How Taggun Reads Documents

Optical Character Recognition (OCR) converts document images into machine-readable text — but text alone isn't extraction. Taggun layers AI models over OCR to understand what the text means and return structured, ready-to-use fields. This page explains how, and how accuracy works in practice.

From Image to Structured Data

The following is a conceptual processing flow. Internal stages, ordering, and techniques may change and are not part of the API contract:

1. Input preparation — the submitted image or PDF is prepared for recognition.
2. Text recognition (OCR) — visible document content is converted into machine-readable text.
3. Field interpretation — models use context to identify candidate totals, products, dates, merchant details, and tax identifiers. Taggun processes supported documents in 205 languages.
4. Data structuring — interpreted values are organised into documented response fields such as line items, subtotals, taxes, and merchant details.
5. Optional processing — dates and currencies may be standardised; documented tax-ID format or checksum checks may run where supported; and enabled features such as fraud signals, categorisation, or campaign rules may be evaluated.
6. Response — the endpoint returns its documented JSON response shape. See Understanding Your Results for how to read it.

Conceptual flow: image to structured data

This diagram summarises the relationship between a submitted document, recognised text, and a structured API response. It is conceptual: the current endpoint schemas, not labels or values in the artwork, define the public contract.

Conceptual document-processing flow; labels and example values in the artwork are illustrative, and endpoint schemas remain the public contract.

Why Receipts Are Hard

Receipts and invoices are among the messiest documents in OCR:

• Endless layout variety — every merchant formats differently, in potentially over 200 languages.
• Poor image quality — crumpled, faded, or badly lit photos from real users.
• Long documents — supermarket receipts with dozens of line items must stay accurate to the last line.
• Ambiguity — is 03/04 March or April? Contextual analysis resolves what raw OCR can't.
• Handwriting — handwritten elements need dedicated recognition models.

Receipt and Invoice Processing

Taggun processes submitted receipts and invoices and returns the documented extraction fields. If a result is incorrect, follow Contacting Support.

Accuracy in Practice

Taggun processes millions of scans per month at 90%+ accuracy. Two things matter more than any headline number:

1. Accuracy is use-case specific. Your document mix — your countries, merchants, and document types — determines your real accuracy. The way to know is to test with your own receipts: see Test With Your Own Receipts. For volume evaluations, we run pilots against your historical documents and share the results for your review.
2. Read response metadata in context. See Understanding Your Results.

Customisation & Performance Tuning

Account-specific performance tuning for selected document types is available on Advanced and Enterprise plans. Contact [email protected] to discuss your document set and the options currently available.

Need something not covered here? Contact [email protected] — see Contacting Support.