The VAT Line Is Always Missing: Why OCR Fails Where It Hurts
OCR tools love the 'Total' amount. They hate the 'Tax' amount. This gap is where you lose thousands in reclaimable cash. Here is why.
The Robot is Stealing Your Tax Refund
I recently audited a set of invoices processed by a popular “AI” tool. It had processed an Uber receipt.
- Total: €14.50 (Correct).
- VAT: €0.00 (Incorrect).
The tool missed the tiny line at the bottom that said “VAT: €0.85”. Because the VAT was zero, the accounting system booked the full €14.50 as a cost.
On one ride, this is 85 cents. On 1,000 rides a year, plus flights, plus software, plus meals… you are looking at thousands of Euros in unclaimed input tax. You are literally leaving free money on the table because your software is bad at reading.
The Audit Risk: The “Standard Rate” Trap
Worse than missing VAT is guessing VAT. Some tools see a total of €121 and automatically strip out €21 as VAT because they assume the standard rate.
But what if that invoice was for insurance (exempt)? Or a train ticket (9%)? Now you have claimed €21 from the tax authority that you were not entitled to. That is not an error; that is tax fraud. When the auditor checks the invoice image and sees “0% VAT” printed on it, but “21% VAT” in your ledger, you are in trouble.
An audit should be boring. Explaining why your robot lied to the tax office is very exciting.
The Control: The Confidence Score
We cannot trust; we must verify.
1. Demand Line Item Extraction Do not buy a tool that only extracts the “Header” (Date, Vendor, Total). You need Line Item Extraction. The tool must read the table. It must see that Line 1 is “Book (9%)” and Line 2 is “Shipping (21%)”.
2. The “Math Check” Logic Good software has built-in logic.
- Does Net + Tax = Total? If the OCR reads the numbers but the math doesn’t add up, it should flag it for human review. It should never auto-post a mathematically impossible invoice.
3. The Spot Check Filter your ledger for “General Expenses” with “0% VAT”. Open 10 random PDFs. Check the footer. If you find VAT on the PDF that is not in the ledger, your tool is failing. Tune it or trash it.
Summary
The “Total” tells you what you spent. The “VAT” tells you what you get back. In my world, the refund is just as important as the expense.
Don’t let lazy software steal your tax credits. Meten is weten. Check the tax line.
FAQs
Why is VAT extraction so hard?
Because every vendor hides it differently. Some put it at the bottom, some in the middle, some call it 'Tax', some 'MwSt', some 'IVA'. Robots struggle with context.
Can't we just calculate 21% of the total?
Absolutely not. Coffee is 9%. Flights are 0%. Alcohol is non-deductible. If you assume 21%, you are lying to the government.
What feature should I look for?
Line Item Extraction. The tool must read every line, not just the footer. If it can't read the lines, it can't verify the tax.