Shortly before the investment company filed their subsidiary for bankruptcy, the assets were transferred to another company, leaving creditors potentially facing multimillion dollar losses. In the following trial a traditional e-Discovery company would play a key role in the investigation to uncover foul play leading up to the bankruptcy. In the aftermath of the trial, the accusers law firm, would carry out a benchmark of the traditional e-Discovery’s results vs Anzyz’ state-of-the-art AI-system. The findings would turn out to be extraordinary.
Language is enormously diverse and nuanced and can be particularly complex for a machine to understand across industry verticals. The exact word order can carry critical information, raising the need for high-precision technology that captures critical nuances. Consequently, standard machine learning approaches struggle immensely as they largely rely on learning representations of individual words.
Anzyz CCL™, on the contrary, incorporates a novel algorithm-based approach that employs unsupervised machine learning algorithms to analyze large corpora of narratives to automatically generate custom-made language models comprising words and phrases of which meanings and relative meanings are also learnt. As such, issues related to misspellings, compound words, and lexical variants are greatly diminished.
The dataset, consisting of thousands of emails and documents, was automatically modelled by the Anzyz’ CCL™ algorithm. Once the AI-model had been created, the lawyers were immediately able to query the knowledge base by introducing voluntary terms into a concept-based search system, the user-interface being the AnzyzLegal product. Finally, the system implements a precision layer of deterministic rules for fine-grained control. Besides allowing the user to tag narrative text with similar accuracy as traditional expert systems, this layer also helps with interpretability and to correct errors reported by end users.
Anzyz AI-system gave insights previously not detected by the traditional e-Discovery system. The AI-model was deployed in a matter of hours, compared to otherwise manual labeling processes stretching for several days. At the same time the user-friendliness was reported to be high, with human-in-the-loop capabilities further enhancing user-experience, accuracy, and transparency of results where the underlying interpretations can to be traced back to the original raw data.