Late last year, the Supreme Court of Victoria released a judgment that resolutely approved the use of predictive coding (or TAR - technology assisted review) for the purpose of discovery in a large litigation matter, the first decision of its kind in Australia.
In McConnell Dowell Constructors (Aust) Pty Ltd v Santam Ltd & Ors (No 1)  VSC 734, the lawyers involved in the dispute faced 1.4 million potentially relevant documents for review, a number already reduced down from a massive 4 million documents using de-duplication software. The Court estimated it would take a solicitor 583 working weeks to get through them. The parties were unable to agree on a solution between them, so turned to the Court for guidance.
After engaging a special referee to assist in answering questions on discovery processes, the Court and the parties agreed to use predictive coding and the Court made an order to that effect.
This is the latest in a series of cases across multiple jurisdictions that recognise and give momentum to this emerging technology. Commentators expect to see an increased use of the technology in Australia now that a court has formally approved it.
So what exactly is predictive coding, is it reliable, and have the New Zealand courts sanctioned its use?
Predictive coding is a sophisticated software that performs the task of reviewing documents for relevancy in a proceeding. Document review is usually carried out by solicitors working on the matter who have an understanding of what may or may not be relevant. As large commercial disputes can involve millions of documents, the discovery phase is incredibly costly and can drag out the litigation process considerably.
Predictive coding greatly reduces the time and costs involved in discovery, as much of the work is done by a computer. The process includes measures that ensure the software performs the task as accurately (and as much like a human) as possible.
The predictive coding process was summarised in the UK decision Pyrrho Investments Limited v MWB Property Limited  EWHC 256 as follows:
- The parties settle on a predictive coding protocol that outlines how the software will do its job
- A solicitor who would otherwise be carrying out the review would then take a sample set of documents, decides on their relevancy and categorises them appropriately
- The software is then 'trained' using this sample batch of documents to complete the same job for all the documents. The software analyses the documents for common concepts and language used
- The software then carries out further quality assurance exercises before a sample is selected for review by the solicitor. Where the solicitor overturns decisions made by the software, this decision is fed back into the system for further learning
- The process of overturning may be repeated several times until a suitable level of agreement is reached
- At this stage, a list of relevant documents can be produced.
Evidence suggests that predictive coding is at least as accurate as, and probably more accurate than, the manual method of identifying documents (examined in Irish Bank Resolution Corporation Ltd v Quinn  IEHC 175). The key is to be consistent during the human-assisted stages, and to agree carefully on parameters beforehand.
New Zealand Courts are yet to examine predictive coding and, perhaps for this reason, the technology is not utilised extensively in discovery here at this stage. However, as far back as 2011, our High Court Rules (HCR) have referred to predictive coding or 'document prioritisation technology' as an option for discovery, and state that parties should consider and agree on whether such methods are appropriate in the case at hand. Judge David Harvey has spoken about the benefits of predictive coding, via obiter remarks in Dotcom v United States of America  DCR 661 and in articles published on the Auckland District Law Society website. In Dotcom, Judge Harvey reminded the parties that "there are procedures available…in the civil arena that will enable the prompt disclosure of relevant information", including predictive coding.
So while predictive coding remains in its early stages and has not been carefully examined in our courts, there is acknowledgement of the technology within New Zealand, and several well-regarded judgments from other jurisdictions that would be persuasive if the question arose here. It's worth remembering that a court sanction is not actually needed. Parties can agree between themselves to use predictive coding but, as the decisions in other jurisdictions indicate, parties seem inclined to first seek guidance from the court.
Under the High Court Rules parties are obliged to co-operate to ensure that the processes of discovery are proportionate, to consider options to reduce the scope and burden of discovery, and to ensure technology is used efficiently and effectively. Predictive coding will no doubt take hold as an effective tool to make information manageable in litigation in this increasingly digitised world.