A Leap Forward for the Recognized Use of AI and Predictive Coding in Insolvency Trials and Investigations

When the High Court of England and Wales handed down judgment in the case of Brown v BCA Trading Ltd, it marked what is believed to be the first test of technology assisted review (TAR) for disclosure at a full trial in England.  The technology is being used increasingly and, combined with recently-proposed changes to the English disclosure regime, could result in more legal cases becoming economically viable to fight and lead to greater recoveries for creditors.

TAR also has applications beyond litigation, which can be particularly useful for insolvency practitioners (IPs, as they are often called) appointed to a company who need to find out key information and secure assets quickly with limited funds.

The BCA judgment

In May 2016, Berwin Leighton Paisner won the first contested application for its client BCA to use predictive coding in Brown v BCA Trading Ltd [2016] EWHC 1464 (Ch). The case built on the progress of the February 2016 landmark ruling in Pyrrho [2016] EWHC 256 (Ch), in which the High Court approved the use of predictive coding for disclosure at the request of both parties. In the BCA case it had been alleged by the petitioner, Mr Brown, that BCA had acted in a manner that was unfairly prejudicial to his minority interest in Tradeouts, an online car-dealing platform in which BCA had purchased a majority interest in 2014. Mr Brown sought an order for BCA to purchase his shareholding and valued the shares at £20m.

The case proceeded to a 12-day trial in October 2017 culminating in a judgment in BCA’s favour. The fact-heavy nature of the case (involving broad allegations of unfairly prejudicial behaviour and bad faith) meant that the disclosure was key to deciding it. Murray Rosen QC, sitting as deputy High Court Judge, found that the documents directly disproved a number of the petitioner’s allegations and cast doubt on other aspects of his evidence. He found in favour of BCA on all issues, dismissed the unfair prejudice petition and awarded BCA its costs.

Predictive coding for the uninitiated

TAR encompasses a variety of document review technologies, but the expression is primarily used to refer to predictive coding (explained below). At its most basic, the sort of keyword searching that is standard practice in electronic disclosure is a form of TAR, but things have come a long way in recent years. The starting point for an efficient TAR exercise is to whittle down the pool of documents for review as far as possible and as intelligently as possible. In addition to keyword searching and restricting by dates and custodians, this can be done by the use of technologies such as ‘concept clustering’ to break down the documents by topic in order to spot-check and (hopefully) eliminate large batches of irrelevant documents or prioritise the review of batches of relevant documents that have been identified.

Once the document set for review has been circumscribed, predictive coding can be used to review those documents. Documents are reviewed for relevance by a computer rather than a human, after a human has taught it how to copy his or her decisions on relevance in a particular case. A senior lawyer reviews a ‘seed set’ of documents and codes them as relevant or not. The results are then analysed by the technology, which breaks the word patterns down into smaller units. The algorithm then generates further batches of documents for the human to review in order to teach the computer. Through an iterative process of refinement, it can reach an acceptable level of review accuracy that can be applied to all of the documents in the pool. The end result should be that the computer identifies relevant documents in a manner that is far more efficient and scalable than a traditional (human) document review.

Further human sampling of the output is generally needed to ensure the statistical soundness of the exercise, and then the final results will usually need to be reviewed by humans in a litigation context for privilege and/or client confidentiality in irrelevant documents. The net result is that humans should need to review far fewer documents than previously.

Impact of TAR on litigation strategy

TAR can play a part in deciding whether to pursue or defend litigation before proceedings have even been issued. The possibility to perform a document review exercise more quickly and cheaply means that in certain situations it will be sensible to conduct an ‘early case analysis’ by assessing how the documentary evidence stacks up and feeding this into merits advice. If the proverbial ‘smoking gun’ is found, that may make it clear from the outset whether the case is likely to be won or lost.

In addition, there are structural changes afoot to the English court disclosure regime. In 2016 a disclosure working group chaired by Lady Justice Gloster was set up with an ambitious task to encourage a wholesale ‘change in culture’ by re-writing and modernising the rules on disclosure. In November 2017, the working group issued a draft CPR (that's the Civil Procedure Rules) practice direction intended ultimately to replace CPR PD 31, which is to be considered by the CPR Committee later this year. After that, a pilot scheme is planned to be run for two years in the Business and Property Courts of England and Wales. The proposed new regime has similarities with disclosure in international arbitration and, rather than having ‘standard disclosure’ as the default, requires parties to disclose the key known documents on which they rely at the same time as serving their statements of case in ‘basic disclosure’. The proposed regime then seeks to tailor the extent of ‘extended disclosure’ required (if any) to the nature of each particular case. The suggested practice direction embraces TAR, obliging parties to discuss and seek to agree the use of technology-assisted software and techniques; and permits the court to include provision in any disclosure order requiring the use of specified software tools.

While the proposed new regime offers an exciting opportunity to rein in electronic disclosure (which, due to the exponential growth of data and media in use, has, by common consensus, got out of hand), parties will be able to agree to opt out of basic disclosure. The ‘menu’ of extended disclosure options is broad enough to effectively replicate ‘standard disclosure’ if desired, but it remains to be seen how bold parties and the judiciary will be in using the new framework and technologies to limit the size of disclosure exercises and conduct them in a more sophisticated manner. If the proposals do result in a reduction in the disclosure burden (often the most expensive portion of the costs of proceedings), this could make litigation a more attractive option for parties and a more viable one where there are limited funds, for example in an insolvent estate. Given the potential difficulty for insolvent estates in satisfying security for costs applications with after-the-event insurance as a result of the Premier Motorauctions appeal at the end of last year (Premier Motorauctions v PwC & Lloyds [2017] EWCA Civ 1872), a reduction in the costs of litigation will be all the more welcome.

Alternative use of TAR in insolvency

TAR is being embraced in litigation because it offers the ability to locate relevant documents more quickly and cheaply than before, however the same technology can also be harnessed in an insolvency (or other investigation) situation. By way of example, when IPs are appointed to a company they may be faced with hostile management and little available help in establishing the workings of the business and whether there has been dissipation of assets. They may have either limited time or limited money (and often, both) to work this out in order to secure those assets. If predictive coding is employed to conduct a review of the company’s books and records to search for leads, it has the ability to rank the documents by relevance so that humans can begin reviewing the most relevant documents, hopefully finding what they need before the time or money runs out. The level of review accuracy of the computer can also be manually set, so that a less refined exercise can be conducted to return more that might be of relevance (to use one analogy, like a fishing net with small holes), or a more pinpointed search can be undertaken, which might miss certain documents but not return too many ‘false positives’ (like a fishing net with large holes).