I truly understand the hesitance when it comes to Predictive Review (Predictive Coding), and maybe that’s because I am a Gen-Xer who grew up with Pac-Man and Donkey Kong. It took me a while to wrap my brain around a computer (or a smartphone or tablet) suggesting anything to me, and I am still sometimes baffled by the recommendations on Netflix. But I must admit those recommendations have led to hours of enjoyable viewing, which made me look at Predictive Review in a new light.
What is Predictive Review?
This is a question that I don’t think anyone can answer for all audiences. Each firm, service provider and software provider has a different perspective, but this is my blog post so here is what it means to me. Predictive Review allows a team to code a document based on its content*, which trains XERA to continuously treat similar documents the same way. The system brings back document suggestions, so XERA helps the team get to the documents they want to review in a linear way more quickly.
Content is Queen of the Predictive Review Castle
*The key phrase of the above paragraph is: based on its content. In an email, this means the date, sender and recipients are not deciding factors in predictive review. Only the body of an email is important, which can throw reviewers for a loop on their first predictive review case.
In Predictive Review, the ideas in the document are the deciding factors. For example, in a Predictive Review project, you may have a highly responsive email due to the date or recipient, but the content of the email is not related to the case. For Predictive Review, we would not want to train the system on this document, but for linear review, we would want our team to be aware of this document. The beauty in XERA’s implementation of Predictive Review is that it is so flexible: tags for predictive review can be mixed with tags for a metadata review, and predictive review can be used for both responsiveness/relevancy and issues.
In our Sword Weaklegs database, tags have been set up in the same project and panel to code for relevance and code for the issues of the case: fraud, doping and slander. Information for privilege and confidentiality is also collected. In the image, you can see that the system brought back document that scored high for Relevance and Doping. And, the reviewer was also able to code the document for Confidentiality and Privilege on the same coding tab.
A properly trained Predictive Review project will start to give users feedback after the first iteration (or round) of documents is reviewed. Predictive Review gives you the flexibility of setting a new iteration size for each project so that you’re not locked into a review size. Also, in each project, automatic batching can be enabled so minimal admin work is needed while review is in progress.
Predictive Review can do for our litigation projects what Netflix does for my snowy Saturday nights: it can take what I have liked and show similar recommendations based on my input. As I uncover more great tv that I didn’t even know existed, I understand the impact Predictive Review can make on a project.