Spraoi's Barrel Platform: Enabling Carriers to Cost-effectively Scale Their Data Science Capabilities

Karan Mishra Co-founder, Spraoi

Carriers consider data to be the most important asset, and driver of both growth (identification of new markets, development of more targeted products/services) and optimization (efficient routing of service and claims, identification of fraud) opportunities. Beginning as a provider of SaaS machine learning solutions, we experienced an issue that most insurance carriers have with their own data science capabilities: the ability to efficiently and cost-effectively scale. Insurance carriers must ingest and process missing and erroneous (e.g. fat fingered) data. To run continuously learning machine learning models in this environment requires integrated development operations (DevOps), data engineering and data science teams. Until now.

Applying the lessons learned from over three years of developing and delivering machine learning solutions, we have built a machine learning platform in Barrel that puts it in a category of one through integration of three core concepts: ETL provision – extract, transform, cleanse and provision data for model building; model building – integration of model building and experimentation environment; pipeline tooling – orchestrate data pipelines to monitor data flows. While there are certainly several options in the marketplace that can be integrated to provide similar functionality, Barrel brings it together in a single, seamless platform.

Spraoi has built its Barrel platform to address the various predicaments of insurers as it provides an infrastructure to create self-describing datasets with built-in error handling. Barrel addresses the impediments to the delivery of machine learning models for insurance carriers, thereby making it truly scalable. Barrel leverages insurance specific, reusable rules, schemas, and processes to speed the machine learning model development process and virtually eliminate DevOps and data engineering from it.

Barrel offers a simple way to assemble workflows using reusable operations. Operations, once defined, can be used against any self-describing dataset making workflow simple to configure and use. It can provision dashboards to specific users and pre-built operations in workflows to keep the dashboards up to date. Further, Barrel combines the experiment infrastructure with the production workflows enabling the data science workflow to span the entire data supply chain.

We aren’t stopping there. We are continuing to advance our platform in an effort to support insurance carrier data science teams, actively driving three additional capabilities to continue to make the development and maintenance of machine learning models even more efficient and effective: machine learning powered schema inference –for dataset ingestion, we are introducing embedded machine learning to create self-describing datasets with no manual intervention; business user workflows – integrating human workflows that can be combined with system workflows to enable business users to be part of the ML process and more tightly integrate models outputs and business outcomes; record level lineage tracking – every record that goes through Barrel will be tracked as will the lineage of how it traveled through the system. While we currently do this at a dataset (file) level, we will now trace lineage at a record level and surface it in the user experience.

Spraoi has learned a great deal about efficiently scaling production data science teams in three years. Leverage our learning. Reach out to us and schedule a demonstration of our Barrel platform.