Predictive Coding Passes the Grade

Last April, the defendants in Global Aerospace Inc. et al., v. Landow Aviation, L.P. dba Dulles Jet Center, et al. asked Judge James Chamblin, of the 20th Judicial Circuit Court of Virginia, for permission to use predictive coding to cull approximately 8 terabytes of data prior to review. Although the plaintiffs protested the validity of this technology, Judge Chamblin allowed it to be used on the conditions that (a) the production be completed within 120 days, and (b) the plaintiffs would be given the opportunity to challenge the results once they receive the records.

The predictive coding process reduced the 1.3 million records down to a manageable 173,000, which were reviewed and produced to the plaintiffs in October. At a hearing in December, the plaintiffs did not raise any objections to the production, paving the way for the court to okay the document set.

​The process quantified the accuracy of both the culling and review of the records. The techniques used, combined with the predictive coding software’s built-in checks and balances, showed that 80% of the relevant records had been found. Recent studies, including the well know TREC initiative, have shown that manual review of large document collections generally identifies about 65% to 70% of the relevant records. This may have explained why the plaintiffs did not dispute the results. Clearly, combining manual review with technology incorporating appropriate checks is much more cost effective and just as accurate, if not more so, than using a manual process alone.