Insurance
Risk and capital modelling platform: the models too big for a desktop, now routine
- Client
- Global insurance and risk firm (name withheld)
- Sector
- Insurance
- Scale
- £10bn+ revenue
We worked with a world-leading insurance and risk firm with annual revenue over £12 billion, operating across highly regulated markets. The business depends on running large, complex actuarial models to price risk and hold capital. They had developed a cloud-hosted risk and capital modelling platform that allowed actuaries to flexibly write and run their own models.
Technologies
- Data Isolation
- Cloud Hosting
- Event-Driven Systems
- Scalability
The hard limit
Whilst the platform was flexible, a major disadvantage in the design was that the models would only run locally on the actuaries' own laptops. There was a hard limit to the size of the datasets that could be used in the models. This reduced the capabilities of the system and the size of the models that could be executed.
There were further issues. Model executions could take hours to complete, and simply closing a browser tab would cause executions to stop mid-calculation.
Scaling up
To enable computations across larger datasets, the system was going to require offloading the model executions onto dedicated cloud servers that could be scaled up accordingly. This would allow models to be run over millions of datapoints, instead of just tens of thousands. An added benefit would be that models would run asynchronously in the background, meaning that executions would continue regardless of whether the user had their laptop running. The users would be able to trigger a model and come back to the result later.
Bringing in the experts
Whilst the client was aware at a high level what needed to be done to improve their system, they lacked the in-house expertise and resources to see this vision to fruition.
We were brought in to work alongside the client to deliver a working solution. DigiLab is experienced in designing and building event-driven, scalable, distributed systems. We have built these types of systems with numerous other clients in financial services and so our skills were a close fit for the project.
The solution
Working closely with technical leadership, we aligned on an architectural design whereby we would:
- Run the model executions on scaled-out compute resources in Azure cloud hosting.
- Store client data on segregated databases to keep each client's data fully isolated.
- Enable actuaries to choose whether to execute models on their laptops or offload the executions onto the cloud-hosted compute resources. This would enable local executions of smaller models.
- Audit every time a model was run for full system traceability.
- Deliver a fully automated test suite that would quickly catch bugs and regressions.
- Fully document the system so that the in-house development team could continue supporting and extending our delivered solution.
- Train and support the development team on the delivered solution.
Outcome
Actuaries can now run the models that matter most, at any size, reliably, instead of being held back by the limits of their own computers. The platform scales with demand, each client's data meets the isolation a regulated insurer must guarantee, and the firm's own team was left equipped to keep building on it. We delivered a capability the business could not have built in-house, and the models that once maxed out at tens of thousands of datapoints now run over millions.
