100% Document Comprehension: Capturing Serial Numbers, Warranties, and Contract Terms From Invoices
The Invoice Nobody Reads Twice
Right now, somewhere in your finance system, there is an invoice for the server running your accounting software. That invoice contains the server's serial number, its warranty expiry date, and the terms for extending support coverage after the initial period.
Your accounting system captured the amount, the date, and the vendor name. The other 35 fields are locked in a PDF nobody has opened since the month it arrived.
When the server needs maintenance and the question is "is this still under warranty?", the process begins: search email for the original purchase confirmation, find the invoice attachment, open the PDF, locate the serial number on page 2, separately search for the warranty terms in the small print. That takes 45 minutes if you find the document quickly. It takes a phone call to the vendor if you do not.
This is the standard outcome when invoice processing captures the regulatory minimum and ignores everything else. And it is the outcome for 85 to 90% of every invoice field in every finance system operating today.
---
What Accounting Software Was Built to Do
Your accounting software was built to satisfy tax authorities. Everything those authorities do not require — warranty terms, serial numbers, delivery references, contract clauses — gets filed in a PDF and forgotten. That is not a data problem. It is a design choice that predates the question by 50 years.
Standard accounting software captures 3 to 5 fields per invoice: date, total amount, VAT amount, vendor name, and a description field that is often truncated or generalized. A typical supplier invoice contains 15 to 40 distinct data points. Invoice number. Line-item descriptions with quantities and unit prices. Tax code per line item. Serial numbers for equipment. Warranty terms. Delivery address and delivery reference. Purchase order reference. Payment terms — net days, early payment discount thresholds, penalty clauses for late payment. Contract reference. Bank account details. Small-print conditions.
Five fields out of 40 is roughly 10 to 15% capture. The other 85 to 90% drives real operational decisions: warranty coverage, payment optimization, contract compliance, delivery verification. All of it dies in the PDF.
The counterintuitive point: this is not a retrieval problem. Making PDF search faster does not solve it. A fast search of an invoice PDF still requires a human to read the document and locate the serial number. Full-field extraction stores the serial number as a discrete, queryable data point the moment the invoice arrives — so the question "Is this equipment under warranty?" takes seconds regardless of how old the invoice is. Speed of retrieval is irrelevant when the data was never captured.
---
The Specific Costs of the Gap
A manufacturing company discovered that 23% of their equipment had expired warranties — because serial numbers had never been extracted from purchase invoices and linked to a warranty tracking system. The data was in every purchase invoice. Nobody had built the extraction layer. The maintenance calls were billed at full cost. The warranty coverage was never claimed.
One software firm avoided a €40,000 overpayment when Stralevo extracted an early-payment discount clause from a supplier invoice that the accounts payable team had never seen — the clause was on line 3 of the small print, never captured by their accounting software. The term was straightforward: 2% discount on the invoice total for payment within 10 days. The AP team processed invoices within 30 days because that was the standard payment run. That particular supplier issued four invoices per year at an average of €125,000 each. The missed discount was running at €10,000 per year. Four years in, no one had noticed.
For French companies, the regulatory risk of missing line-item detail is concrete: a distribution company resolved a TVA — France's sales tax — dispute because Stralevo had captured the line-item service descriptions that determined the applicable tax rate. Their accounting software had summarized those descriptions as "services rendered" in a single field — standard practice. The DGFiP auditor required the detailed line-item breakdown to apply the correct rate to each service category. The original invoice had the breakdown. The accounting system did not. The company with full-field extraction had the answer in seconds.
---
What SightCapture Actually Reads
Stralevo's SightCapture is not optical character recognition with better marketing. It reads documents the way a trained accountant would: checking the vendor name against known aliases, reading every line item including the fine print, noting payment terms and comparing them to the contract on file, capturing serial numbers and warranty dates from wherever they appear in the document layout.
SightCapture handles 50-plus document formats: standard PDF, scanned image, photographed receipt from a mobile camera, Word document attachment, Excel invoice, EDI format. For each, it applies comprehension rather than positional matching — identifying fields by context rather than location. A serial number in a table, in a header block, or in a free-text paragraph is identified as a serial number because of what it is, not because it always appears in the same cell position.
A supplier invoice in German with a French small-print section yields every field in a unified, queryable format. Multi-language document processing is a routine case, not an edge case, in any company operating across European markets.
The extraction result for each invoice is not a set of text strings — it is a structured set of entities with types: vendor entity, product entity, serial number entity, warranty term entity. These entities are linked in the Knowledge Graph and queryable as first-class objects. "Show me every piece of equipment with a warranty expiring in the next 90 days" crosses every extracted serial number against every extracted warranty term, across all purchase invoices in the archive.
---
The Data Debt That Accumulates
Organizations accumulate invoice data debt with every document processed at regulatory-minimum capture. After five years of accounting software capturing 5 fields per invoice, a company processing 3,000 invoices per month has 180,000 invoices in its archive — each containing roughly 30 fields that were never captured. That is 5.4 million data points locked in PDFs.
Each invoice processed at this minimum level is permanently incomplete. The missing fields cannot be recovered from the accounting system — only from the original document. Over time, as vendors retire their document portals and archived PDFs are lost or degraded, even the original document becomes unavailable. A warranty dispute from equipment purchased four years ago requires finding a specific PDF in an email archive. A contract term referenced in a 2021 invoice requires a document search that may produce no results.
Delivery references are one specific case where the accumulation becomes directly costly. Delivery references extracted from invoices cross-reference against purchase orders to confirm goods were received before payment is authorized. Standard accounting software captures neither field and cannot make the connection. Full-field extraction makes it automatic: invoice, purchase order, and goods receipt are linked in the knowledge graph at the moment the invoice arrives. Payment for goods not yet delivered is blocked before it posts.
France's B2B e-invoicing mandate — taking effect progressively from 2025, with full implementation required by 2027 — will significantly increase the volume of structured invoice data entering finance systems. The organizations that build full-field extraction before that mandate expands will have a complete dataset. Those that wait will have more incomplete records at higher volume.
---
Four Questions to Test Your Current System
The gap between what accounting software captures and what invoices contain is personally testable before the end of this week.
Ask your current accounting system: "Is the server purchased in March 2023 still under warranty?" If the answer requires opening a PDF, the serial number was never captured.
Try this one: "Which of our current supplier invoices have early-payment discount clauses?" If the answer requires a manual review of invoice documents, payment term data was never stored as a queryable field.
Then ask: "Which invoices in the last 12 months reference contract number C-2024-117?" If the answer is "we don't have that data," contract references were processed and discarded.
Last one: "Which invoices received in October have delivery references that do not match a confirmed goods receipt?" If this question requires a multi-system project to answer, delivery reference fields were never captured.
Any question on this list that requires manual PDF retrieval reveals a capture gap. Those gaps accumulate with every invoice your organization processes — until the extraction layer is built to close them.