You Cannot Force Parties to Use Predictive Coding

United States Magistrate Judge Sallie Kim refused to order Defendants to use predictive coding to identify the location of responsive ESI. In re Viagra (Sildenafil Citrate) Prods. Liab. Litig., No. 16-md-02691-RS (SK), 2016 U.S. Dist. LEXIS 144925, at *51-53 (N.D. Cal. Oct. 14, 2016).

This was the right decision based on the law, but not necessarily the best use of technology.


Judge Kim noted that no case has ever allowed a requesting party to dictate a producing party to use “predictive coding” to find responsive ESI. Id., *51.

On the surface, I agree with the plaintiff’s approach more for conducting a reasonable inquiry or responding to requests for production. The Plaintiff’s proposal was for the Plaintiffs and Defendants to work together in creating key words to identify the locations of relevant information and then responsive ESI with predictive coding. Id., *50-51.

The Defendant’s plan was to use search terms and sampling of ESI to find responsive discovery. The Defendants outlined they would test search terms by searching a sample set of documents, conduct review, and then verifying the search had a high rate of responsive documents. This process would involve the parties exchanging search terms. The Defendants would then use any of the search terms that appeared on both of the parties’ lists. Id. Additional sampling and negotiating search terms would follow.

The Court explained from prior case law that the responding party is the “one best situated to decide how to search for and produce ESI responsive to discovery requests.” Id., at *52.

The issue is whether the requesting party can show the producing party made an inadequate production. Id., at *53. That will not be known until an actual production is made and reviewed by the requesting party. Even then, the standard is not perfection, but whether the “search results are reasonable and proportional.” Id., *52 citing Fed. R. Civ. P. 26(g)(1)(B)).

Document review is a multiple step process. The attorneys conducting the review cannot waste time looking at millions of irrelevant documents. Review can identify terms of art used by the parties, phrases, and insight on how the parties communicated. Search terms can identify emails to review, but there is no promise they will actually be responsive.

Search terms are one step in identifying what to review for relevancy. However, they are not the only step. Responsive ESI might not contain any “search terms” at all. For example, a case about cars with defective transmissions might not ever use the word “car.” Trying to think of every permutation of how to say “car” would be a frustrating process. Imagine the gamesmanship of a producing party saying, “We never agreed to search for ‘automobile’ or ‘vehicle.’”

The issue with search is not terms, but understanding what you are trying to find. Are you trying to show breach of contract? Fraud? Discrimination? This requires testing search terms, searching again, and then leveraging advanced analytics, whether it is a data clustering, concept search, “transparent search” to identify different forms of words (like earn* could include earnings or earnest), proximity searches, and predictive coding. Knowing what you are trying to find, whether it is email messages pertaining to contract terms or hiring policies, can help develop a search plan beyond reviewing files with search term hits.

It is entirely possible search terms alone can find responsive ESI. I prefer using “trust but verify” approach with a workflow that has involves search, review, and predictive coding to identify what could be responsive. However, one party cannot force another to use a specific search technology. That is a question up to the producing party on how to best find what is responsive. Now if there production is inadequate, that is another matter.