Predictive coding — it is a frequent topic of conversation at national e-discovery conferences, in legal technology-focused publications, and any other place where e-discovery practitioners are likely to congregate. In 2012, Da Silva Moore v. Publicis Group thrust the concept of predictive coding upon the legal world, and the e-discovery landscape has never been quite the same. It’s easy to say that no other case has generated as many CLE approved sessions yet delivered to such a relatively small percentage of the legal profession.
But, Da Silva Moore is not without its faults – most notably, it has led to several erroneous conclusions. The first is the question about predictive coding and when “seed sets” (the documents used to train the predictive coding engine) must be shared. This, perhaps more than any other aspect of the case, made many people believe that if Da Silva Moore represented what predictive coding was all about, then they would pass. It also led many people to believe that all predictive coding review processes were basically the same, another factually incorrect conclusion.
The last two years, however, saw a transformation of this erroneous opinion. Although predictive coding continues to dominate most e-discovery conversations, the discussion around the technology and process has increasingly become more sophisticated and nuanced. People, generally, are no longer primarily concerned with “Can they use predictive coding?” The focus now seems to be “How can they use predictive coding?”
Of course, it is not surprising the discourse has evolved, but what is surprising is how quickly the evolution has occurred, especially considering the glacial speed at which most lawyers move when adopting new technologies. Nevertheless, predictive coding has still not been fully unleashed; its power (and therefore the initial promises that came with the technology) has not been fully realized. The question is, why?
One culprit may be the technology vendors themselves. Many of the technology companies that offer predictive coding solutions have needlessly restrictive workflows for using predictive coding — “use it this way or else the technology will fail,” they frequently warn their prospects. Not surprisingly, when the technology is thus constrained, so too is the imagination, and solutions to discovery challenges tend to remain elusive.
Hence, despite the introduction of predictive coding document review costs continue to plague the litigation process because not enough litigants have used predictive coding and other advanced forms of review. Many of the vendors offering predictive coding claim it is a way to eliminate “first pass review.” But what is first pass review? It is merely a culling method using, traditionally, expensive human labor — “this stack is garbage, this stack might be worth looking at, this stack is potentially privileged,” etc.?
If your stated purpose for predictive coding is to eliminate first pass review, then your main objective for predictive coding is to cull documents. A laudable goal, but using predictive coding in this limited capacity belies the strength of the technology. Although potentially faster, it’s also a misuse of otherwise powerful resources. Unfortunately, many people eschew the use of predictive coding believing it simply wasn’t right for their particular case.
This limited understanding of predictive coding is simply incorrect. It can be a culling tool, but it is not just a culling tool. It’s one of the reasons the future will see more integration of predictive coding solutions in review strategies. As more review platforms offer predictive coding and, more importantly, offer predictive coding without restrictive workflows, the rate of adoption will increase dramatically. Once more people have this powerful technology in their hands and have the power to use it how they want — that is when true innovation is possible.
Fortunately, there are some solutions in the market that already stress workflow adaptability to accomplish the particular goals of a case. If your goal is to simply remove 80 percent of the documents from human eyes — there is a workflow for that. If you need to understand the communications between two people and any other person in the organization who might have also been discussing the same concept (albeit using different verbiage) — there is a workflow for that too. The reality is that there should be a workflow that can meet the needs of any case.
Predictive coding solutions that only offer one workflow are not versatile enough to meet the needs of a widening customer base. But, when considering the multitude of ways a user can structure a workflow around predictive coding technology, two inescapable conclusions present themselves: This flexibility is not the widely held view of what predictive coding is and is not even the way almost all software vendors talk about the technology, and if this rigid and static view is what predictive coding really is, then predictive coding (as we have come to know it) is dead.
When we consider what’s next for predictive coding, we need to consider first what we mean by predictive coding. If predictive coding means a restrictive workflow designed only to eliminate first pass review (i.e. culling), then predictive coding will likely decrease in importance (as better ways to cull documents rapidly evolve). But, if we embrace a more robust and flexible view of predictive coding, then we should see why predictive coding will become even more prominent in 2014.
More usage will allow creative users to develop more creative solutions to old problems and predictive coding will go from a technology limited in utility for the AmLaw 50 and their Fortune 100 clients to a much wider, more diverse audience. Predictive coding will become a valuable tool in the tool box any time you need to rapidly develop the facts of a case or develop a deeper understanding of who said what and when. After this evolution, predictive coding will be truly ready for the masses.