Skip to content Skip to footer

Loading Results

Can Active Learning cut through the knot of privilege reviews?

The complexities of identifying privileged documents have made a standardised workflow that fits every case difficult to incorporate into eDiscovery processes. Active Learning could provide a way forward, though is yet to offer a universal panacea.

The past few years have accelerated technological advances, with the impact of the COVID-19 pandemic being a particular driver. With the ever-increasing scale of document reviews, we’ve seen greater adoption of Technology Assisted Review (TAR) amongst eDiscovery practitioners and lawyers.

As many will know all too well, however, screening out privileged documents is one of the areas that is challenging. We’re often asked by lawyers whether TAR, and Active Learning in particular, could ease the burden of privilege reviews in the same way it has done for relevance reviews. Well, yes and no.

Difficulties with identification

It’s worth reminding ourselves why privilege reviews are such a headache within eDiscovery.

An initial relevance review will be undertaken by the first level review team, who may not have a legal background and are therefore unable to identify privileged material correctly. It is hence typically necessary for more expert legal teams, many of them senior lawyers, to perform the privilege review.

Keyword search terms designed to identify privileged material are often broad by necessity (to mitigate the risk of privileged documents being disclosed). Consequently the set of keyword-responsive documents is frequently very large, and this places a heavy burden on the expert legal teams.

eDiscovery practitioners have found ways to reduce the number of false positives resulting from the search terms; for example, here at PwC, we’ve developed a bespoke method for removing email footers/disclaimers before running privileged search terms. Nonetheless, the volume of documents for review can still be a large hurdle.

Much of the problem lies in the fact that identifying a privileged document is rarely as straightforward as simply looking for ‘privileged’ in the subject, or finding all emails between a company and its legal team. This means that highly experienced lawyers frequently have to step-in to judge what is and isn’t privileged material.

“We've worked with a number of clients struggling with the burden of privilege review. Technology assisted review has been proven to help firms prioritise documents, and reduce time and costs for disclosure. Could implementing Active Learning be as beneficial on privilege reviews as it is for relevancy reviews?"

Is Active Learning the solution?

Using Active Learning to prioritise documents and then concluding the review after a designated cut-off (determined using various reporting metrics) has the potential to greatly reduce the number of documents in a privilege review.

However, difficulties remain. Prominent among these is the ‘four corners rule’ recommended for TAR, under which documents should be considered and reviewed on their own merits in isolation from any associated metadata or family documents (such as attachments to an email). This is largely incompatible with privilege reviews, where consideration of this broader context is often key.

In a related issue, initial Active Learning implementations did not necessarily keep family documents together - though this option has since been implemented in the Relativity review platform, for example.

Still some work ahead

It could be argued that including families and reviewing every document contained within them may not always be the best approach. It could also mean that lawyers are spending too much time wading through documents that are unlikely to impact on the privilege determination. Despite losing the context within the family, the reviewers’ time may be better spent solely focusing on documents more likely to determine privilege.

This leads us to the key question of whether an Active Learning model could learn what a privileged document ‘looks like’ in order to identify and present others which are similar.

Given that the complexities and nuances of a privilege review can make determinations challenging even for experienced lawyers, and the fact that the parameters of Active Learning mean emphasis is placed only on the contents of a document, Active Learning is unlikely to be a practicable solution in its current implementation. For now, this technology is best suited to relevancy, rather than privilege, reviews. Whether this changes in future remains to be seen.

Here to help

We at PwC have extensive experience of using Active Learning to prioritise documents in a review, reduce the number of documents in the search and therefore reduce the time and cost spent on review. If you would like to discuss whether Active Learning is appropriate for your eDiscovery project, please feel free to get in touch.

This article is the first in our new series discussing Technology Assisted review. Look out for our next topic on portable models - could this be the silver bullet for speeding up document reviews?

Contact us

Matt Joel

Matt Joel

Partner, PwC United Kingdom

Tel: +44 (0)7809 552273

Christopher Dean

Director, Digital & Forensic Investigations, PwC United Kingdom

Tel: +44 (0)7841 570 567

Follow us