Continuous Active Learning

Continuous Active Learning (CAL) is a modern form of Technology Assisted Review which can drastically reduce the time and cost associated with a large document review. With a CAL review, a random selection of documents are classified manually by an expert and used as a training set for an AI. The AI will then predict which of the remaining documents are most likely to be relevant. These are given to the review team to begin classifying. Based on each new document classified, the AI updates it’s model and reorders documents based on the new expected relevance. This process is repeated on a continuous basis throughout the review.

By presenting documents expected to be relevant first, CAL can help reviewers gain an understanding of the case more quickly than in a linear review.

Using our tools and experience IRS can provide both the platform for hosting a CAL review and can assist you in conducting it.

In practice, we run CAL reviews on Reveal and RelativityOne, whose Continuous Active Learning is well suited to large New Zealand discovery matters. Because CAL surfaces the most relevant documents first, it helps you meet your High Court Rules discovery obligations proportionately — reviewing far fewer documents to reach the same result, with a defensible, measurable stopping point.

Continuous Active Learning FAQs

How does CAL reduce review time and cost?

Because CAL continuously re-ranks documents by likely relevance, reviewers see the most important documents first and reach a defensible stopping point sooner — so fewer documents need to be reviewed to find the material ones, which lowers both time and cost.

Is CAL better than traditional TAR?

CAL avoids the ‘cold start’ and one-off training limitations of older TAR protocols by learning continuously from every reviewer decision. A 2014 peer-reviewed study found CAL achieved superior results compared with other machine-learning review methods.

Does CAL replace human reviewers?

No. CAL prioritises and orders documents, but experienced reviewers still make the relevance and privilege decisions that train the model. CAL makes their work faster and more focused, not automated.

Want to cut the time and cost of a large review with CAL? Contact us or call 0800 WITNESS (0800 948 637).