Guide

AI OCR for BIR Receipts and Official Invoices in the Philippines

BIR-compliant invoices carry a fixed set of fields: TIN, VAT breakdown, registered name, totals. Here's how AI OCR extracts them automatically, and what the BIR's e-invoicing mandate (EIS) means for bookkeeping by December 2026.

Axia ExtractSeptember 13, 20267 min read

AI OCR for BIR receipts reads the TIN, VAT breakdown, and total off a scanned or photographed invoice and returns them as structured data instead of a wall of text. For Philippine bookkeepers processing hundreds of receipts a month, that turns a manual keying task into a review task.

The timing matters beyond convenience. The BIR's e-invoicing system (EIS) becomes mandatory for large taxpayers and e-commerce sellers by December 31, 2026, and it expects invoices as structured, machine-readable data (Revenue Regulations No. 11-2025). Firms still keying totals by hand are building a manual step into a process the BIR is actively pushing toward automation.

Take a small distribution business issuing 40 to 60 sales invoices a day to retail buyers. Each one needs a TIN match, a VAT breakdown, and a total that reconciles before it's filed. Multiply that by a month and the bookkeeping load isn't the invoicing itself, it's re-typing numbers that were already printed correctly on the source document.

What a compliant BIR invoice has to carry

Every BIR-registered invoice or receipt carries a fixed field set, regardless of the business issuing it. Miss one and the document can't support an input VAT claim on the buyer's side.

Diagram showing the required fields on a BIR-compliant invoice: TIN and branch code, registered name and address, and VAT breakdown
Three of the six required fields. Miss one and it isn't compliant.
FieldExample formatWhy it matters
TIN + branch code000-000-000-0000Matches the seller to their BIR registration
Invoice/OR numberSequential, per BIR permitTraces the document to an accredited series
VATable salesPeso amount, ex-VATBase for the 12% VAT computation
Zero-rated / exempt salesPeso amount, if applicableExcluded from the 12% VAT line
VAT amount12% of VATable salesWhat the buyer can claim as input tax
Total amount dueSum of all lines aboveShould reconcile with the VAT breakdown

Sales invoice vs official receipt, post-EOPT

Since the Ease of Paying Taxes Act took effect, the distinction between an invoice and an OR changed for extraction purposes too. A schema built before EOPT often still treats the OR as primary, which now produces the wrong document type on the output side.

Sales invoiceOfficial receipt
Role after EOPTPrincipal document, goods and servicesSupplementary proof of payment
Supports input VAT claimYesNo, on its own
Issued atPoint of sale or billingPoint of actual payment
Required fieldsFull VAT breakdownAmount received, reference to invoice

Where the BIR's e-invoicing mandate is heading

EIS doesn't just want the invoice image on file. It wants the same fields listed above as structured JSON, submitted within three calendar days of the transaction, secured with a digital signature (EDICOM's EIS breakdown covers the technical format). That's exactly the shape AI OCR extraction already produces. The gap between "we have the paper receipt" and "we have the compliant e-invoice" is one extraction step.

Flow diagram showing a paper invoice extracted into structured JSON and filed with the BIR's electronic invoicing system within three calendar days
The same extraction step that saves bookkeeping time also produces an EIS-ready record.

First-wave coverage (large taxpayers, e-commerce sellers, and firms under the BIR's Large Taxpayers Service) hits the December 31, 2026 deadline. Smaller businesses aren't yet mandated, but the direction is set: manual invoice books are being phased toward structured, submittable data.

An online retailer already selling through Shopee or Lazada, for instance, likely falls under the e-commerce wave regardless of size. Waiting until late 2026 to build the extraction step means doing it under deadline pressure instead of on a normal implementation timeline.

What automating the read actually saves

A bookkeeper keying a typical BIR invoice, TIN, VAT lines, vendor details, total, spends roughly eight minutes per document once you count double-checking the VAT math. AI extraction returns the same fields in seconds, with the arithmetic already verified against the stated total.

Comparison chart showing manual BIR invoice entry taking 8 minutes versus AI extraction taking about 15 seconds
At 500 invoices a month, that gap is 65+ hours of bookkeeping time.

The time isn't the only reason to automate the read. A person keying numbers by hand from a faded thermal receipt makes transposition errors at a fairly predictable rate, and a transposed digit in a TIN breaks input tax matching on the buyer's end. Confidence-based routing catches that before it becomes a filing problem.

Diagram showing extracted BIR invoice fields with confidence scores, where a matching VAT computation and valid TIN auto-post and a mismatched total is flagged for review
The VAT math and TIN check out automatically. Only the mismatch needs a person.

What to check before automating BIR receipt processing

Confirm the tool recognizes both the pre- and post-EOPT document types, since a lot of receipts still in circulation predate the invoice-first rule. Confirm it validates the VAT arithmetic rather than just reading the printed total, since a printed total can itself be wrong on a poorly configured POS. And confirm it returns confidence per field, not one blended score, so a smudged TIN gets flagged instead of silently accepted.

Also worth checking: whether the tool distinguishes a BIR-accredited Computerized Accounting System (CAS) receipt from a plain thermal printout. Both carry the same required fields, but a CAS-generated invoice usually includes a machine-readable barcode or QR code the extraction step can cross-check against the printed VAT figures, an extra validation layer a hand-typed receipt doesn't offer.

For the general mechanics behind schema-based extraction and confidence scoring, see our guide to AI data extraction. If invoice processing broadly, not just the BIR-specific fields, is the immediate problem, invoice extraction software covers the workflow end to end.

FAQ

What fields does a BIR-compliant invoice need?
At minimum: the seller's TIN and branch code, registered business name and address, invoice or OR serial number, date, a VAT breakdown (VATable sales, zero-rated sales, VAT-exempt sales, and the 12% VAT amount), and the total amount due. Missing any of these makes the document non-compliant for input tax claims.
Is the official receipt still required after the EOPT Act?
The Ease of Paying Taxes Act (RA 11976) made the invoice, not the OR, the principal document for both goods and services. Official receipts still exist but now serve mainly as supplementary proof of payment rather than the primary sales document.
When does BIR e-invoicing (EIS) become mandatory?
Revenue Regulations No. 11-2025 sets December 31, 2026 as the deadline for the first wave: large taxpayers, e-commerce and internet businesses, and taxpayers under the BIR's Large Taxpayers Service. Covered transactions must be submitted as structured, machine-readable e-invoices within three calendar days.
Can AI OCR read handwritten or faded BIR receipts?
Yes, with lower confidence than a clean thermal receipt. A good extraction tool scores each field separately and flags a smudged TIN or an illegible total for review instead of guessing, which matters more here than on other document types since a wrong TIN breaks input tax matching.
Does AI OCR replace a bookkeeper?
No. It replaces the keying step, entering TIN, VAT lines, and totals into a ledger by hand, not the judgment calls: which expenses are deductible, how to classify a transaction, or what to file. Bookkeepers spend the time saved on the parts that need a person.

Keep reading

Free demo

Bring us your worst document.

A crumpled receipt, a handwritten form, a scan someone took at an angle. We'll run it live and show you the fields that come back, confidence scores and all.

  • Your own documents
  • Per-field confidence
  • No setup required