UK-based legal AI company RAVN Systems has formally launched a new application to review documents for legal professional privilege (LPP) using its cognitive engine.
The new service will be branded the ‘LPP Robot’ and will be offered alongside RAVN’s other AI capabilities. The system reduces the time taken in such reviews by over 80% and greatly reduces the level of manpower needed to conduct such a task. Accuracy levels are also improved.
This is an important step as the ability to spot LPP documents is of critical importance to lawyers, especially where investigations of large company data sets are involved or when part of a wider discovery process due to litigation.
It matters because LPP material represents communications between a lawyer and their client, which could also include privileged material between an inhouse lawyer and an executive in their company. Other parties involved in litigation would not normally be allowed to see such communications, hence LPP requires significant attention. Yet, it remains a very slow process to review this material and law firms have used relatively antiquated methods until now.
LPP is also of major importance to regulatory investigations where company files may have been turned over to a Government department or agency for analysis, or where a request has been received to produce documentary evidence.
As with other ‘cognitive processes’, such as due diligence or large-scale lease review, this work can be laborious and errors can be made by large teams of tired junior lawyers and paralegals.
Peter Wallqvist, CEO at RAVN, said: ‘We are excited to add the LPP Robot to our portfolio of [AI] applications, completely transforming the review process, resulting in an overall improved service to end clients.’
While it may seem that LPP reviews are similar in methodology to other AI-enabled review work, such as with property leases, the subject matter is in fact quite different.
RAVN explained to Artificial Lawyer why this AI challenge was different:
‘When RAVN started looking at the most common LPP review process, which involves keyword searches agreed between both parties in order to identify which material should undergo LPP review, the inherent inefficiencies of such an approach became apparent. The default behaviour was to err on the side of caution with the use of very far reaching search criteria.
Over the past few years, and with the exponential growth of data, it has become apparent that this approach is no longer sustainable. With an average yield of confirmed LPP material of 15%, with extremes as low as 1%, LPP reviewers therefore spend most of their time determining non-sensitive material.
Our initial approach was cautious: Could we help by picking the lower hanging fruit and narrow down the scope of the review? This has traditionally been the intention of standard Technology Assisted Review (TAR) e-discovery tools, but after careful investigation it was determined that a new approach was needed.
Traditional TAR methods are purely data-centric, relying on being able to infer similarities between documents. The basic assumption is that similar documents have similar relevancy.
Whilst this is an approach that can help in identifying clusters of similar documents and has been used successfully to help on disclosure obligations, it was determined too generic to yield the accuracy required.
Therefore, our approach had to remove any assumptions. [Our approach] allows the LPP Robot to model both the data and the reviewer behaviour with the use of supervised iterative machine learning models.
It’s a dynamic process that follows the initial review process and learns the idiosyncrasies of the on-going LPP review with a very high degree of accuracy. Furthermore, the longer the review takes place the more accurate the LPP Robot becomes.
Finally, our approach provides a consistent performance compared to manual processes. Its Confidence Measure allows the end users to meet very conservative adoption of this technology whilst at the same time significantly reducing the time and cost of the LPP review.’