An AI agent reviewing every upload.

CostCTRL is a SaaS platform that turns operational and financial data into profitability insights its customers can prove. Every implementation begins the same way: the customer uploads CSV files with their data.

• A duplicated ID, a misread date or a mistyped column can quietly corrupt an analysis or stall an onboarding
• Basic format warnings existed; the costly failures were the edge cases: stray special characters, numbers read as text, missing values
• Errors surfaced later as calculations that did not add up, and tracing the cause back to one cell could burn hours of expert time

In a five-day Forward PoC, Flipkick built an AI agent on Amazon Bedrock that watches CostCTRL’s upload bucket. When a file lands in Amazon S3, the agent works out its structure and column types on its own (AWS Glue and Amazon Athena, querying files in place) and hunts for the silent errors: duplicated IDs, misread date and number formats, numbers stored as text, stray special characters. Because CostCTRL’s data models vary from customer to customer, the agent adapts to each file it receives instead of relying on hard-coded rules that would need constant maintenance.

The uploader gets clear feedback by email, with suggested fixes. No new tools to learn and no change to how customers work: same CSVs, same upload, but now every file gets an expert review in minutes. Everything runs in CostCTRL’s own AWS account (data never leaves their environment), the agent suggests fixes but never modifies source files, and costs scale with actual uploads. The PoC did what a PoC should: prove it. The agent evolved into production and is now part of CostCTRL’s live workflow, at no cost to CostCTRL for the PoC, backed by AWS funding.