SightCapture™ ICR: Why OCR Was Never Enough for Financial Documents
Beyond Optical Character Recognition: Document Understanding for Accounting
---
Your accounting software processed that invoice. It captured the amount, the date, the vendor name, and the VAT number. The serial number, the warranty expiry, the payment terms, the delivery reference, and the product code are in the same PDF — but they are not in your database.
Somewhere in the last three years of invoices, there is a €45,000 warranty claim you never filed because nobody captured the warranty terms on page three of the purchase invoice. OCR technology converted those terms to text. Nobody structured them, stored them, or made them queryable. When the equipment failed at month nineteen of a twenty-four-month warranty, your finance team paid the repair bill as an operating expense.
That gap is what SightCapture™ ICR closes — capturing everything on that invoice, not just the financial fields, but every field a human reader can see.
---
OCR Reads Characters. SightCapture™ Reads Documents.
Optical Character Recognition — OCR — converts images to text. That is all it does. Give it a scanned invoice and it will tell you the characters in the image. It will not tell you which characters constitute a serial number, which are payment conditions, and which are warranty terms. That distinction requires context, and OCR has none.
Standard accounting software uses OCR to pull 3 to 5 fields from each PDF invoice: amount, date, VAT number, vendor, and description. Those five fields satisfy the regulatory minimum required for a valid accounting entry. The other 35 fields — serial numbers, warranty periods, payment conditions, delivery references, product codes, early payment discount terms, cost center allocations, and contract clause references — are converted to searchable text in the archived PDF and never touched again.
SightCapture™ ICR — Intelligent Character Recognition — belongs to a different class of technology entirely. It reads documents the way a human accountant would: understanding that "12/03/2025" in the top right corner is the invoice date, that "90 jours fin de mois" in the payment section means payment is due 90 days after the end of the invoicing month (a common French payment term equivalent to "net 90, end of month"), and that "SN: 47821-FR" is a serial number that belongs to a specific product category. Every field, identified, structured, stored, and queryable.
French mid-market companies receive an average of 840 supplier invoices per month. Standard accounting software captures 3 to 5 fields from each. SightCapture™ ICR captures 15 to 40 fields per invoice. The gap — approximately 90 percent of each document's information value — is currently locked inside PDFs that nobody reads twice.
---
The Six Fields That Pay for the Deployment
Six specific field categories that standard OCR processing discards represent measurable annual financial value for a 200-employee company:
Serial numbers link purchased equipment to warranty terms and service history. Without them, warranty claims require a manual search through procurement email to find the original purchase confirmation. With them, the question "which assets expire warranty in the next 90 days?" takes 8 seconds.
Payment terms determine when invoices must be paid and whether early payment discounts are available. A company with €8 million in annual payables and 3 percent average early payment discount availability has €240,000 per year in available discounts — which are only capturable if the payment terms are in the database. Payment terms are on every invoice. OCR processes them as undifferentiated text.
Contract renewal dates captured from supplier agreements and service contracts prevent automatic renewals on unfavorable terms. Auto-renewal clauses are a standard feature of supplier contracts. Eight untracked auto-renewals at €15,000 average annual value each represent €120,000 in committed spend that nobody reviewed before the renewal date passed.
Duplicate invoice detection requires that invoice numbers, vendor identifiers, and amounts are all structured and cross-referenced. When only amounts and dates are captured, duplicate invoices from the same vendor with slightly different dates pass through processing unchallenged. Structured field extraction makes duplicate detection automated.
Price condition clauses capture the contractual pricing terms against which actual invoice amounts can be compared. When a supplier quietly raises unit prices by 8 percent, the increase is invisible if the accounting system captures total invoice amounts but not per-unit pricing and contracted rate references.
Delivery references resolve logistics disputes. When a supplier claims delivery was made and the buyer claims it was not, the delivery reference on the invoice — which maps to the carrier's tracking system — is the evidence. Without it, the dispute requires manual retrieval of shipping documentation.
For a 200-employee company, capturing these six categories produces estimated annual value of €170,000 to €435,000. That value is currently being discarded by systems that stop at five fields.
