Extract data from VAT and GST returns

A VAT or GST return is the periodic filing a business sends to the tax authority to declare the tax it charged on sales, the tax it paid on purchases, and the net it owes or reclaims. Finance teams at a company registered for VAT in Germany file quarterly, reporting output tax collected on invoices raised, input tax paid to suppliers, and the difference that settles with the Bundeszentralamt für Steuern. Groups operating across the EU file in several member states at once, each with its own rate, its own deadline, and its own form, while a UK business sends a return to HM Revenue and Customs every quarter under Making Tax Digital. The headline is always the same three numbers: total output tax, total input tax, and the net VAT due. The detail underneath is where a return stops being simple. Sales split into standard-rated, zero-rated exports, and exempt supplies, and only some of them carry output tax. Purchases split into domestic input tax, intra-community acquisitions from other member states, and import VAT paid at the border, and not all input tax is recoverable. One reverse-charge supply moves the liability to the customer, so it shows in two places at once. Prepayments and prior-period corrections adjust the balance before it settles. Registration numbers such as DE123456789 anchor the filing, and the tax period from 2026-04-01 to 2026-06-30 fixes the window that has to reconcile. Talonic reads the return and returns the taxpayer identity, the period, the sales and purchase summaries, and the output, input, and net figures as structured fields. Output tax of €91,200 less recoverable input tax of €34,580 leaves a net €56,620 due, checked against the return so a filing that does not reconcile is caught before it is submitted.

What gets extracted from VAT and GST returns

How extraction works for VAT and GST returns

VAT and GST returns arrive as portal PDFs from filing platforms, exports from accounting software such as DATEV, Sage, and Xero, and scans of paper submissions, and the box numbering differs by country even when the tax logic is shared. Talonic classifies each return by jurisdiction and period, then maps it to the VAT return schema in the Field Registry, which models output tax, input tax, and the sales and purchase categories rather than one country's box layout. Standard-rated, zero-rated, and exempt sales are separated, reverse-charge and intra-community lines are tagged so they are not double counted, and every amount is normalized to its ISO 4217 currency. The arithmetic is checked: total output tax less recoverable input tax is reconciled against the net VAT declared, and a return that does not balance is flagged. Each captured value returns with a confidence score and a pixel-region pointer under DIN SPEC 91491 conformity, so a tax manager can trace the net VAT due back to the box on the source return before filing.

Sample extraction

A German quarterly VAT return for Q2 2026

{
  "document_number": "VAT-2026-Q2-00417",
  "taxpayer.name": "Meridian Retail GmbH",
  "taxpayer.tax_id": "DE123456789",
  "tax_period_start": "2026-04-01",
  "tax_period_end": "2026-06-30",
  "currency": "EUR",
  "taxable_sales": 480000,
  "zero_rated_sales": 62000,
  "exempt_sales": 0,
  "total_output_tax": 91200,
  "total_input_tax": 34580,
  "intracomm_acquisitions": 28400,
  "import_vat": 4900,
  "tax_due": 56620,
  "return_type": "periodic_quarterly",
  "filing_status": "submitted"
}

Frequently asked