Between: (1) George Allen Cowan (2) George Allen Cowan, on Behalf of Equis Special L.P. (Previously known as Equis Asia Fund Special L.P.) Plaintiffs v (1) Equis Special L.P. (Previously known as Equis Asia Fund Special L.P.), Acting by Its General Partner Equis Special GP (2) Equis Special GP (Previously known as Equis Asia Fund Special GP), in Its Capacity as General Partner of Equis Special L.P. (3) David Charles Russell (4) Adam Bernhard Ballin (5) Lance Michael Comes (6) Joseph Thomas Carmody (7) Rajpal Singh Chaudhary (8) Tony Gibson Defendants
| Jurisdiction | Cayman Islands |
| Judge | Justice Raj Parker |
| Judgment Date | 20 September 2022 |
| Court | Grand Court (Cayman Islands) |
| Docket Number | CAUSE NO: FSD 22 OF 2018 (RPJ) |
The Hon. Justice Raj Parker
CAUSE NO: FSD 22 OF 2018 (RPJ)
IN THE GRAND COURT OF THE CAYMAN ISLANDS
FINANCIAL SERVICES DIVISION
Discovery-directions-use of technology to assist review-methodology-keyword searches—GCR O.24—GCR O.1 Overriding Objective.
Paul McGrath QC, instructed by Walkers (Cayman) LLP, on behalf of the Plaintiffs
Ms Annaliese Day QC instructed by Maples and Calder (Cayman) LLP, on behalf of the Third, Fourth, Fifth, Seventh and Eighth Defendants
Mr Erik Bodden of Conyers Dill & Pearman on behalf of the Sixth Defendant
This is the judgment following the hearing of a summons dated 21 October 2021.
The Plaintiffs (Ps) apply for directions in respect of discovery. It is an application made as a result of the developments in technology assisted document reviews. Traditional manual methods are said by Ps to be inadequate for this case and are also said to involve disproportionate cost.
The parties agree that some form of technology-assisted process should be adopted but differ on what that should involve.
The essential issue for determination concerns how technology should be applied to the process of identification of relevant documents in this case.
Both parties have served evidence consisting of numerous affidavits and exhibits in respect of the Application 1.
In essence, Ps argue that the technology should be applied across an ‘unmanipulated’ data pool.
D1-D5 and D7-D8 (the “Maples Ds”), say the data pool should be manually reviewed by use of keyword searches prior to the implementation of the technology.
D6 has taken a neutral position with his attorney attending on a ‘watching brief.’ The discovery protocol arising from this decision will be binding on him.
The dispute concerns the management and operation of a group of related private equity funds known as “Equis”. The focus of the funds was on renewable energy projects in Asia and the Far East.
P1, Mr Cowan, is an engineer with significant experience in the Asian energy and power sectors. Mr Cowan was employed by the Equis Group in 2011 in connection with the establishment of the first Equis fund and was a foundation partner (along with D3-D7) of D1.
D1 is an Exempted Limited Partnership which was formed in the Cayman Islands under the Exempted Limited Partnership Act in October 2011. At all material times, D1 acted by and through its General Partner, D2 (“Special GP”). D2 is sued in that capacity. D8, Tony Gibson, was General Counsel to the Equis Group.
In overview, Mr Cowan brings claims in his own capacity and derivatively on behalf of P2 (the General Partner).
In the derivative claims, Mr Cowan alleges that monies to which P2 was entitled by way of investment and/or performance fees in respect of a number of the Equis funds were wrongfully diverted from P2 pursuant to an unlawful means conspiracy to which D3-D8 were party.
This was allegedly done by an assignment from P2 to a parallel vehicle, which Mr Cowan alleges caused loss of approximately US$90 million to P2. The unlawful means alleged involved breaches of a series of contractual and fiduciary duties owed variously by Ds to P2 and to Mr Cowan himself, as well as other wrongdoing.
In his personal claims, Mr Cowan complains of the wrongful diversion by D3-D8 of monies to which he was entitled by reason of his position as foundation partner and investor in the funds. This includes monies which would have been paid to him but for the wrongful diversion of monies from P2 and other distributions to which he was entitled from P2 but which he says D3-D8 wrongfully caused to be withheld from him.
