Predictive coding endorsed again by English High Court

In brief

In its second decision on predictive coding this year, the English High Court has again accepted the use of this innovative technology in discovery, this time in a contested application. Partner Duncan Travis, Managing Associate Kate Austin and Law Graduate Yi-Ling Ng examine the case and its implications.

Recap on predictive coding

Predictive coding utilises sophisticated software to enable the efficient and cost-effective identification of relevant documents within large document sets. The software 'learns' from decisions on relevance made by a human reviewer over a sample set of documents, and then 'predicts' automatically which documents will be relevant across a larger document set.

The decision in David Brown

In David Brown v BCA Trading Limited and Others², handed down on 17 May 2016, Registrar Jones of the English High Court considered the respondents' application that discovery be undertaken using predictive coding. The proceeding concerned a shareholder dispute in which the plaintiff sought to recover a sum of more than £20,000,000.

The first respondent, BCA Trading Limited (BCA), held approximately 500,000 potentially relevant documents. BCA submitted that predictive coding would be the most reasonable and proportionate method of discovery. It also provided evidence that the costs of predictive coding would be in the region of £132,000, compared to the costs for a traditional key word search approach of at least £250,000 (with that figure potentially as high as £338,000 in a worst case scenario).

The plaintiff opposed the use of predictive coding, apparently on the basis that it was a new technology that may not be as effective as traditional but more expensive methods.

In endorsing the use of predictive coding, the Registrar was heavily influenced by the significant cost savings that BCA submitted could be achieved, observing that:

• there was no factual or expert evidence to contradict the assertion that predictive coding was the most reasonable and proportionate method of discovery;
• in particular, there was nothing to suggest that predictive coding would not be able to identify the documents which would otherwise be identified through traditional key word searching; and
• he was 'encouraged' by the fact that directions were in place requiring the parties to narrow the issues in dispute and to sit down before predictive coding began to discuss the criteria to adopt and the general process of disclosure. These discussions were between experienced solicitors who could be relied on to take an approach consistent with their duties as officers of the court and with their duties to their clients.

The Registrar also considered that the applicable factors set out in Pyrrho were satisfied, stating that the Pyrrho decision 'must encourage and sustain my decision' to order predictive coding.

Implications

Both Pyrrho and now David Brown reflect the growing momentum in overseas jurisdictions towards the use of predictive coding to minimise the time and cost of civil litigation. These developments can only be of assistance to parties to litigation involving significant document volumes, although use of the technology is not without risks which need to be considered carefully and managed appropriately. We would expect that the Australian courts will take a similar approach before long.

For those wishing to use predictive coding in discovery, David Brown highlights the importance of providing evidence to satisfy the court that it is the most reasonable and proportionate method of discovery in the circumstances. In the event that predictive coding is ordered, David Brown also suggests that the parties to litigation may be required to confer on the process and parameters of predictive coding and that any ensuing disputes may need to be resolved by the court.

Finally, the volume of documents in David Brown (approximately 500,000) was significantly less than in Pyrrho (approximately three million) demonstrating that the technology is not only suited to handling claims with large datasets but can also be critical to achieving proportionality in claims involving more typical size datasets.