Cases involving disputes over when and how predictive coding should be used have recently come before judges in the US, triggering a robust debate in legal circles on both sides of the Atlantic.
A London panel discussion a few weeks ago brought together experts united in the view that the enormous increase in information – fuelled by the prolific use of email, mobile devices and social media – meant that traditional models of document review were increasingly ineffective from cost and practical perspectives.
There was also broad agreement that technology was the only means by which lawyers could cut through vast amounts of information quickly and efficiently.
Lawyers and their clients are used to using keywords as a means of searching information, but this debate concentrated on whether – and how – parties should use next-generation technologies to achieve further efficiencies. The panellists recognised lawyers continued to be reluctant to adopt technology-assisted review and that reluctance is fuelled by several factors.

Creatures of precedent

The first of these is a lack of judicial approval. In the experience of Judge Andrew Peck of the Southern District of New York, lawyers are ‘creatures of precedent’, waiting for cases to establish judicial approval before taking the plunge to use new methods. Luckily, precedents are forthcoming. There have been two cases so far this year in the US where parties have received judicial approval for the use of predictive coding, and he predicts that an increase in such cases will prompt lawyers to use the technology more widely.
But, from the British perspective, Senior Master Steven Whitaker made the point that lawyers in the UK are less likely to be waiting for precedent and more likely to be reluctant to use predictive coding because of a lack of understanding of how the technology works.
The second factor is fear of cost. Master Whitaker’s experience is that lawyers try to avoid employing complex technologies because of the expense involved. However, in his view, they fail to take into account that those costs are part of the end process. If the parties are able to locate the most relevant documents more quickly they will, for example, be able to consider whether settlement is appropriate at an earlier stage.
Robert Lewis, the global director of cyber-forensics and investigations at Barclays Bank, said that while many companies are not yet using predictive coding, the functionality to do so is likely to be available in the disclosure systems they currently use. At Barclays, predictive coding is to be included in the next upgrade of its existing tool and the organisation intends to use it once the upgrade is tested and reviewers trained.

Box phobia

Another factor is the black box. Lawyers are comfortable using all sorts of technology without understanding how they work – for example, Google searches or Microsoft Office. And yet they fear using predictive coding because they do not have a detailed understanding of how it operates.
Master Whitaker said he is interested only in the techniques adopted by the parties in training the software and how it is used. If good quality information is used, and effective quality control is deployed, good results will emerge from the exercise.
He added that, when considering predictive coding, litigants should ‘trust that it does what it says on the tin’, rather than requiring the investigation of complex algorithms. Parties need to be careful about setting accuracy levels and must understand that machines will not give perfection – but nor will human review. He concluded there must be an acceptable margin for error if one is adopting a pragmatic and proportionate process.
Judge Peck agreed that it would be sufficient, in his opinion, for the quality control statistics to show that parties were producing the right documents and that they had undertaken the appropriate sampling of non-relevant documents. Master Whitaker also stressed the importance of the parties’ agreeing on a protocol establishing how the software is to be trained, the level of input opponents have, and how disputes should be resolved.
Both judges agreed that, for predictive coding to be effective, the parties will need to agree that both sides will have an opportunity to consider the seed set of data to ensure the process of training the software is transparent and open. Doing so will ensure it is difficult for opponents to undermine the methodology adopted by a party at a later date.

Greg Wildisen is international managing director at e-disclosure specialists Epiq Systems.