In the recent decision of Triumph v Primus, shortly before his appointment to the Court of Appeal, Mr Justice Coulson ordered, on application by the Defendants, that the Claimants carry out a manual review of a statistical sample of 25% of 220,000 documents (that should have previously been reviewed) within 3 weeks. This is a useful indication of how the TCC is adapting and using modern technology to deal with disclosure.
The underlying action concerned a US$65 million claim for alleged breaches of warranty arising from the sale of the defendants' aerospace business to the claimants via a share capital purchase.
At the directions hearing, the parties had agreed to standard disclosure of documents pursuant to part 31.6 of the CPR. This requires parties to search for and disclose (a) the documents on which he relies; and (b) the documents which (i) adversely affect his own case; (ii) adversely affect another party’s case; or (iii) support another party’s case.
A process for the review of electronic documents held on the Claimants' shared drive was agreed. This required the data to be searched by keywords agreed between the parties, with the Claimants to then manually review all of the responsive documents and give disclosure of those documents required under standard disclosure. The parties had not agreed to or discussed any use of computer assisted review ("CAR"), which is understood here to be computer learning and predictive prioritisation of documents.
The amount of data on the Claimants' shared drive was voluminous – containing over 860,000 identified folders and over 3TB of data. The Claimants sought to restrict their searches to around one-third of the folders based on assumptions as to which were most likely to contain relevant documents. The agreed keywords applied to the shared drive brought back around 450,000 responsive documents which were required to be reviewed manually pursuant to the approach agreed between the parties and endorsed by the Court.
The Claimants carried out a manual review of a substantial volume of documents (around 200,000) through a combination of a manual searches and a CAR process. This resulted in 12,476 documents being disclosed by list. Three weeks later, the Claimants served a first supplemental list, disclosing a further 4,163 documents. By this point, around 230,000 of the 450,000 documents had been reviewed by manual searches aided by CAR.
The remaining 220,000 documents were assessed by the CAR system (based on the training and learning it had gleaned from the Claimants' review of the 230,000 documents) which predicted that only 0.38% of the remaining 220,000 documents would be relevant and the Claimants subsequently carried out a sampling exercise of 1% of the remaining documents. Based on this, the Claimants' unilaterally decided that it was not reasonable or proportionate for the remaining documents to be searched further, no doubt on the basis that the Claimants had already incurred the significant cost of reviewing 230,000 documents.
The Claimants subsequently produced almost 3,000 further documents by way of 10 further supplemental lists of documents, following pressure from the Defendants. Around two thirds of these documents responded to the agreed key so should have been reviewed as part of the agreed process following CMC. The Defendants sought an order that: (i) the full list of 860,000 folders on the shared drive be disclosed; and (ii) the Claimants to undertake a manual review of the remaining 220,000 documents.
Coulson J remarked that "Disclosure has been something of a running sore in this case". Whilst the Claimants had acted unilaterally and without the agreement of the Defendants in using CAR and ceasing the review, Coulson J nevertheless, declined to grant the orders sought. This was on the basis that: (i) the process adopted by the Claimants for the folders had been clearly set out in the Claimants' original list of documents and the Defendants had delayed by well over a year in raising any objection; and (ii) the method adopted by the Claimants had been sensible and proportionate on the facts. As to the review itself, Coulson J considered that the use of CAR and the subsequent sampling exercise had not been transparent or independently verifiable and it was unsatisfactory that the Defendants had not been allowed to provide any input into the process, although he accepted that the use of CAR was acceptable.
Coulson J ordered that the parties to agree a methodology by which a sample of 25% of the remaining documents (i.e. 55,000) be manually searched and reviewed by the Claimants within three weeks.
This decision demonstrates that the failure by the Claimants was not their decision to restrict the folders to be searched. It was not their decision to utilise CAR in the review methodology. Indeed, it is clear that the use of these technologies (be that (i) predictive coding, (ii) technology aided review such as analytics or (iii) computer aided review such as prioritisation) is now approved by the Court.
Further, it was not the Claimants' decision to stop the review when the CAR system suggested that there would be diminishing returns of continuing. What the judge was critical of was the Claimants': (i) failure to consult with the Defendants as to the proposed process to be adopted, (ii) deviation from an agreed approach and; (iii) failure to ensure that the Claimants' approach was transparent and independently verifiable. For this reason, the Court ordered further searches but restricted these based on proportionality.
This decision highlights the risk a party takes if it fails to carry out early analysis of its data in advance of agreeing directions at CMC and/or fails to co-operate and consult with the other parties around disclosure.