Predictive coding in the disclosure process

In a landmark ruling, an English court has approved expressly the use of predictive coding technology in the electronic disclosure process.

In the modern disputes world, document disclosure is often dominated by the enormous exercise of sifting vast quantities of electronic documents produced in the ordinary course, looking for those documents relevant to the issue in the case. That process and the associated costs of providing such disclosure are frequently critical and sensitive issues in commercial litigation both for the courts and the parties to the litigation.

The English Civil Procedure Rules provide that in giving disclosure: ‘a party is required to make a reasonable search for documents’. However, neither the relevant rule or the practice direction deal with how the search for electronic documents should be conducted. Practice Direction 31B states: ‘it may be reasonable to search for Electronic Documents by means of Keyword Searches or other automated methods of searching if a full review of each and every document would be unreasonable’, yet it does not address the extent to which it is permissible to conduct electronic disclosure through the medium of a computer program.

Simply put, predictive coding, sometimes described as technology assisted review, is the review of documents undertaken by proprietary computer software after a human reviewed seed set is used to ‘train’ the software. The software analyses the documents and then ‘scores’ them for relevance to the issues in the case. The purpose of the process is to identify the documents relevant to the case while reducing the time and cost of the review by reducing the number of irrelevant documents.

The case concerns claims by companies against directors in respect of payments allegedly made as a result of breach of fiduciary duty.

The parties agreed case management directions subject to the approval of the court, including as to the use of keywords and predictive coding to be employed in the electronic disclosure exercise. The scale of the exercise concerned more than 17.6 million documents which was reduced to 3.1 million by a process of electronic de-duplication. The parties sought the Court’s approval to use predictive coding.

Master Matthews approved the use of predictive coding technology. The court identified ten factors in favour of the use of predictive coding in the electronic disclosure process, whilst noting that there were no factors of any weight pointing in the opposite direction. The factors identified by the court were as follows.
Experience in other jurisdictions shows that predictive coding software can be useful in appropriate cases. In the absence of English case law on the application of predictive coding, Master Matthews turned to decisions in other jurisdictions notably, the US Federal Court case of Moore v Publicis Groupe, 11 Civ 1279 (ALC)(AJP) and the Irish High Court case of Irish Bank Resolution Corporation Ltd v Quinn [2015] IEHC 175, both of which had endorsed the use of predictive coding.

There was no evidence to show that the use of predictive coding software leads to less accurate disclosure than manual review alone or keyword searches and manual review combined, and indeed there is some evidence to the contrary.

There will be greater consistency in using the computer to apply the approach of a senior lawyer towards the initial sample to the whole document set, than in using dozens of lower-grade fee-earners, each independently applying the relevant criteria in relation to individual documents.

There is nothing in the CPR or Practice Directions to prohibit the use of such software.

The number of electronic documents which must be considered for relevance and possible disclosure in the present case was huge.

The cost of manually searching these documents would have been enormous, amounting to several million pounds. Therefore a full manual review would be ‘unreasonable’ under Practice Direction 31B where a suitable automated alternative exists.

The cost of using predictive coding software would depend on various factors, including whether the number of documents was reduced by keyword searches. The estimates given in this case varied between £181,988 (plus monthly hosting costs of £15,717), to £469,049 (plus monthly hosting costs of £20,820). In either case, it was considerably less expensive than the alternative of a full manual review.

The ‘value’ of the claims made in this litigation is in the tens of millions of pounds. Therefore the estimated cost of using the software was proportionate.

The trial in the present case is not due to take place until June 2017, so there would be plenty of time to consider other disclosure methods if for any reason the predictive coding software route turned out to be unsatisfactory.

The parties had agreed on the use of the software, and also how to use it, subject only to the approval of the Court.

In keeping with the Jackson reforms, which focussed on costs management, a theme of this judgment is cost control and the comparison between the cost of a manual review against the cost of predictive coding. Master Matthews noted that, ‘provided that the exercise is large enough to absorb the up-front costs of engaging a suitable technology partner, the costs overall of a predictive coding review should be considerably lower’. Moreover, the Master expressly recorded that the decision is in accordance with the overriding objective in CPR1.1(1) of, ‘enabling the court to deal with cases justly and at proportionate cost’.

The decision is important as being the first time an English court has provided a detailed judgment closely examining and approving the use of predictive coding. Also of assistance is the court’s confirmation that predictive coding can satisfy disclosure obligations, which will provide litigants with encouragement to explore the method in future cases (although industry consensus is that predictive coding has been used for some time by parties to litigation). It is now almost inevitable that parties to litigation will use developments in technology to reduce the costs of litigation as a whole and disclosure exercises in particular. In many cases, predictive coding will help to achieve this.

However, it should be stressed that this decision does not open the way for predictive coding to be used in every case. Several of the factors identified by Master Matthews were specific to the particular case. The judgment notes that whether predictive coding would be right for other cases will depend upon the particular circumstances in each case. Predictive coding is but one option.

The case therefore provides an important incremental step in the development of the law and practice around predictive coding, but parties will have to continue to consider carefully what methods of disclosure are appropriate in any given case.

This article was co-written by Sarah Bramley.