Unpublished opinions cannot be cited in any motion, but can be very illuminating on how state courts are handling ediscovery issues. In a case where plaintiffs sued a retailer for alleged violations of requesting and recording personal identification information during sales, the defendants were able to recover predictive coding costs as a prevailing party. Dremak v. Urban Outfitters, No. D071308, 2018 Cal. App. Unpub. LEXIS 1925 (Mar. 23, 2018).
The trial court had exercised its discretion to award the Defendant costs for processing ESI, coding analytics to identify relevant documents, and create a review database. The Court of Appeal explained these costs are neither expressly included nor excluded under Code of Civil Procedure section 1033.5(a). This enabled the Defendants to take a document database with over 400,000 documents and identify 1,658 for production. The trial court held that the database was “specially required in order to process the unique discovery in this litigation.” Dremak, at *23.
The Defendant explained in an affidavit the “time-consuming process” to make the production of 1,658 records. The CA Court of Appeal agreed that the costs were “reasonable and necessary to the litigation.” Dremak, at *24.
Bow Tie Thoughts
The context of this opinion makes the predictive coding sound like it required a seed set to train the predictive coding on relevance. Predictive coding models that run on continuous active learning (meaning the predictive coding learns from the reviewers) does not require the training of a seed set. This sound take less time then creating a seed set. Granted, this system does need quality assurance testing, where admins do check their prediction models to validate there are no gaps in review. This is to ensure no relevant information is missed in training the prediction models.
The technology aspects aside, it is very refreshing to see a trial court and Court of Appeal recognize the use of predictive coding was “reasonable and necessary to the litigation.” Also a big high five to the review team that was able to identify 1,658 records out of 400,000 with predictive coding. The big lesson here is to document work that is done to train predictive coding, so affidavits can be filed with sufficient detail necessary for cost recovery.
Just too bad it is not a published opinion.