Reading bank statements at scale — accurately.
A European financial-services company was drowning in manual bank-statement review. CONE RED built the automation engine that now extracts and classifies their transactions — measured, in production.
Measured in a production reporting period
The challenge
Bank statements arrive in dozens of layouts — every bank formats dates, running balances, and debit/credit columns differently. Reviewing them by hand is slow, and a single mis-read of a transaction’s direction (money in vs. money out) corrupts every downstream reconciliation. The client needed statements turned into clean, structured, correctly-signed transaction data without a human checking every line.
What we built
CONE RED engineered a document-processing pipeline that normalizes any supported bank’s statement into a common schema, extracts each transaction, and classifies its direction and category. The system was built to be measurable: every run logs volumes, processing time, and a transaction-level accuracy check, so quality is a number the client can see rather than a claim they have to trust.
The measured result
Across a production reporting period the engine processed hundreds of statements and 4,716 transactions spanning 21 banks, holding a 0.24% transaction-direction error rate with 84% of statements returned in under 30 seconds. Those are operational figures from live runs — not a projection.
A note on these numbers. The metrics above are measured operational results from one engagement’s production reporting period. They are specific to that client’s data and volumes; results vary by engagement, document quality, and scope. AI systems are probabilistic, so we report measured accuracy honestly rather than promising a fixed figure. Client identity withheld by request.
Have a document-processing bottleneck?
If your team is re-keying data from PDFs, statements, or forms, we can scope an automation that reports its own accuracy.
Talk to us