Sogeti and HSB – deploying AI insurance models in Azure
Sogeti provides HSB a new predictive model, new architectures, and DevOps automation.
"HSB, a Munich Re company, partnered with Sogeti to launch our Enterprise Information Management Program that includes operationalizing AI/ML models, a transformative initiative that leverages Microsoft Azure Data Stack to enable fact-based real-time decision-making. During this journey, Sogeti leveraged its extensive Microsoft expertise to help us navigate the various technical and process features critical for moving to production. Their deep knowledge of Microsoft as well as their strong relationship with the Microsoft team ensured effective and efficient collaboration, as well as a successful launch."
Louis DiModugno, Chief Data Officer at HSB

About the Client
HSB provides specialty insurance coverage and services, both directly through independent agents and brokers and through reinsurance arrangements with over 200 multi-line insurance companies. Like all insurers, HSB uses a Loss Cost Model (LCM) to determine its insurance premiums to ensure its profitability.
Client challenge
HSB wanted to create an analytics model deployment solution that could be leveraged for Loss Cost model scoring to enable underwriters to do the costing in real time as well as batches, reducing the manual process. To achieve this, it sought a technology solution enabling the creation of a deployment framework in Microsoft Azure that would integrate internally with various cloud components. This would utilize the global API management of its parent company Munich-re for exposing services.
HSB turned to Sogeti, Microsoft’s Azure Expert Managed Service Provider for three times in a row. Leveraging the depth and breadth of experience we already had from implementing their enterprise information management platform using azure, we were now ready to help HSB meet the business objectives of creating a solution that:
- Could primarily be utilized by internal pricing tools
- Was robust enough for any kind of analytics model deployment such as EB Loss Cost and HSP/SL Loss Cost models (LCM)
- Provides a complete end-to-end and seamless analytics model solution in the Azure cloud environment
- Could be integrated with applications such as rating tools or any other front-end user interfaces
- Would improve quality, performance and stability of LCM 2.0
HSB needed to setup best practices to put their predictive models into operation. After the model is validated, the model is moved to production by implementing a scoring system where the model is applied to new data. Production environment was necessary for models built in R and Python. However, HSB needed the capabilities to deploy AI/ML models for both real-time scoring as well as batch scoring a portfolio.
We followed the three high-level workstreams: Model Request Orchestration, Analytic Model Services and Containerization, and Platform Build/Release Lifecycle.
Within this, HSB sought to:
- Create DevOps automation to deploy each cloud service into different environment
- Leverage ETL process to pre-capture scoring model data
- Leverage a full Azure technology stack or all-cloud approach: Azure Function App, Azure PaaS SQL DB, Azure APIM, Azure DevOps, Azure Kubernetes, Azure App Insight, Azure Log Analytics, & Azure Key Vault

The solution
Two new logical target state architectures were defined: one for loss risk and the other for loss cost.
In building these architectures, the team created specific architecture patterns, including the following:
- Internal applications accessing external APIs over internet (Outbound) – no transformation
- Internal application accessing internal (MR private) API (Inbound)
- External users/application accessing Munich Re public API over internet (Inbound)
- Publishing an API for internal or external consumption
- Internal application sending a file to an internal application (Transformation/no-transformation)
- Leverage the Publisher and Subscribe pattern to manage large amounts of the locations to be scored or pricing.
- Leverage Function Apps to build Realtime and Batch Processing of Data Science Model Data.
- Leverage Log Analytics which enable audit history, and monitoring on model requests and model outputs.
- Complete end to end, Azure DevOps platform that automates the process of updating versions of the model.
- Leverage the Global APIM, which built upon Azure API Management platform.
- Ensure compliance with Fortify scan tool for standardization and code quality.
- Ensure compliance with Vulnerability & Penetration SecOps Testing.
Sogeti Cloud Automation is a structured collection of scripts that DevOps teams can inject in their DevOps pipeline and enable on the cloud platform. Adopting a DevOps approach enabled by automation was a key component of the Platform Build/Release Lifecycle phase of this transformation. Cloud automation helped the DevOps teams to keep the focus on creating business value. Everything that was a recurrent activity in production was automated.
Key Client benefits
HSB can now expose the predictive models to both internal & external customers and deploy complex models in production easily with this framework.
The solution supports real-time transaction with a minimum of 5 batches/ day with capability to scale up, if business expands, to a batch up to 500K locations.
Automating the process resulted in fewer errors and quicker turn-around time (cut down to just half a day than many days earlier).
We are averaging 1000 to 2000 location per min in near real-time, which produce results within hours reducing turnaround time for results to just half a day for jobs which would take several days to complete.
Leveraging the Azure Cloud platforms with serverless, microservice, and service bus patterns has helped HSB achieve greater elasticity with scaling & performance for bigger and smaller batch counts, while helping maintain low costs and low run rate for HSB's IT budget.