image of html

BCLP obtains new and unique order on the use of Predictive Coding technology in disclosure


Posted by , on

Summary: BCLP has obtained a new and unique order in a contested application on the use of Predictive Coding technology in disclosure. The order permits BCLP’s client to use Predictive Coding technology to identify documents that are highly likely to be irrelevant for the purposes of disclosure, and to dispense with the manual review by an individual of those irrelevant documents.

This will result in significantly increased efficiencies, and further demonstrates BCLP’s unique in-house forensic technological capabilities.

Approaches to disclosure review

BCLP is acting for a large financial institution in a commercial dispute.

BCLP employed various Predictive Coding technologies in its disclosure review process in order to prioritise potentially relevant documents for manual review, and to designate potentially irrelevant documents for further testing and analysis.

The other parties’ solicitors adopted a “traditional” approach to the disclosure review process, involving a manual, linear review of hundreds of thousands of documents. With the technology now available, this approach is outdated, inefficient and expensive.

How do Predictive Coding technologies work?

A human reviewer manually reviews and codes a sample set of several thousand documents, using coding assignments such as “Relevant – To Disclose” or “Irrelevant – Do Not Disclose”. The Predictive Coding software creates an algorithm based on the coding assignments made by the human reviewer. This algorithm tries to model what characteristics make a document “Relevant – To Disclose” or “Irrelevant – Do Not Disclose”.

The software then applies that algorithm to the unreviewed documents, and gives each document a score of between -1 and 1, using increments of 0.1, according to that document’s perceived relevance. A score of -1 denotes that the document is highly likely to be irrelevant, and a score of 1 denotes that a document is highly likely to be relevant. Similarly, a score of -0.4 denotes that a document is moderately likely to be irrelevant, and a score of 0.4 denotes that a document is moderately likely to be relevant.

The software model can be rerun at any stage and an updated algorithm generated, which takes into account the coding on further documents that have been reviewed.

What are the benefits of using Predictive Coding technologies in this way?

  • Speed: The algorithm can be applied to a population of tens or even hundreds of thousands of documents in a matter of minutes. The document population can then be analysed holistically, and the number of potentially relevant documents immediately identified.
  • Prioritisation: The manual review of documents can be prioritised according to the “relevance” score a document has been ascribed. Documents that have been identified by the software as more likely to be relevant can be prioritised for manual review over those documents which have been identified as more likely to be irrelevant.  
  • Accuracy: The use of a single algorithm based on machine learning is more likely to conduct an accurate and consistent review than a team of human reviewers. The algorithm itself can evaluate how accurate it considers it is. It can be stress-tested, updated and improved at any stage of the process.
  • Efficiency: Documents that have been identified by the software as more likely to be irrelevant can be set aside for further testing and analysis, either by way of manual spot-checking or application of further technologies. This means that time is not wasted manually reviewing a large number of completely irrelevant documents.

The facts of the case

The dataset in this case comprised approximately 250,000 documents. The Predictive Coding technology, having been applied in the manner described above, identified 63,000 documents as being highly likely to be irrelevant.

BCLP requested the consent of the other parties to dispense with the manual review of these 63,000 irrelevant documents, on the basis that the cost of manually reviewing these documents would be significantly disproportionate to the number of relevant documents that were likely to be identified.

Upon the other parties’ refusal to consent, BCLP applied for an order to permit them to dispense with the manual review of these 63,000 documents, subject to stress-testing the algorithm and spot checking the unreviewed document population.

The court’s decision

The court noted that BCLP’s proposal would be highly beneficial in terms of efficiency and cost, and ordered that BCLP were permitted to dispense with the manual review of 63,000 documents that had been identified by the technology as being highly likely to be irrelevant.

This is the first order of this kind, and is likely to represent the beginning of a trend of highly sophisticated technologies being permitted, and even actively encouraged, by the court in the disclosure process.

BCLP’s use of legal technology

BCLP is at the forefront of this innovative wave of sophisticated disclosure technologies. We are one of only a few firms with an in-house data processing, hosting and document review capability, and are almost unique in having an in-house Predictive Coding resource.

Stay informed

Sign up to receive email alerts from our award winning Expert Insights team

Sign up now

See more insights by category

This site uses cookies to help us manage and improve the website, your browsing experience, and the material/information we send to our subscribers. For further information about cookies, including how to change your browser settings to no longer accept cookies, please view our Privacy Notice. Otherwise we will assume you are OK to continue.