Predictive coding: the future of electronic document production?

In brief

A recent decision of the English High Court may pave the way for the use of 'predictive coding' in large scale discovery and regulatory investigations in Australia. Partners Nick Rudge and Duncan Travis, Managing Associate Kate Austin and Associate Emily Giblin look at the benefits and risks of the new software.

What is predictive coding?

Predictive coding utilises sophisticated software to enable efficient and cost-effective review of large document sets. It involves the software learning review decisions from a human reviewer and applying these review decisions automatically to large document sets. At a high level, the steps involved in predictive coding are as follows:

• the use of culling techniques to arrive at a target document population;
• review by a senior lawyer of a ‘training’ set of documents;
• software ‘learning’ from human coding and applying predictions to an additional population of documents;
• any conflicts or issues identified between human coding and predictions are resolved to bring a sufficient degree of certainty to the software's predictions;
• the software makes predictions on remaining documents in the population; and
• relevant documents are sampled or reviewed by human reviewers and then discovered or produced. The extent of human review at this stage will depend on the particular case.

Predictive coding is viewed as a solution in cases involving very large document sets where traditional culling techniques, such as key word searching, still leave a very large set of potentially relevant documents that would be impracticable to review. The software learns from the human reviewer by applying complex algorithms to the human reviewer's coding decisions so as to identify similar documents. In this way, it is similar to internet advertising which is designed to display advertisements similar to content you have previously viewed on other websites. Predictive coding can also be used alongside more traditional approaches to prioritise reviews, speeding up review times and the ability to allocate documents to the most suitable resource (lawyers, a document review team or LPO) in order to achieve the most cost effective and quality outcome.

Acceptance of predictive coding in Australia?

Predictive coding is sometimes referred to as the 'future of discovery'. While the benefits of predictive coding are clear, there has been a question mark over whether courts will accept predictive coding as a valid discovery tool.

Internationally, predictive coding has been judicially considered in jurisdictions including America, Ireland and England. In America, there is a growing body of case law around various aspects of this technology and its application. A number of American decisions have endorsed the use of predictive coding in cases where a technology driven discovery process is appropriate (for example, where a very large number of documents need to be reviewed).

In Ireland, the High Court has also recently approved of the use of predictive coding. In Irish Bank Resolution Corporation Ltd & Others v Quinn & Others¹, the parties disagreed over whether predictive coding should be used to review approximately 680,000 documents. The court held that predictive coding was an acceptable method of document review. The court's decision was heavily influenced by evidence suggesting that the technology, when applied to large document sets, is as accurate as manual document review but more efficient and cost effective.

The recent English decision of Pyrrho Investments Limited & Anr v MWB Property Ltd and Others² was the first time predictive coding has been considered by an English court. The Pyrrho case was a high value, complex, multi-party civil dispute involving some 3.1 million documents. The parties had agreed between themselves that predictive coding was appropriate, however they sought the court's approval for its use given the novelty of the technology. In a decision handed down on 16 February 2016, the court approved the proposal to use predictive coding. The court based its decision on a number of factors, including:

• evidence showing that predictive coding is at least as accurate as a manual review;
• the increased consistency of review results likely to be produced by predictive coding (as opposed to human decision making);
• the very large volume of documents to be reviewed;
• the cost savings provided by predictive coding when compared against the cost of manually reviewing such a large set of documents;
• the high value of the claims in dispute, meaning that the cost of the software was proportionate;
• the fact the parties both consented to the use of predictive coding; and
• the substantial length of time before trial, meaning that traditional discovery methods could be reconsidered if the results produced by the predictive coding technology were unsatisfactory.

While there has been no judicial consideration of predictive coding in Australia to date, Pyrrho is likely to be persuasive authority if the issue comes before an Australian court. Given the growing use of predictive coding internationally, and the endorsement it has received from the judiciary in America, Ireland and England, the software is likely to find support in Australian courts. This is particularly the case given the increasing focus of Australian courts on the efficient and cost effective conduct of civil litigation and the potential uses of technology to support these aims. Parties wanting to use predictive coding would, however, need to provide evidence to satisfy the court that employing this technology would be appropriate in the particular circumstances of their case.

Risks

The use of predictive coding may also give rise to risks which parties will need to consider how to manage. One such issue is the extent to which responsive documents are human reviewed before being discovered. In the absence of human review prior to discovery, the risk of privileged, commercially sensitive and confidential, or irrelevant documents being inadvertently disclosed is heightened. If a party discovers large quantities of irrelevant documents, for example, they may be exposed to adverse cost orders.

Is predictive coding right for your dispute or regulatory investigation?

Whether predictive coding is right for your case will depend on the subject matter, volume and type of documents involved. For example, predictive coding may not be appropriate for construction disputes which involve a large number of drawings and plans.

Where predictive coding is appropriate, it should not be viewed as a simple 'one click' answer which removes the need for lawyers to be involved throughout the review process. Decisions about relevance and privilege will need to checked by lawyers and conflicts and issues resolved on an ongoing basis. Further, it must be remembered that the review decisions of the senior lawyer on the training set of documents are critical to setting the foundations of the review. This means that time must be invested upfront by senior lawyers to understand the key issues in the case and how the documents under consideration are relevant to those key issues.