Big Data Cloud

Zenith Actuarial and “evo-insight”: The platform that turns workbooks into calculation APIs

Redefining Data Management”, Martin Sher, COO/CTO

 

Zenith Actuarial is an innovative and agile consulting firm specialising in actuarial and technology services for the insurance and financial services sectors.  Leveraging its extensive experience and adaptability, Zenith navigates the complexities of these industries, delivering bespoke solutions tailored to each client’s unique circumstances and objectives.

For Zenith, technology plays a critical role in serving its clients, as much of the advice is based on models that need to scale, be well governed, and offer robust controls. Many of these models start life as client spreadsheet models or Python scripts, often referred to as “end-user computing” (EUC).

While EUCs are excellent starting points for prototyping and rapid analytics development, they present challenges in terms of scalability and governance.

CLIENT
Zenith Actuarial
Industry
Finance
Services
Cloud, Data
Technology
Azure Batch, Azure Function, Azure Datalake
To address these challenges, Zenith Actuarial partnered with Plain Concepts to develop evo-insight, an award-winning platform. evo-insight transforms workbooks into compute APIs hosted in Azure, integrating them into highly scalable, well-governed, and tightly controlled web applications running in Azure.
01

Challenges

We faced three main challenges: scalability, governance, and project design.
1. Scalability: The model often needed to be repeated tens of thousands of times, a requirement that Excel struggled to meet, especially with multi-scenario calculations. Overcoming this challenge necessitated in parallelizing the calculations and storing the results in a data file.
2. Governance: Ensuring robust governance in operational models was crucial due to market regulations. The platform needed to provide flexible workflows that incorporated approvals, documentation, and audits to ensure compliance.
3. Project Design: Plain Concepts had to ensure IT scalability to accommodate Zenith’s rapidly growing customer base. The infrastructure needed to be .NET compliant to run C# computations and support rapid scaling, as the computations were highly parallelizable. Additionally, it had to adjust flexibly to variations in usage, scaling up or down as needed, and maintain an easy-to-maintain architecture.
02

The Process

To implement the project and overcome the aforementioned challenges, we followed a five-step process:
1. Creation of initial Proofs of Concept (PoCs) to evaluate architecture performance, parallelization capabilities, cold boot times, and overall simplicity.
2. Kubernetes was discarded due to its operational complexity and the high level of technical expertise required.
3. The first PoC with Azure Container Apps completed the computing process, but configuration limitations in KEDA prevented the desired parallelization.
4. A second PoC with Azure Batch was tested, resulting in a significant leap in performance, meeting scalability and efficiency expectations.
5. Azure Datalake was integrated to store and handle the large output generated by the parallel execution of analytics.
Plain Concepts brought deep technical expertise in Azure cloud services to design and refine an architecture that fully met Zenith’s requirements. Continued collaboration is planned for the next phases of Zenith’s roadmap, which includes developing a richer set of extract-transform-load capabilities and supporting a broader set of EUC analytics.
03

Results

``Empowering companies to harness the latest in cloud computing and data management, seamlessly transitioning to innovation while maintaining full control over their core knowledge.`` Javier Echavarri, Delivery Director, Plain Concepts

• Performance optimisation with a simple and efficient infrastructure.
• Scalable architecture to expand Zenith's market offering.
• Reduced computing time from 24 hours to 20 minutes.
• Achieved a 50% reduction in computational operating costs.
• Platform customers can now produce their regulated economic scenarios in 30 minutes instead of 6 hours.
• Running thousands of simulations much faster, allowing rapid model adjustments as new information emerges.