At the Federation of Defense and Corporate Counsel (FDCC) annual meeting, I recently attended an excellent program on predictive coding presented by Howard Merten of Partridge, Snow & Hahn in Providence, RI, Matt Nelson of Symantec, and Magistrate Judge Andrew Peck of the Southern District of New York (who has written a leading article and one of the leading court decisions on this issue, see my March 9, 2012 post for more). For those who are not familiar with it, predictive coding is a relatively new software-based document review platform that uses computer algorithms to identify relevant/responsive documents. First, a review set of potentially-relevant documents is gathered by the producing party and loaded into the system. Second, a set of more senior-level attorneys (not junior associates or contract lawyers) reviews a sample set of documents for relevance, coding them for their degree of relevance. Then the software itself identifies relevant documents in the remainder of the review set. The human reviewers then review again to help the computer refine its analysis, through an iterative process.
Here are some key points I gleaned from the presentation:
- Predictive coding is clearly cheaper than traditional review. A case study cited by the presenters suggested that with 100,000 documents, traditional manual review might cost approximately $140,000, whereas with predictive coding the attorneys’ fees might be as low as $9,600, plus the electronic discovery vendor fees, and the cost of reviewing the documents identified by the computer as relevant to confirm relevance and that they are not privileged. Predictive coding has also been shown to be more accurate because the computer doesn’t get tired unlike young lawyers who are spending countless hours in front of a screen.
- The system will rank documents by degree of responsiveness, which is a key time-saver, helping the litigation team get to the most relevant documents faster. The secondary review process for a large production can be segmented so that the senior lawyers review the documents that the computer identifies as most relevant and junior lawyers or contract attorneys review the documents the computer identifies as least relevant.
- In the limited court decisions on this, where the producing party has asked the court for approval, courts have uniformly approved the use of predictive coding. Where the requesting party asks the court to force the producing party to use predictive coding there has been some resistance, but this still might be possible if you ask early, before the other side spends a bunch of money on manual review. (In class actions that is essentially irrelevant because it would be the very rare case where the plaintiffs would have a large number of documents anyway.)
- The key sticking point can be if the opposing party expresses a concern that the producing party will “game the system” by marking “hot” documents as non-responsive. There are various approaches to deal with this involving quality control checks and some transparency.
- Because predictive coding is word-based, it will not work for photographs, drawings, spreadsheets and the like, which need to be dealt with separately.
- Judge Peck reminded everyone about taking advantage of Fed. R. Evid. 502(d), which allows a court to issue an order regarding non-waiver of privilege for privileged documents that slip through the review process. He suggested asking for a Rule 502(d) order in cases involving large productions. In my experience, this is often also built into a confidentiality/protective order.