attaining the Global Aerospace level of Precision and Recall – and even much better – on a very regular basis,” said Roitblat. “You can see on the OrcaTec dashboard the percentage of precision and recall you’re achieving as you code. The clients here were aiming for greater than 75%, which is vastly better than human review, and they got 80%. We can’t help but be proud of that.”
In October, Gricks produced the document set that his team had gathered and reviewed using predictive coding. After a hearing on December 7, the court okayed the document set without objection.
“While there are other cases allowing or ordering predictive coding or CAR, this is the first time a court has said, essentially, ‘The way you used predictive coding to cull your data set was fine,’ “ said Roitblat.
“The true value to U.S. companies of this ruling on predictive coding is in allowing litigants to move a case that really matters to trial, rather than being forced to settle to save the millions in review costs that would come without it,” Gricks said. “Corporate litigants can use this technology to truly understand their cases without waiting months or years until mountains of documents are manually reviewed.”