Landmark Irish Judgment on use of TAR and Predictive Coding in Discovery
Earlier this month, the commercial division of the Irish High Court gave its judgment in IBRC & Ors. v. Sean Quinn & Ors.1 on the proposed use of technology assisted review ("TAR") and predictive coding methods to assist the Irish Bank Resolution Corporation ("IBRC") to review some 680,000 documents. In an important milestone for the future use of the technologies, Fullam J ruled that TAR using predictive coding complied with the rules of court governing discovery.
The Discovery Process
A large component of the fees in litigation is the cost of making pre-trial disclosure of documents to an opponent (known in Ireland as "discovery"). These fees often comprise as much as 50% of the total costs of a case; a function of the explosion of documentation created in modern workplaces, and the need for lawyers to review large swaths of that documentation prior to making discovery.
The discovery of soft-copy data typically follows a five stage process involving:
(a) the identification, preservation and retrieval of potentially relevant data;
(b) the upload of that data onto a data retention system;
(c) the processing or filtering of the data to remove duplicates, program files, etc., and to create a 'document pool';
(d) the manual review of each document in the document pool by a team of junior lawyers using a document review platform; and
(e) the production of the relevant documents to the opponent.
Technology has been assisting lawyers to perform all stages of this process for some time. However, there has traditionally been no alternative but to manually perform step (d). While the pool of relevant documents could certainly be reduced with the use of keyword searching and the identification of key custodians and date ranges, the decision-making process of identifying relevant documents in the document pool has always been manually performed by lawyers.
Typically, manual review requires multiple reviews of the document pool. Though methodologies can vary from case to case, a first review will typically be performed to assess the relevance of each document in the document pool. A second review will then be performed on the documents classed as relevant during the first review, in order to assign them to the categories of documents that the party has agreed, or has been ordered, to discover. A third review will then be performed in order to assess whether any of the relevant and categorised documents are privileged or require redaction to prevent the disclosure of irrelevant commercially sensitive information. Quality control reviews are also performed at each stage by senior lawyers.
In commercial cases involving tens of thousands, if not hundreds of thousands of documents, this methodology requires a lot of man-hours to implement and, as a result, can incur very significant costs.
What are TAR and Predictive Coding?
TAR is a term used to describe a variety of different computer-aided review technologies, including data mining, analytics and predictive coding. Predictive coding operates by learning from the coding decisions made by a small team of senior lawyers on a small subset of documents, and extrapolating those decisions across the entire document pool. Different providers have different software but each uses sophisticated algorithms to identify patterns in the documents the subject of the relevance decisions by the lawyers. Over a series of iterative rounds of testing, learning and quality control, the software's decision-making is honed until the user can be confident that it will capture a statistically-justifiable proportion of the relevant documents in the document pool. Similar mathematics are used in many industries today, including in the spam filters commonly applied to email inboxes.
Predictive coding is, however, not a panacea for all of discovery's ills.
First, a relatively large and homogenous document set is required so that the software can identify consistent patterns in the language of the individual documents, and for predictive coding's economies of scale to operate effectively.
Second, documents containing little or no text, or foreign language, are not amenable to predictive coding's methods. They must be extracted from the document pool and manually reviewed.
Third, predictive coding is at present suitable only to replace the first pass relevance review in a discovery project. While predictive coding can emulate and replace the binary choice between relevance and irrelevance, it cannot cope with the multiple choices involved in the categorisation of documents. Therefore, after predictive coding has identified the relevant documents in the document pool, manual review must be performed at second pass (to assign the relevant documents into their appropriate categories) and third pass (to identify and prevent the disclosure of privileged documents).
After discovery was ordered by the Court, IBRC collated its data and used keyword searching to reduce it to a pool of 680,809 potentially relevant documents. IBRC calculated that it would take 10 lawyers nine months manually to review these documents, at a cost of €2m. Rather than incur this cost and delay the proceedings, IBRC proposed instead to use predictive coding to perform the review. It relied upon expert evidence and a number of studies that showed TAR to be more accurate in identifying relevant documents than a traditional manual review. To assuage any concerns of the defendants, and ultimately the Court regarding the use of predictive coding, IBRC had produced a draft protocol and had invited input from the defendants on its provisions.
While the defendants initially engaged with IBRC, they ultimately refused to agree to the use of the technology. IBRC asked the Court to determine whether TAR, and predictive coding in particular, complied with the Irish Court's rules on discovery and, if they did, the most appropriate methodology for the use of the technology.
Fullam J stated that while the rules of court are silent as to the means that a party should employ in order to comply with its discovery obligations, the principles of economy and expedition are integral considerations in that context. Having regard to US case law and to the academic studies which asserted that TAR was more accurate than manual review, the Court posited that, even assuming TAR to be only equally as effective as manual review, the use of the technology would – by virtue of its relative speed – still permit a more expeditious and economical discovery process.
However, because TAR involves man and machine interacting, Fullam J ruled that an appropriate methodology for its use must contain appropriate checks and balances to render the accuracy of each stage capable of being independently verified. Fullam J further found that a balance must be struck between the right of the party making discovery to determine the manner in which discovery would be provided, and participation by the requesting party in ensuring that the methodology chosen by the discovering party was transparent and reliable. Provided that these conditions were complied with, it was held that a TAR project using predictive coding could be used to discharge a party's discovery obligations under the rules of court.
The Court then assessed IBRC's proposed TAR protocol and, subject to one minor amendment, was satisfied that the protocol contained adequate controls on reliability, and provided for appropriate levels of participation by both sets of defendants.
The Irish judiciary has been keen to impress upon parties to litigation the need to contain costs during the discovery process. Experience in the United States suggests that the use of TAR and predictive coding can dramatically lower the cost of making discovery in document-intensive actions, while also expediting the discovery process.
Seeking engagement and input from an opponent on the details of a TAR protocol while still exchanging pleadings, or at the close of that process, is advisable. However, a party is not required to implement its opponent's suggestions. Ultimately the party making discovery and its legal representatives are accountable to the court and not to their opponent for the quality of their discovery, and thus their TAR protocol.
TAR using predictive coding now has the Irish High Court's imprimatur and parties are free to use it to comply with their discovery obligations and so reduce costs. However, that approval is caveated, and a party's legal representatives and technical advisors must devise a protocol that ensures the reliability, integrity and transparency of the predictive coding project.
1 Irish Bank Resolution Corporation Limited & Ors v. Quinn & Ors  IEHC 175;  3 JIC 0306. Please note that the Senat defendants have since appealed the judgment and order to the Court of Appeal.