In a recent decision in the UK High Court, dealing with scope of collection and predictive coding, the Court confirmed the importance of a defensible and transparent process when it comes to data identification, data collection and disclosure review. The Court also highlighted the need to use senior reviewers when completing a predictive coding project.
In Triumph Controls UK Ltd & Anor v Primus International Holding Co & Ors  EWHC 176 (TCC) (07 February 2018) the Defendants sought orders from the Court that:
- A list of 860,000 folders and file paths which had been identified by the Claimants be provided to the Defendants so that they could see whether there were any folders or file paths which had not yet been, but should be, searched.
- The Claimants undertake a manual review of the balance of 220,000 documents (out of a total of 450,000) which had been identified as potentially disclosable following the keyword search but which, other than a very limited sampling exercise, had not been further searched.
In outlining the approach to be taken, the Claimants' Electronic Documents Questionnaire (EDQ) provided that following keyword searches all responsive documents would be manually reviewed. The EDQ made no mention of Predictive Coding. The parties agreed on the keywords including a number of refinements which subsequently returned around 450,000 responsive documents.
Based on a review of these documents, the Claimant provided two sets of disclosure.
After this disclosure, it became apparent that the Claimant had used Computer Assisted Review (CAR) to prioritise the review workflow and that after reviewing the first 230,000 documents had made a decision not to review all of the remaining 220,000 documents. In explaining this decision, the Claimants' solicitors put forward the claim that having reviewed a 1% sample of the remaining documents, their CAR workflow predicted that only 0.38% of the remaining documents would be relevant. Based on these figures, they decided that it was disproportionate to review a further 220,000 documents to return a potential set of 836 further relevant documents.
This approach was never discussed, let alone agreed, with the Defendants.
In his decision, the Hon. Justice Coulson ruled that as the Claimant had outlined the approach taken in identifying the folders and file paths to be searched when providing the original disclosure list and that he was satisfied with the steps taken by the Claimant and that the approach was "sensible and proportionate". This decision can be seen to be partly based on evidence that a further review of the folders only revealed a potential two further file paths that may have produced potentially relevant material. Justice Coulson did, however, comment that it would have been preferable if the approach taken had been discussed with the Defendant prior to this point. We can perhaps assume that if more potentially relevant folders had been found, Mr Justice Coulson's decision may have been different.
In regard to the approach taken towards the predictive coding workflow however, Mr Justice Coulson took the stance that it was "unsatisfactory" that the Claimant had not done "what they said they would do in the EDQ", which was a manual review of all 450,000 documents and that they had at no time provided "relevant details as to how CAR was set up or how it operated".
Mr Justice Coulson also considered that further disclosure after the first 2 tranches raised concerns about the Claimants' disclosure process, calling the predicted 0.38% relevance projection provided by the Claimant into question. In looking at the percentage of documents disclosed from the 450,000 documents responsive to keyword searches, Mr Justice Coulson felt that this was, in his experience, "very modest". Taking into account the low percentage arrived at by the sampling exercise, which could not be said to be "reliable", Mr Justice Coulson ruled that it was "reasonable and proportionate" to order "some form of manual search" and that based on the facts of the case, he thought this would tend towards a "document-heavy trial".
Mr Justice Coulson raised further concerns with the number of people and the seniority of those involved in the CAR process, with the Claimant apparently using "perhaps" ten paralegals and four associates, stating that it was also "not apparent that there was any overseeing senior lawyer and certainly not one undertaking [a principal role]".
Given that "both the CAR exercise and the sampling exercise that it produced cannot be described as transparent and cannot be said to be independently verifiable" and the doubts cast upon the accuracy of the results provided by the Claimants, Mr Justice Coulson ruled that it was "reasonable and proportionate" to order "some form of manual search" and that the parties should agree on methodology to sample 25% of the remaining 220,000 documents.
Whilst proportionality is a recurring theme in the disclosure space, the decision in Triumph Controls also reinforces the importance of tracking decisions made at every stage of the disclosure process, from initial identification of potentially relevant data through to the final review of documents.
If you are in doubt as to what should be tracked, seeking advice from an expert would be the sensible first step to take.