Predictive Review combines advanced machine learning technology with attorney review. Servient actively learns from the document decisions made by the legal team during legal review and separates the relevant documents from the irrelevant material. Predictive Review produces meaningful results after the review of only a modest number of documents. As the legal team reviews more documents, Servient continues to learn and further refines the automated document decisions.

Servient allows the legal team to immediately analyze the likely responsive documents at an early stage of the case. No longer must the trial team's case analysis wait for completion of formal review; it can now be run in parallel.

Servient covers both the Search & Culling and Document Review phases of the e-discovery process. As a result of the unified platform, the learning derived from Predictive Review can be applied back to the entire litigation hold data set to find additional relevant documents that have not been assigned for review.

Servient provides powerful new tools to balance the need for transparency in the selection of search criteria with the need to avoid the assignment of large volumes of irrelevant documents to expensive review. The legal team can negotiate narrowly tailored and more meaningful search criteria, yet supplement the search with the intelligence derived from the ongoing review.

In essence, Servient improves the recall of e-discovery retrieval and significantly increases the precision. Servient improves the quality of the e-discovery while providing huge costs savings.

Servient cuts the cost of e-discovery by more than half in most cases. Because Servient actively learns from the decisions made during legal review and can automatically separate the relevant files from irrelevant files, Servient provides new options to cost control.

With Predictive Review, the legal team can avoid the expensive, wasteful review of irrelevant documents. Predictive Review identifies relevant documents with a much higher accuracy rate than what is achieved with simple keyword searching and linear review; indeed, studies are finding that the technology exceeds the accuracy of subjective manual review (which accounts for over 80% of the total cost of e-discovery). Integrated statistical validation of the document decisions provides a new level of defensibility not seen with keywords and "brute force" linear review.

Predictive Review also enables new prioritized review work-flows. With Servient, the review manager can assign documents to different levels of reviewers based upon a document's probability of relevance. For example, documents with a high probability of relevance can be assigned to the trial team reducing the need for multiple level review; neutral documents can be assigned to contract review teams; and likely non-responsive documents can be assigned to lower cost resources for first pass manual validation. Assigning document review tasks to the appropriate cost-level resources creates significant cost savings.

The legal team can apply a new level of quality assurance to the review process because Servient's active learning technology is tightly integrated into the Servient E-Discovery Platform. Because Predictive Review is based upon statistical modeling of various document features, the decisions made by the technology are generally more consistent than the decisions made during subjective manual review.

Servient tracks the consistency of the document decisions made during the ongoing manual review with the automated document decisions made with Servient's active learning technology. The analysis allows the legal team to:

• Measure the accuracy of each reviewer's subjective document decisions as compared to the other member of the review team; the review manager can flag reviewers that have consistency rates that deviate from the average team consistency rates and evaluate the actual performance of the identified reviewer
• Automatically detect inconsistencies in review decisions that are being applied to similar documents; the review manager can identify confusion within the review team over the factual basis for the review decisions and improve the quality of the legal review

Servient also enables the legal team to apply the knowledge gained through Predictive Review to locate relevant documents that were missed in the standard search process. Servient increases the quality of of the culling process by supplementing the review set with documents that have a high probability of relevance but do no otherwise meet the standard search criteria set by the parties to the litigation.

Servient provides the legal team with the statistical verification necessary to prove the defensibility of the search protocol.

Servient automates the validation of the active learning output. Servient applies k-Fold Cross Validation of the learning model at various points during the process to measure the reliability of the document decisions.

Servient's intelligent selection of documents to assign to review incorporates statistical sampling of document decisions at different probability measures. As the review exhausts the documents that are predicted to be responsive, Servient's active learning technology continues to sample non-responsive documents to provide a defensible validation of the process. The combination of accepted machine learning validation techniques, such a k-Fold Cross Validation, and statistical sampling provides a reliable measurement of the accuracy of the process.

Given that the accuracy can be scientifically tested through the application of accepted statistical theory, the question of the defensibility of the process really comes down to an acceptance of the measured accuracy and the method used to derive the measurement. Servient's statistical validation process exceeds the current level of quality assurance utilized in the standard keyword search and linear review approaches.

Servient's analytics technology identifies various document relationships that are used in the active learning process. Relationships such as near duplicates, email threads and conceptual clusters not only provide insight into the data, they also streamline the review process.

Servient factors in the knowledge of conceptual cluster membership, email threads and partial duplication within the data set to determine the best documents to train the learning model. For example, Servient selects the "top of the thread" of an email chain and all unique attachments associated with the threaded communication to assign to review. This provides better exemplars for the software to understand the important document features and the attorney's relevance decision.

The Servient E-Discovery Platform implements an intuitive work-flow to leverage the identified document relationships. Compare the differences in near duplicates, rather than reviewing them as separate documents. Prioritize the review of "top of the thread" to avoid the unnecessary review of earlier email in the thread. Visualize any search result within conceptual clusters to gain a better understand of the data.

All of the analytics are performed upon ingestion into the Servient E-Discovery Platform. There is no third-party service to buy; instead the analytics are fully integrated into work flow.

Servient manages all data collected pursuant to a litigation hold. From forensic images to email to social networking feeds, Servient automatically ingests all collected data and prepares the files to be analyzed, culled and reviewed. Servient alleviates the need for traditional e-discovery processing because all base tasks (metadata extraction; duplicate ID; flatten containers etc.) are performed as the collected files are imported into the platform.

Our cloud delivery reduces the IT burden and costs. Servient's secure, private cloud allows legal departments to access the advanced technology resources they need without incurring internal costs for software, hardware and specialized staffing. Through the cloud model, companies can pay for the technology solution based upon their actual use and needs while ensuring that they are leveraging the "best of breed" technology to control e-discovery costs.