---
What a 350-Employee Company Found in Its Own PDFs
One French industrial company with 350 employees deployed SightCapture™ ICR as part of Stralevo in 2025. The accounting software had been processing their invoices for years. The invoices were filed. The system indicated they were complete.
In the first 90 days of SightCapture™ deployment, the system identified:
€67,000 in unclaimed warranty repairs on equipment purchased in the previous 36 months. Warranty terms had been on page three of every purchase invoice. OCR had converted them to text. Nobody had ever queried them because the accounting software had no warranty expiry field — because it had never captured warranty terms.
Twenty-three duplicate invoices totaling €31,000. The duplicates had passed through the standard approval process because the accounting system checked amounts and dates but not invoice numbers combined with vendor identifiers and line-item descriptions simultaneously. Structured extraction made the duplicates visible immediately.
Eight supplier contracts with auto-renewal clauses triggering within 60 days that nobody had flagged. The contracts had been processed by OCR and filed. The renewal dates had never been extracted. Three of the eight renewals were for services the company no longer needed.
Total first-quarter value identified: €98,000 — from data sitting inside PDFs that the accounting software had already marked as processed.
---
Why Accounting Software Vendors Built OCR-Only
Regulatory minimum drove the architecture. Tax authorities require amount, date, VAT number, and vendor identifier for a valid accounting entry. Everything beyond that is optional for compliance purposes — so vendors built for mandatory fields and stopped. In 1995, building full document comprehension would have required technology and engineering investment that had no obvious regulatory justification. Vendors built what regulators required.
By 2025, that logic is visibly broken.
EU e-invoicing standards are accelerating the change. EN 16931, the EU structured invoice standard, requires that electronic invoices carry structured data — not just PDF scans. French regulation requires all business-to-business invoices to be transmitted in structured electronic format by 2026 for large companies and by 2027 for SMEs. The regulatory direction is toward full structured data, not the five-field minimum of 1995.
Companies that deploy SightCapture™ ICR now are building the extraction infrastructure the regulation will require — capturing all fields from current documents during the transition period, rather than retrofitting extraction capabilities when compliance becomes mandatory for their entire supply chain.
---
Sovereignty in Document Processing
Cloud-based OCR services from major technology providers process document data through US-jurisdiction infrastructure. Financial documents sent for extraction — invoices, contracts, receipts — contain supplier pricing, commercial terms, payment conditions, and strategic relationship data. Under the CLOUD Act, US law enforcement can compel any US company to produce data processed through their infrastructure, regardless of where the customer is located.
Running on Stralevo's sovereign architecture, SightCapture™ ICR processes documents locally. Documents are processed on your infrastructure. No third-party API receives your financial documents. Every extraction result stays within your system, subject to your access controls, visible only in your audit trail.
Document processing sovereignty is not the headline capability of SightCapture™ ICR. It is the baseline. The headline capability is the 35 fields per invoice that your accounting software has been discarding.
---
What Five Years of Full Extraction Looks Like
Any company that has deployed SightCapture™ ICR for five years holds a structured data repository that replaces several systems built precisely because the accounting software never captured the relevant data: manual asset tracking spreadsheets (because serial numbers were never captured), warranty tracking systems (because warranty terms were never structured), supplier contract databases (because contract clauses were never extracted), and delivery reference logs (because logistics codes were never stored).
All of that data was always in the documents. Every invoice the accounting software processed contained it. SightCapture™ ICR captures it automatically, at ingestion, without changing the accounting team's workflow.
Five years of fully structured invoice data gives the CFO answers that competitors cannot answer. Which assets are under warranty right now? Which supplier contracts auto-renew in the next 90 days? Which suppliers have increased unit prices above contracted rates? Which invoices paid in the last three years could have qualified for early payment discounts that were missed? Each answer comes from data that was always available — in PDFs that the old system processed and considered done.
The question "is the production line equipment still under warranty?" should take 8 seconds. At most companies, it takes 45 minutes. That gap is not a resource problem. It is a data capture problem — and it has been solvable since the moment someone built document comprehension that goes beyond converting pixels to text.