Keywords aren't enough anymore

Search for ‘predictive coding’ in the UK law reports and (as at the
end of July) you will find only one reported case, Goodale and
Ors v the Ministry of Justice and Ors [2009] EWHC B41, which contains a passing reference
at paragraph 27 that there is “software that will effectively score each document as to its likely relevance and which
will enable a prioritisation of categories within the entire document set.”

Similarly, there have only been a limited number of predictive coding cases in the US. The first was Da Silva Moore v Publicis Groupe SA in February 2012,
and there have been about a dozen since. However, the landscape is changing both here and in the US and there is a need to highlight a couple of the practical points to bear in mind. For, as US District Judge Michael Baylson emphasised in May 2014 in re Domestic Drywall Antitrust Litigation, to ignore these electronic capabilities “is like pretending businesses still communicate by smoke signals.”

Simply put, predictive coding allows you to use technology
to review not just hundreds of thousands of documents, but millions, cost and time efficiently.

Disclosable documents

From a bird’s-eye perspective, predictive coding operates
much like the way some internet companies track information about your purchases and other customers’ purchasing similar items in order to suggest what you should buy next.

Similarly, predictive coding learns from the solicitors reviewing the electronic documents in your case to understand which ones are
(or are not) disclosable.

It assesses every document as
a whole, then individually scores (as a percentage) the likelihood that the unreviewed documents in your database are disclosable. In this way, predictive coding suggests to your reviewers
which documents it thinks are disclosable. It is, therefore, much more sophisticated than keyword searching, which only looks for a word or phrase in your document.

There are several different vendors offering their own predictive coding software.
All use documents from the specific case, which have been reviewed by the solicitors instructed on it, to train on use
of their software. The use of predictive coding is, therefore, always bespoke to the case in which it is used.

Practical points

• Predictive coding can be particularly useful if the keywords you have agreed to use to search for documents are imprecise, i.e. if you believe your database contains a large number of irrelevant documents. Predictive coding can be used to help you further refine the number of documents for review. Equally, if you have received a large amount of disclosure, predictive coding can rank/refine the number or order in which you think it is worth reviewing those documents.
• Predictive coding can also be used more widely, for example, as a valuable tool for quality control purposes, checking the decisions your reviewers have made about the discloseability of documents. This is especially useful in cases where you have a large number of people reviewing your client’s documents for disclosure.
• Using predictive coding means that a percentage will appear next to each document in your database. This indicates how confident the software is that a document is disclosable – ‘disclosable’ as opposed to ‘relevant’, because, for example, if a document scores 99 per cent it does not mean that it will be the smoking gun in your database. Conversely, a document with a low percentage, for example 25 per cent could be very relevant and important to your case. It is worth considering what additional searches or checks should be combined with the use of predictive coding to complete any review.
• The decisions made by those within your legal team about the relevance of any documents (particularly at the early stages of your review) are used to train how the predictive coding works across the rest of the documents. These initial review decisions are fundamental to how the rest of the process will work. It is, therefore, important that the early stages of the review should be conducted by solicitors who have the best knowledge of the issues in dispute. This seems an obvious point, but it is worth bearing in mind as historically the first stages of document review in many cases have been conducted by more junior lawyers with more senior lawyers reviewing these results. Predictive coding turns this idea on its head. SJ

Celina McGregor is an associate at Herbert Smith Freehills and an executive committee member of the Junior London Solicitors Litigation Assocation