Financial KPI Extraction: Enhancing Efficiency in Financial Document Analysis
Introduction: An Unavoidable Economic Reality for CTOs
The rise of AI is transforming data exploitation, but its execution remains costly, especially on non-optimized cloud infrastructures.
In a context where GPU resources (such as Nvidia H100, H200) are billed hourly and rarely utilized at 100%, CTOs must find solutions to optimize costs without compromising performance.
Ryax addresses these challenges by efficiently structuring AI workflows to optimize resource usage and reduce expenses. Its platform enables Data Scientists to create, deploy, and manage their analyses simply, freeing them from the constraints of data engineering and infrastructure limitations.
A Use Case: Extraction of Financial KPIs
Consider a concrete example: extracting Key Performance Indicators (KPIs) from financial reports.
In the financial sector, manually extracting key performance indicators (KPIs) from annual reports represents a major operational hurdle. Financial institutions process thousands of reports each year, each containing critical indicators buried within complex document structures. This process traditionally requires considerable human effort, is prone to errors, and struggles to scale with increasing document volumes.
Many companies have begun automating this process and face a new challenge: the resource consumption required for this processing.
This process is broken down as follows:
• File import
• Data extraction
• KPI calculation
• Storage in a database
The following diagram represents the application’s workflow:
Principle of Analysis
In this use case, KPI extraction is performed from 1,180 financial reports, executed daily in 32 independent workflows.
This process relies on:
• Standard CPU tasks, executed in less than a minute.
• Intensive GPU operations, requiring AI models for document analysis and computer vision.
One of the major findings of this study was the inefficiency of current provisioning models.
An Example of Inefficiency
In a traditional cloud environment, companies are forced to reserve entire GPU instances, even if their workload only partially utilizes them.
As a result:
• A cost of €34.41 per execution on AWS (p5.48xlarge).
• A cost of €30.41 per execution on GCP (a3-highgpu-8g).
• Largely suboptimal consumption, with GPU resources used only 67% of the total workflow time.
The challenge is clear: how to reduce resource overallocation while maintaining efficient execution of AI models?
Ryax’s Proposal: Advanced Orchestration for Multi-Level Optimization
Ryax has developed a multi-step approach to optimize AI applications on the cloud, adapting to the specific constraints of the models executed.
1. Decomposition of Workflows via Containerization
One of the first optimizations is to fragment AI pipelines into independent actions, executed in dedicated containers. This allows for:
• Dynamic resource allocation, with each task receiving only what it needs.
• A 34% cost reduction, as CPU and GPU instances are used more intelligently.
With this approach, companies no longer pay for oversized monolithic instances but only for the resources actually consumed.
2. Automatic Resource Optimization (CPU/GPU/Memory)
Once workflows are containerized, the next step is to optimize resource usage through Ryax IntelliScale technology. Leveraging machine learning techniques, Ryax partitions physical resources for allocation tailored to the needs of the containers.
With this approach:
• Workloads are distributed over fractions of GPUs, thus avoiding underutilization of resources.
• Execution costs decrease by an additional 26% without compromising performance.
• The total number of GPU instances can be reduced, improving utilization density and reducing cloud expenses.
3. Multi-Cloud Allocation and Dynamic Orchestration
Beyond simple resource allocation, Ryax offers advanced orchestration of AI workloads, allowing for:
• Selection of the optimal cloud provider based on costs and performance.
• Distribution of the computational load based on real-time available CPU/GPU capacities.
• Utilization of energy consumption metrics to favor more sustainable infrastructures.
This dynamic allocation model ensures an optimal balance between cost, performance, and availability, while adapting to workload variations.
Results: Over 50% Savings with Minimal Impact on Performance
By combining these three approaches—containerization, GPU partitioning, and multi-cloud orchestration—we achieved a total cost reduction of 51.6%, with a minimal increase of 7.1% in execution time.
This represents nearly €6,000 in annual savings for a single use case. Applied on a larger scale, the financial impact becomes substantial.
Why Adopt Ryax Today?
Companies leveraging artificial intelligence face increasing pressure to reduce operational costs while maintaining high performance. Ryax offers a pragmatic and proven solution to achieve this balance.
Concrete benefits for your organization:
• Multi-Level Optimization: A holistic approach combining containerization, GPU partitioning, and dynamic orchestration.
• Massive Cost Reduction: Up to 50% savings on your cloud expenses by eliminating resource wastage.
• Flexibility and Scalability: A scalable model capable of adapting to real-time needs and workload variations.
• Transparent Implementation: Rapid integration with your existing infrastructures without complex workflow overhauls.
With Ryax, you regain control over your AI infrastructure costs while ensuring optimal performance and smooth execution of your models.
Read about other RYAX use cases
RYAX tackles new use cases every day.
Tell us about your projects.