Bank statements are universal in financial services. Whether you're underwriting loans, investigating fraud, or conducting legal discovery, you need accurate transaction data. Extracting this data from documents at scale still presents hurdles.
Two companies recently tackled this challenge with Sensible's document automation platform, but their journeys to successful implementation were very different. One needed bulletproof accuracy for legal and forensic use cases. Another prioritized speed and cost efficiency for high-volume lending.
Company A: financial forensics in the UK
A UK-based financial services company—Company A— needed to extract data from bank statements for estate settlements, divorce proceedings, and anti-money laundering investigations. They use the data to serve law firms and accountancy firms.
Their previous extraction vendor could handle simple, clean statements. But its accuracy fell apart when faced with real-world complexity such as skewed scans, random page ordering, and multi-currency transactions.
The Challenge: Messy documents
The reality is that bank statement automation in legal discovery is messy. For the most part, you’re not receiving the statements directly from financial institutions, which would ensure clean, digital documents. Instead, you’re receiving scans from a nephew whose great aunt just died. Her bank statements are in a jumble in some desk drawers. The nephew snapped pictures of 50+ pages of bank statements in the order in which he found them (so it’s page 45, then page 1, then page 30). The nephew also used a camera without OCR or deskew functionality, so many of the pages are heavily tilted. Even aside from the digitalization problems, the statements they process are inherently complex. Transaction order is variable, with some banks listing transactions newest to oldest, others oldest to newest. Some statements span multiple years (e.g. Dec - Jan) without explicit year labels on individual transactions. There are multiple currencies given the company works with customers across Europe, and inconsistent negative value notations ("overdraft," "DR," etc.)
The Solution: Advanced logic and normalization
Their head of engineering asked Sensible to build a proof of concept around the two most difficult bank statement formats they’d encountered. If Sensible could accurately extract and normalize the data from those highly contrasting statements, they knew it could handle anything. The CEO noted that their initial caution evaporated once they saw the extracted data was accurate and could be easily tweaked. When they got back their data, their reaction was that the normalization was “magic.”
After the proof of concept, Company A took a hybrid approach: they built some templates themselves while relying on Sensible's managed services for complex formats.
Their implementation involved sophisticated data transformation and validation. For example:
- Detect when transactions cross a year boundary and append the correct year to each transaction. For example, if a December 2023 statement shows transactions from both December and January, the system automatically assigns the correct year to January transactions.
- Perform OCR and deskew on poor-quality scans, then correct OCR errors automatically (for example, commas often are misread as periods by OCR engines).
- Standardize negative values regardless of whether they’re notated with "overdraft," "DR," negative sign, or parentheses.
- Consistently sort transactions whether or not they appear in chronological, reverse-chronological order, or in misordered pages.
Sensible’s head of engineering actively participated in the Slack channel, exploring solutions for edge cases.
The Results: Handling the "impossible" documents
After implementation, the company can handle a greatly increased number of bank statements. Within 6 months they were automatically processing statements from about 80 UK banks. They plan to grow this number and switch all bank statement processing to Sensible. As their scale increases, they’re able to model out their increased costs using a per-page pricing model with Sensible (which they chose over a per-document pricing model for apples-to-apples comparison with their previous vendor). The validation capabilities—being able to check totals and balances within the template itself—was something their previous vendor simply couldn't do. The upshot is that their client law firms and accountancy firms receive consistent, validated data despite messy source documents.
Company B: Automating small business lending in Canada
A Canadian small business lending platform—let's call them Company B—faced a different but common problem: their existing document extraction vendor was too slow and too expensive for their rapidly scaling operations. As their loan volume grew, response times lagged and costs escalated.
The Challenge: Balancing speed and cost
Their chief product officer explained that they needed to process statements faster and at a fraction of the cost, but couldn’t sacrifice accuracy. Incorrect transaction data could mean approving bad loans or rejecting good ones. The good news was that their target bank statements were generally clean, digitally generated documents. The challenge was that they needed to process large volumes of bank statements from dozens of banks with one business day turnaround times. They also needed to maintain fraud detection capabilities for US banks and validate extracted data automatically to catch errors.
The Solution: Managed services with built-in validation
Rather than building their own extraction templates, Company B opted for Sensible's managed services. The Sensible team builds and maintains their bank statement configurations, delivering new templates in as little as 45 minutes.
Their data analyst noted that the managed service approach was a game-changer. They get new formats within a day, and have a direct Slack channel with the Sensible team, with no waiting and minimal back-and-forth.
The platform implemented a critical validation strategy using Sensible's logic capabilities. Every extracted statement undergoes an automatic reconciliation check:
Opening balance + Deposits - Withdrawals = Closing balance
If this equation doesn't balance, the document gets automatically flagged for human review. This simple validation catches OCR errors, missed transactions, and formatting issues before they impact underwriting decisions.
The Results: Faster and cheaper
After two years with Sensible, they’ve significantly reduced their costs, expanded their Canadian bank coverage to most of the banks in Canada, and they’re expanding their US coverage. They use a multi-region deployment to keep Canadian data in Canada for regulatory compliance, and US data in the US. For fraud detection, they make use of Sensible’s integration with a third-party fraud detection API.
Two customers, two strategies, one platform
These stories illustrate a key advantage of modern document automation: flexibility in implementation strategy.
The UK forensics firm (Company A) needed to handle chaos. With documents from non-technical users and high-stakes use cases, they required sophisticated normalization and validation. The hybrid self-serve model let them tackle routine formats, while handing over edge cases to Sensible’s team.
The Canadian lender (Company B) optimized for operational efficiency. With high volume and standardized needs, managed services provided the perfect balance of speed and cost. Their validation logic ensures accuracy without slowing processing.
Both approaches work because Sensible's platform supports both deterministic and LLM-based extraction methods that can handle real-world document variability, combined with powerful logic for validation and normalization.
Key takeaways for bank statement automation
If you're considering automating bank statement extraction, consider a few key points:
- Validation is non-negotiable: Whether it's reconciling balances or detecting year crossings, build validation into your extraction logic directly within the Sensible platform rather than catching errors downstream or having to build out all validations on your end.
- Plan for document chaos: Real-world statements arrive skewed, out of order, and in inconsistent formats. Your extraction solution must handle this reality.
- Consider data sovereignty: Multi-region deployment isn't just nice to have, it's often legally required for handling financial data across jurisdictions.
- Start with your hardest formats: Like Company A, test Sensible with your most difficult edge cases. If the Sensible can handle these, then routine documents will be straightforward.
Get started with bank statement automation
Whether you're processing statements for lending, legal discovery, or financial analysis, Sensible's team can help design an extraction pipeline for your specific needs.
Book a demo to discuss your loss run processing requirements, or explore our managed services to see how we can handle template creation and maintenance for you.
