AWS Cost Optimization Engine

About the Customer

isaga helps Indian mid-market companies improve business ROI from the cloud through two tightly integrated pillars:

• FinOps and cloud cost optimization: The team analyses existing AWS infrastructure to reduce waste, inefficiencies, and unnecessary spend.
• GenAI: isaga starts from business problems - not technology - identifying high-leverage use cases where GenAI can reliably deliver value and then designing and implementing those solutions on AWS.

Together, these capabilities deliver measurable ROI. isaga is vertically integrated - from problem identification to production.

Customer Challenge

Before building the AWS Cost Optimization Engine, isaga relied on manual processes to analyze AWS consumption reports for customers seeking cost optimization. Teams spent over 80 hours per month parsing PDF reports, with a single report taking 1 to 3 hours to analyze, which slowed down customer engagement and pre-sales cycles. AWS reports have roughly 1,800 permutations including regions, operating systems, engines, instance families, and tenancy options, and human reviewers missed around 20% of potential savings because they could not catch everything.

Standard regex-based parsing failed because it could not handle variations in report formats and struggled with complex billing structures like Savings Plans and Reserved Instance deductions, with multi-region and multi-engine deployments making it worse. Prospects expected rapid assessments, but long turnaround times made them disengage, and customers had no self-service option, depending on isaga consultants for even initial estimates.

This model limited isaga's ability to scale, increased the cost of each assessment, and weakened their competitive position in fast-moving evaluation cycles. If left unaddressed, these challenges would have continued to constrain growth and reduce customer satisfaction.

Partner Solution

To improve customer experience, isaga built an AI-powered, self-service platform - the AWS Cost Optimization Engine - using Amazon Textract and Amazon Bedrock with Claude models. Customers upload their AWS consumption reports and receive instant, service-level savings assessments, while isaga consultants focus on higher-value advisory work before and after implementation.

The solution runs as a fully synchronous, stateless web application on AWS Elastic Beanstalk. The process is immediate: customers upload invoices through a secure interface and get a breakdown of current spend and potential optimizations right away. This breakdown forms the basis for follow-up strategy calls where isaga FinOps specialists validate recommendations, design optimization roadmaps, and support customers through rollout and ongoing governance.

Amazon Textract extracts text from PDF consumption reports and preserves complex table structures. Textract was chosen because it handles AWS billing tables that standard OCR tools misread or flatten, processes consumption reports 2 to 3 times faster than alternatives, and supports both synchronous processing for small files and asynchronous workflows for larger documents using temporary Amazon S3 staging.

Amazon Bedrock with Claude Sonnet and Claude Opus interprets extracted data and converts it into structured JSON with intelligent line-item extraction. Larger documents go to Claude Sonnet for speed, while smaller or more complex reports use Claude Opus for maximum accuracy. Claude's reasoning capabilities extract key metadata accurately, including Region, Instance Type, Engine, Family, OS, Usage Hours, and Cost, handling AWS-specific terminology and nested billing constructs. isaga consultants review AI-generated outputs during follow-up conversations.

A deterministic pricing engine performs sub-millisecond lookups across more than 1,800 AWS pricing permutations, while a high-performance pricing cache calculates potential savings instantly. The architecture prioritizes privacy: processing happens in memory, temporary S3 objects used for asynchronous Textract runs are deleted immediately after use, and no customer data is stored. AWS Elastic Beanstalk manages auto-scaling, health checks, and deployments, letting isaga's team focus on FinOps logic, responsible AI prompt design, and white-glove customer support across pre- and post-implementation phases.

Results and Benefits

The AI-powered AWS Cost Optimization Engine delivered measurable improvements across time, cost, and accuracy. Time to insight dropped from 1 to 3 hours per report to under 5 minutes, plus it removed dependency on humans to share these insights with customers. isaga achieved roughly a 95% reduction in manual analysis time. The effective cost per analysis fell from roughly $31 in manual labor to about $1.50 in AWS infrastructure spend, enabling far more assessments without adding headcount and leading to annual savings of $17,236.

Accuracy of savings detection improved from around 80% with manual review to approximately 98%. The platform achieved a 99.9% JSON parse success rate and 99.95% availability. isaga identified 15% to 25% more cost-saving opportunities than traditional methods.

The customer experience shifted dramatically - previously, customers waited 2 days for recommendations, but now they upload a report and receive a clear, service-level breakdown within minutes. This transformation represents the most critical aspect of the solution's impact, enabling faster decision-making and improved customer satisfaction.