Using computer assisted review in construction disputes

Out-Law Analysis | 16 Nov 2016 | 9:52 am | 2 min. read

FOCUS: Computer assisted review (CAR) of trial documents can save time and money, but should be used with care on construction matters where there are often lots of discrete, complex issues to consider.

CAR, or predictive coding, is an optional process that can be used as part of the document review stage in disclosure to filter out irrelevant documents from your wider pool of electronic documents. It effectively involves 'training' the computer system by reviewing a relatively small, but representative, sample, so that the system can work out what is relevant to the case and what isn't and can use its understanding to categorise the wider document population.

From a construction disputes perspective, CAR is specifically referred to in the e-disclosure protocol jointly published by the Technology and Construction Solicitors Association (TeCSA), the Technology and Construction Bar Association (TECBAR) and the Society of Construction in Law (SCL). Use of the e-disclosure protocol has been championed by the Technology and Construction Court (TCC), the division of the High Court which deals with the majority of construction disputes.

A practical experience

In a recent case, in preparation for disclosure, we carried out a targeted collection of electronic documents. We then used key word filtering to get the bundle down to about 1.5 million documents for review and decided on a two-stage process:

  • CAR to cull the obviously irrelevant documents;
  • manual second stage review, removing any irrelevant documents and flagging for issues and privilege.

The immediate difficulty we faced was the number of discrete items of claim. As a complex piece of litigation, the case involved around 20 discrete claims and 10 discrete counterclaims, amounting to 30 separate issues.

As we had come up with our list of key words on a claim by claim basis, we were able to use these same key words to produce bundles specific to each claim and counterclaim. We used a cross section from across each of these bundles to create a representative sample, and then started reviewing the batches of documents until the computer caught on and its findings became more accurate. After reviewing about 10,000 documents, we reached the point of diminishing returns.

The ultimate result was that the computer was able to cull about half of the electronic documents. A quality control check showed that the computer was correctly categorising the documents with a 98% accuracy level.

CAR: the pros and cons

There is a fairly even split between the pros and cons of incorporating CAR into your document review process. However, the technology is likely to become more and more popular, particularly in cases with large numbers of documents for review.


  • time and cost savings. CAR avoids the need to employ an army of paralegals, and allows you to get to a defensible position without someone having to look at every single document;
  • there is no set number of training rounds for the system, so you can keep going until you get it right;
  • it avoids the potential for human error to creep into the process, which is almost inevitable after a person has reviewed a large numbers of conceptually similar documents;
  • the computer learns from one or two individuals who have a real understanding of the issues in the case.


  • it's quite tricky to engineer a genuinely representative sample, so you need a technically capable e-disclosure provider on board;
  • it takes quite a while to train the system, so the cost and time savings are not as big as you might think;
  • depending on the complexity of your dispute, use of CAR potentially opens you up to another area for criticism from the opposing party.

David Greenwood is a construction disputes expert at Pinsent Masons, the law firm behind