Bank’s Process Transformation Delivers 90% Cost Reduction and 52x Productivity Gains with AWS-powered MLOps and GenAI
Financial spreading process greatly enhanced with innovative platform for machine learning operations (MLOps) and generative artificial intelligence (GenAI) built on AWS.
Summary
The corporate division of a leading Asia Pacific (APAC) bank was looking to modernize a business-critical financial data analysis process known as financial spreading. The goal was to improve process speed, scale, and cost-efficiency without compromising performance. With ML and GenAI identified as potential enablers, the bank engaged Amdocs Cloud Studio to harness relevant tools on AWS. The project resulted in successful deployment of an end-to-end MLOps and GenAI platform that is set to revolutionize financial spreading at the bank.
- >90% reduction in operational expenditure (OpEx) costs.
- >90% uplift in credit assessment accuracy (especially for net new clients).
- 52x productivity improvement (process time reduced from hours to seconds).
- Governance ensures process compliance and ethical use of GenAI.
Challenge
Resource-Intensive Financial Process Urgently Required Modernization
Globally, financial spreading is a major pain point for corporate banks. It involves the extraction and transfer of data from structured and unstructured third-party sources such as annual reports and cash flow statements into a bank’s internal financial frameworks. Labor-intensive and error prone, it creates bottlenecks in credit decisioning which directly impact revenue opportunities and risk management. The process is facing heightened regulatory scrutiny too, with new requirements continuing to emerge.
A leading APAC bank’s corporate division determined that its financial spreading process for credit risk assessment was in urgent need of modernization. Under the bank’s existing set-up, specialized accountants typically receive more than two years’ dedicated training before they conduct financial spreading analyses. The process for inputting, reviewing, and interpretating complex financial data is time-consuming and demands a high level of skill. Even so, outputs are not always consistent; different accountants may reach different conclusions from the same data source. Improving the efficiency, accuracy, and cost profile of financial spreading was imperative.
Solution
Advanced Cloud Tools Harness GenAI and ML to Automate Financial Spreading
Amdocs Cloud Studio was engaged to help modernize financial spreading for the bank using the power of AWS’ tools for automation, ML, GenAI, and MLOps. We devised a granular, multiphase approach to tackle the complexity of automating and improving this business-critical process.
First, we established data readiness to lay the essential foundation for an advanced solution leveraging ML and GenAI capabilities. This involved robust data strategy development and ongoing engagement with stakeholders. We also defined what the data products would look like, confirming that the end goal was ‘augmented intelligence’ whereby AI and ML outputs support end users.
Initial data analysis was rapidly followed by two proof of concept (PoC) executions, first for a non-production environment, then in production. These implementations yielded insights which guided the development of a comprehensive, centralized MLOps and GenAI platform to automate the bank’s financial spreading.
“We have achieved a lot and are now well positioned to solve a hard problem for our business (financial spreading automation) using cutting-edge tech (LLMs), and in the process position us to scale what we have done and how we have done it.”
How the Platform Works
The solution combines several AWS tools: SageMaker for MLOps, Bedrock for serverless GenAI capabilities and frameworks, S3 for data storage, and Textract for document processing. Together, these tools were used to build a cohesive, automated MLOps platform with an integrated GenAI application.
The end-to-end MLOps solution encompasses a wide range of modules and tasks. Key areas that benefit from MLOps include data compliance and governance, data ingestion, feature engineering, and experiment tracking. These modules are orchestrated into a unified workflow, with performance and analytics insights available for individual steps and phases, and across the overall business process.
GenAI is used for language-based tasks such as document evaluation and text extraction. For example, based on established financial document requirements and constraints, the model assesses whether new input documents align with the parameters of historical documents. Suitable documents then proceed to the next stage where relevant content is identified and extracted.
When productionizing data science, the focus is generally on deployment, automation and scalability rather than output quality and integrity. Yet subjectivity is an industry-wide problem for financial spreading, with tangible impact on the bottom line. To address this, we introduced strict constraints for data contracts, data quality, and data integrity, supported by a wide range of data engineering pipelines. For example, some pipelines use AWS Textract to extract data, enrich it via LLMs, and then perform quantitative checks to verify accuracy. Each data artefact is subject to lifecycle management during every iteration of the end-to-end process.
These measures ensure both input and output data are of a consistently high standard by applying the exact same logic and decision-making criteria to each financial spread. Data for new and existing customers alike is now subject to wholly objective analysis and interpretation.
Onboarding new data science use cases is also streamlined within the MLOps and GenAI-embedded environment, which is rigorously designed for governance, quality, and financial control. Since the centralized platform seamlessly scales to support new banking and data use cases, development costs are amortized over time and use cases, minimizing short-term impacts on profitability.
Ethical and Responsible Practices
In addition to technical best practice, the solution incorporates strong data and AI governance principles. This includes working only with datasets that comply with relevant markets’ regulatory and banking standards. There is a strong focus on embedding explainable and responsible AI (XAI and RAI) to ensure transparency and ethical use. For instance, if outputs are inconsistent, accountants can now pinpoint when, where, and why they differ.
Working in a stepwise manner meant we earned trust and buy-in from the bank’s leadership team and other stakeholders as activity progressed. The approach also reduced risk and meant potential obstacles were identified then resolved on an iterative basis. This was a major factor in the solution’s successful transition from PoC to deployment in a production environment.
Outcomes
Financial Spreading is Quicker, Cheaper, and Highly Accurate
The MLOps and GenAI platform significantly outperforms the bank’s financial spreading benchmarks across speed, cost, and consistency.
Empirical studies on the original, manual process reveal that a senior accountant can prepare three to five financial spreads per day, resulting in a high cost per spread. There is also an average discrepancy of around 21% when the same spread request is assigned to multiple accountants. For net new clients, this inconsistency increases exponentially.
Using the MLOps and GenAI platform, more than 1,300 financial spreads from the bank’s historic, compliance-approved datasets were completed in less than five days. This represents an efficiency level 52x higher than the most senior in-house team members.
Furthermore, the associated cost per spread was less than 10% that of manual processing. The platform also achieved overall accuracy of more than 90% when working with new, previously unseen datasets. For known clients, accuracy exceeded 91%. In short, the average number of errors was reduced by more than 50% compared to the manual process.
ML model deployment is renowned for high rates of failure, but this solution bucks that trend. The bank’s corporate division is now poised to benefit from a scalable, rapid, and cost-effective solution for financial spreading, giving the bank a major competitive advantage. AWS’ own ML and GenAI experts have described the project as ‘innovative’ and ‘revolutionary’.
