When we started building our system, one of our goals was to design the platform to normalize unstructured contractual data using proprietary taxonomy libraries we had built with the intent of making it amenable to dashboarding and reporting. As a result, we had a huge corpus of structured data originating from contracts of every shape, size, and language. This has proved to be a key differentiator for us because the same structured data was later used to train the machine learning models, which power our contract analytics engine today.

Our focus firmly remained on getting the basics right when we were developing machine learning models for the text extraction engine, which has been through scores of iterations over the past several years. We synthesized the engine using a variety of Bayesian network, named entity recognition and n-gram language models along with syntactic rules. Soon enough, we built prototypes and started testing some of the earlier AI-driven features in collaboration with select customers.

Fast forward to 2019-20, and we have now filed several patents and rolled out a stand-alone AI solution for contract management that helps businesses gain full visibility into and control of risks and obligations. What started out as a simple idea to automate document extraction has evolved into something much more. Today, our customers can use SirionAE to process up to a million documents a day. We have transformed our AI engine from a backend efficiency enabler into a customer-facing, self-service solution. Some of our clients are already training Sirion’s AI engine themselves to extract specific fields and clauses, which means that the solution keeps evolving well after it has been deployed, learning about a new structure or clause language with every contract it ingests.