Technology Assisted review in large dispute case management: Australian and international developments

​By Andrew Stephenson (Partner, Corrs Chambers Westgarth), Lindsay Hogan (Senior Associate) & Jaclyn Smith (Associate)

With the size – and complexity – of large commercial disputes rapidly increasing, using technology to assist in their case management is becoming more and more important.

As noted by the Arcadis Global Construction Disputes Report 2016, arbitrations and mega disputes in Australia are on the rise.

Disputes like these, and the factual circumstances underlying them, typically involve vast numbers of documents that are, or may be, relevant to the issues requiring resolution. As such, the use of technology assisted review (TAR) to assist with their case management is becoming increasingly common.

In this article, we take a look at some recent case law developments – both in Australia and internationally – that offer some insight into how the Courts view TAR and may use Special Referees to assist with its implementation in the case management of large disputes.


In McConnell Dowell Constructors (Aust) Pty Ltd v Santam Ltd & Ors (No 1) [2016] VSC 734 (‘McConnell'), the Supreme Court of Victoria confirmed that the Court will enlist a Special Referee to examine the implementation of TAR to assist with the case management of large disputes. 

The proceeding concerned claims arising from the design and construction of the QCLNG pipeline in Queensland, and related policies of insurance that McConnell Dowell (one of the parties to a JV who constructed the pipeline) took out with Santam,
QBE and Liberty.

The principal procedural concern, from a case management perspective, was the significant volume of potentially relevant documents. A related issue was how they might be managed in a manner consistent with the principles of proportionality, and also section 9 of the Civil Procedure Act 2010 (Vic) which requires the Court to further the overarching purpose of this act when making orders or giving directions.[1]

Initially, McConnell Dowell had identified some 4 million documents generated by the contract in question and an associated arbitration. It had reduced this number to 1.4 million documents that were relevant to the proceeding. Justice Vickery pointed to the enormous time and cost of potentially reviewing such a volume of documentation according to a traditional discovery process, even after de-duplication and other automated processes had cleaned up the data.[2]

This appreciation of time and cost resulted in his Honour appointing Mr Anthony Nolan QC as a Special Referee to answer questions, and produce a report, on the appropriate management of the discovery process.[3]

His Honour observed that the parties had worked together during this review process, which required discussions about the use of TAR and establishing protocols and procedures around this procedure.[4]

While construction disputes are not the only kind of disputes that can result in large volumes of documentation for review, the increasing use of technology on construction projects means that electronic data is growing exponentially (which may well be exacerbated once BIM becomes standard on construction projects).

In McConnell, Vickery J advised that he would make orders for the use of predictive coding to be used in discovery, and made orders concerning the appropriate protocol for this to occur. Given his Honour’s reference of the issues to a Special Referee, this case offers a great insight into the way expert assistance might be used to assist the Court to manage large scale discovery processes.

His Honour further foreshadowed the recent issue of a new Practice Note by the Supreme Court of Victoria which provides guidance on the use of TAR.[5]

This protocol provides that TAR can be ordered for use in discovery, with or without the consent of the parties, and that parties may be directed to agree to a protocol for the TAR process. Among other things, the Practice Note requires the parties to consider:

  1. the system to be implemented, including the learning protocol (including whether it is a continuous active or passive learning protocol);
  2. the method of analysis to be adopted (including the statistical measures implemented for quality assurance purposes); and
  3. arrangements for the clawback of privileged and confidential documents.


While still a relatively novel development in Australia, TAR has been utilised in litigation in the UK, US and Ireland.

In Pyrrho Investments Ltd v MWB Property Ltd [2016] EWHC 256 (Ch) (‘Pyrrho’), the UK High Court endorsed the use of TAR in circumstances where its use was determined to be ‘proportionate.’

The dispute in question generated approximately 17.6 million documents. While this was narrowed to 3.1 million after de-duplication, the number of documents requiring review still posed considerable issues to the discovery process. Like the Civil Procedure Act 2010 (Vic), the English Civil Procedure Rules require parties to make ‘a reasonable search for documents,’ and for such discovery to be ‘proportionate.’[6] Reasonableness, in this context, is determined by factors including:

  • the number of documents;
  • the nature and complexity of the proceedings;
  • the cost of disclosing; and
  • providing inspection of the documents.

In Pyrrho, Master Matthews explained predictive coding (an expression used interchangeably with TAR) as a software analysis.

The common process identified by the Master can be summarised as follows:[7]

  • the parties agree on a coding protocol;
  • the parties define the criteria of inclusion and exclusion;
  • the included documents are ‘cleaned up’ (e.g. by de-duplication);
  • a representative sample is manually reviewed by a single senior lawyer (who is best placed to apply the criteria of relevance consistently), so that the software can be ‘trained’ to review documents a certain way;
  • the software processes all of the documents;
  • the software’s categorisation is tested by further sampling and review (often three or four times);
  • the results of this further testing are fed back into the software for continued learning.

Master Matthew explained that this predictive coding process costs considerably less in comparison with manual review.

In Irish Bank Resolution Corporation Ltd v Quinn [2015] IEHC 175 (‘Quinn’), the documents initially identified as potentially relevant to the proceeding numbered 1.7 million (which was reduced to 680,809 following de-duplication).

There was a dispute between the parties as to whether TAR was sufficiently accurate to ensure that the discovery given was adequate. While noting that there was no express concept of proportionality in Ireland, the Irish ‘courts increasingly refer to [proportionality] as a relevant factor in assessing whether the necessity requirement has been satisfied on the facts of a particular case.’ [8]  

His Honour’s findings in Quinn regarding the accuracy of predictive coding suggest that it should be adopted in all large cases:

the evidence establishes, that in discovery of large data sets, [TAR] using predictive coding is at least as accurate as, and, probably more accurate than, the manual or linear method in identifying relevant documents.[9]   

The use of this process has also become increasingly common in the US. Vickery J referenced a decision of the Southern District of New York in which TAR was accepted as black letter law in the US.[10]


Given the cost savings it promises, the use of TAR in large disputes is likely to increase. However, there will be continuing concerns about the accuracy of such a system. 

Like other computer models, TAR relies upon an algorithm, which is proprietary confidential information of the owner of the software, who is unlikely to be willing to disclose that algorithm, making it difficult to establish that it is sufficiently reliable. This could perhaps be overcome if there was an appropriately independent certifying authority which issued certificates as to its capacity.  

Similarly, any protocol would require the party receiving documents identified as relevant by TAR to satisfy itself about the accuracy of the manual coding done by the party making discovery. This requires sufficient transparency about what has been done, which may lead to very time consuming arguments about the manner in which the coding is done. 

Depending on how those disputes are resolved, much of the value of TAR could be lost.

[1] Civil Procedure Act 2010 (Vic) ss 7 and 7.

[2] [5].

[3] [8] – [9].

[4] [13].

[5] [30]; Technology in Civil Litigation Practice Note SC Gen 5.

[6] Practice Direction 31B, specifically provisions 20 and 21, provide for what constitutes a ‘reasonable search’ for e-discovery.

[7] [17] – [24].

[8] [48].

[9] [66].

[10] [25]; Rio Tinto v Vale 14 Civ. 3042 (RMP) (AJP) (2 March 2015).