You scan an invoice. The OCR reads every field perfectly: supplier name, date, amounts, line items. Then you spend the next ten minutes looking up GL codes, checking VAT calculations, and figuring out which cost centre it belongs to.
This is the fundamental problem with basic invoice OCR: it reads everything and understands nothing about your business.
The Gap Between Text Recognition and Business Data
Most invoice scanning tools stop at extraction. They'll pull "Office Equipment Rental - £240.00 + VAT" from a PDF, but they won't know that in your system, this maps to GL code 6300 (Equipment Hire), requires approval from the Operations Manager, and needs flagging if it exceeds your monthly budget threshold.
You're left manually applying the same business logic to every invoice:
- Converting supplier trading names to your internal codes ("SCREWFIX DIRECT LTD" becomes "SCR-001")
- Mapping generic descriptions to specific GL accounts ("Fuel" to 7200, "Office supplies" to 6201)
- Checking VAT calculations match rates and codes
- Applying approval workflows based on amounts or suppliers
- Cross-referencing purchase order numbers in various formats
Without these validation rules built into your extraction workflow, you're essentially doing data entry twice: once by the OCR, then again by you when you clean up the results.
Why VAT Validation Can't Be Optional
Here's what happened to a client last month: their basic OCR tool processed 200+ supplier invoices perfectly. Every field extracted, every amount captured. But when they ran their VAT return, Xero flagged £3,400 in discrepancies.
The problem? The OCR couldn't distinguish between:
- Standard rate supplies at 20%
- Zero-rated items that suppliers incorrectly showed as standard
- Reverse charge VAT on construction services
- Mixed-rate invoices where different lines had different treatments
They spent two days manually checking invoices against VAT calculations, correcting entries, and resubmitting their MTD return. The "automated" system created more work than manual entry would have.
This is why common OCR invoice extraction errors often centre around tax calculations - the software reads the numbers but doesn't understand the business context that determines whether they're correct.
The Supplier Recognition Problem
Every supplier has quirks. Screwfix emails come from "trade@screwfix.com" but their invoice header says "SCREWFIX DIRECT LIMITED". Your accounting system knows them as "SCR-001". The OCR tool sees three different entities.
Next month, Screwfix changes their email format. Or sends a credit note with a different layout. Or their PDF generation software puts the VAT number in a different position. Basic OCR treats this as a completely new supplier.
You end up maintaining mental notes about every supplier's format variations:
- "BT invoices sometimes split line VAT, sometimes show totals only"
- "Fuel Express rounds to nearest penny, creates 1p VAT differences"
- "City Plumbing shows labour separately, but it all goes to 5000 code"
This supplier-specific knowledge needs to be built into your extraction process, not recreated manually every time.
When Approval Workflows Break Down
Your approval process probably looks something like this:
- Under £100: auto-approve
- £100-500: department manager approval
- Over £500: finance director sign-off
- Emergency repairs: bypass normal workflow
- Certain suppliers: always require additional documentation
Basic OCR tools push everything to the same queue. You're manually sorting invoices into approval categories, checking amounts against thresholds, and routing documents to the right people. The automation stops exactly where your business rules should start.
Meanwhile, urgent invoices sit waiting for approvals they don't need, and high-value purchases slip through without proper authorisation.
The GL Mapping Nightmare
Your chart of accounts is specific to your business. "Motor expenses" might split across three codes: 7200 (fuel), 7201 (repairs), 7202 (insurance). But supplier invoices just say "Vehicle costs" or "Fleet maintenance".
Without automated mapping rules, you're constantly switching between:
- The extracted invoice data
- Your supplier's description patterns
- Your accounting system's code structure
- Any special treatments (capitalisation thresholds, department allocations)
Every invoice becomes a mini decision tree. Multiply that by 200+ invoices monthly, and you understand why invoice processing feels endless despite "automated" extraction.
How Business Rules Transform Extraction
The solution isn't better OCR - it's building your business logic into the extraction workflow. This means rules that:
Validate calculations automatically: IF VAT ≠ Net × 0.2, flag for review. IF Total ≠ Net + VAT, block processing. Catch errors before they reach your accounts.
Map data to your systems: Convert "Office Equipment" to GL 6300, "SCREWFIX DIRECT" to supplier code SCR-001. Build lookup tables that reflect your business structure.
Apply workflow logic: Route invoices over £500 to approval queues, auto-process utilities under £200, flag new suppliers for verification.
Learn supplier patterns: Remember that Supplier X always formats PO numbers as "Order: 12345" and Supplier Y includes delivery charges that need separate GL coding.
This is what the Rules Engine: applying business logic to extracted data concept delivers - validation that understands your specific requirements, not generic best guesses.
The Real Cost of Manual Correction
Let's quantify the hidden cost of "good enough" extraction:
- 15 minutes per invoice checking and correcting OCR output
- 200 invoices monthly = 50 hours of manual work
- At £25/hour for accounts assistant time = £1,250 monthly
- Plus delayed payments, approval bottlenecks, and month-end correction cycles
You're paying for automation tools while still doing manual work. The promise of straight-through processing remains exactly that - a promise.
Building Validation Into Your Workflow
Effective invoice automation requires extraction plus validation plus business rules. The system should:
- Learn your suppliers - Remember document layouts, email patterns, and data quirks
- Apply your business logic - GL mappings, approval thresholds, VAT validation
- Enforce your workflows - Route documents based on your rules, not software defaults
- Prevent errors upstream - Block incorrect data before it reaches your accounting system
This integrated approach means invoices arrive in your accounts system clean, coded, and ready for payment. No manual correction, no month-end surprises, no compliance anxiety.
To see how Harold's validation system works with supplier-specific learning and custom business rules, the difference becomes clear: extraction that understands your business, not just your documents.
The goal isn't perfect OCR - it's perfect data. That requires business rules, not just text recognition.