Our team is composed of data scientists, data management specialists, eDiscovery experts, and senior leaders with decades of experience working with dispute and compliance technologies. Many members of our team have advanced degrees, including doctorates in machine learning and master’s degrees in statistics and econometrics and other data science fields.
What We Do
The use of artificial intelligence and machine learning in the legal industry can be seen in several examples. Legal research, contract review and creation, due diligence, compliance, and litigation all have use cases of AI and ML technology. Although many people still see the use of AI and ML as being at an early stage of technology adoption, we believe that these technologies will quickly become standards. We also believe that using these technologies will ultimately lead to courtroom battles over the quality, care, and ethics of AI/ML.
Machine learning uses algorithms to build mathematical models that are used for prediction. In our daily lives, we see many examples of AI/ML programs: when we shop online, when we are recommended TV shows and movies based on others who watched similar shows, and when our music app curates playlists. The methods used to create those recommendations are based on ML algorithms that improve as they acquire more data. For example, a collective 4.0 stars on a movie has more relevance when there are 5,000 recommendations versus 50. More data also allows for better statistical analytics, which in turns leads to better model improvements that help the machine continuously learn and improve its predictions.
Our team supports and augments existing litigation and discovery technologies. In litigation settings much of the work we do ends up in review technology software programs). We perform all the required eDiscovery services (processing, hosting, review support, and productions) and augment them with AI techniques such as similarity scoring of documents, handwritten document detection, enhanced keyword-search techniques, and non-keyword privilege document identification. These augmented techniques are added to review platforms so they are integrated with the technology people are already using.
BRG’s AI and ML practice also creates an analytics dashboard for use by litigation teams. The dashboard uses advanced AI and ML technologies to perform entity extraction (including related entities), heatmaps of custodian relationships and topics, timeline and sentiment analysis, and clustering.
In non-litigation settings, our professionals provide technology solutions for compliance and audits. We use a scientific approach to create predictive models that can identify anomalous transactions or events. Examples include finding:
- Suspicious transaction for further review
- Anomalous events in compliance processes
- Outlier communications between parties
We use the public cloud to perform many of our services. This provides us virtually unlimited computing power, security, and compliance with various privacy statutes.
Our Experience*
Our professionals’ experience includes recent and historical matters with voluminous data involving shared databases, multiple parties, and government agencies. Our experience involves multi-party discovery as well as productions to the Department of Justice and Federal Trade Commission. We have experience in antitrust, securities, manufacturing, food service, product liability, employment and health care matters.
Some key measurements of our success are:
- Demonstrated cost savings in document review
-
- We have built privilege models that predict the likelihood that a document is privileged. Using ML output, a team of attorneys reviewed 25,000 documents in ten weeks and produced 2.7 documents from that review. Our client benchmarked our work against the review cost of a previous antitrust matter and estimated cost savings of more than $3 million.
- We built an ML model using semantic clustering to group similar documents together for attorneys to review. The model returned a similarity score of indices of the documents, which substantially increased the efficiency of the review. The client estimated this efficiency saved more than 30 percent over the cost of a traditional document review.
- BRG uses a proprietary predictive coding model than can augment and even replace human review. This model was accepted by the IEEE International Conference on Big Data for presentation. It features ensemble learning features and can be provided as an academic article to the proposal reviewers.
- We have built ML models that predict the likelihood that a document has handwriting on it, the likelihood that a document is an organizational chart, and the likelihood that two or more people are working in an organizational structure that is not readily identifiable.
- Multiparty litigation
- Our team manages cases with dozens of clients represented by multiple firms in multidistrict litigation and class action actions. We provision applications and data to groups, sub-groups and experts based upon a case need or for an ML project such as labeling documents so algorithms can learn faster.
- We have built predictive ML models to calculate commercial damages and used advanced dimension reduction and variable selection techniques to identify the most relevant predictors. Such models were used successfully in some of the largest cartel cases in the past decade.
- Compliance
- We have used AI and ML technology to help companies evaluate the effectiveness of their compliance programs. Using anomaly detection tools that we have developed our team has:
- Helped financial institutions evaluate lookbacks on suspicious transactions
- Helped pharmaceutical companies monitor gifts and hospitality compliance by highlighting outlier transactions
- Predictive Analytics
- We have built settlement prediction models that provide a timeline and expected settlement value of a case.
- We have helped corporate legal departments predict the value of a case prior to determining where work should be performed (in-house or with outside counsel).
- We have helped counsel determine the likelihood that a witness or series of witnesses were coached prior to their testimony.
- *Includes professionals’ experience prior to joining BRG.