The procedural history is somewhat convoluted and it is not necessary to set it out here. There have been multiple iterations of Ps claims which has caused the Maples D's significant work in relation to discovery and this application has further put discovery on hold 2. The main cause for the delay in the proceedings is that Ps sought to introduce a significant claim (the Japan Solar trust claim) and were ultimately unsuccessful.
Paul McGrath QC appeared for Ps. He argued Ps proposal logically recognises that the technology-driven approach rests on searching for concepts within documents and not simply keywords. This he says broadens the scope of the search and ensures a more accurate sorting of documents into the categories of relevant/non-relevant. Mr McGrath QC put the point as follows:
“… the introduction is really this, my Lord, that at the heart of our proposals is a recognition that the methods of conducting discovery have moved on in recent years, and that the software that's being developed to enable that discovery to take place has become increasingly more focussed and sophisticated. And that software removes much of the guesswork that's been previously involved in methodologies that have replaced manual review of all documents, and that includes the keyword search.”
He says it is widely recognised that a keyword search method, by its very nature, is an overly—rigid approach and likely to exclude documents dealing with similar concepts on the basis that they happen to employ different terminology. He argues that excluding such documents would materially impact the accuracy of the training for the predictive model.
Applying the technology to ‘unmanipulated data’, by contrast, enables the predictive model to receive the data it requires for effective training. There is accordingly no need, on Ps' proposal, for any remedial steps to overcome initial shortcomings from the use of keyword searches.
He goes on to argue that P's proposal would involve the consistent use of advanced techniques throughout the process, from the identification of a computer-generated mixed “seed set” to use for manual review and to teach the review algorithm, to the automated identification of further
documents of likely relevance. This, he says, is the most reliable approach, because it ensures that the model can learn effectively, through an iterative process by which documents are selected in an unbiased, unmanipulated fashion, and the algorithm is continuously refined by the results of manual review. It is also ultimately the most efficient and least expensive approachPs approach is in summary steps:
-
a) Compile an algorithm-generated mixed seed set 3 of 3,000 documents drawn from the Total Document Population 4, and conduct a manual review of that seed set.
-
b) Use the CAL 5 algorithm to review the total document population, on the basis of the output from the manual review of the seed set, assigning a relevancy score to the remaining documents in the Total Document Population.
-
c) In tandem with the CAL review, conduct a manual review of those documents in the total document population which the algorithm has ranked highest for relevance, creating a ‘feedback loop’ by which the CAL algorithm continues its learning by the manual identification of (ir)relevant documents, and thus refines its identification of other likely relevant documents which are prioritised for review.
-
d) When this process reaches a point where the relevant documents have been sufficiently sifted from the irrelevant, carry out an elusion test 6 by identifying a sample of
unreviewed documents for manual review. If that reveals abnormalities, the review continues. When it shows no abnormalities, the review process is complete.
Mr McGrath QC argued that there were four essential reasons why Ps' approach is to be preferred:
-
i) Generation of seed set: the best way to set the model off on the right track is to generate a mixed computer-generated seed set. The use of keywords to generate a seed set is an unreliable approach which would exclude large numbers of relevant documents, making for an inherently defective model from the outset.
-
ii) Document types in the seed set: Ps' approach would also involve a diverse set of documents. The Maples Ds' approach of using only parent emails, by contrast, will again artificially skew the model and miss relevant documents.
-
iii) Elusion test: the Maples Ds' proposal that these problems can be cured retrospectively is both unsatisfactory in principle and wrong in practice.
-
iv) Cost: the Maples Ds' suggestion that Ps' approach would be lengthier and more costly is unfounded. On the contrary, the Maples Ds' approach would produce a false economy at the start of the process which would lead to a longer and more costly process overall.
Ms Annaliese Day QC appeared for the Maples D's. She argued that P's approach was unorthodox, unfocussed and wasteful. She submitted that the P's proposed methodology would impose an enormous and disproportionate burden on the Maples Ds. It would require them to conduct an extensive trawl through irrelevant documents in the hope that something of relevance might emerge that would not be caught by a more conventional set of searches. It should therefore not be ordered even if there was a legal basis to do so (which she argued there was not).
The Maples Ds suggested approach would first involve the application of keywords to reduce the overall document population from 3.1 million to 90,000. The...
To continue reading
Request your trial