A typical Nepali trading company receives 30 to 80 supplier invoices every week. Each invoice arrives as a physical document - sometimes handwritten, sometimes a printed form with a stamp, sometimes a carbonated copy where the third sheet is barely legible. The accountant's job is to read each one, type the vendor name, invoice number, date, line items, VAT amount, and total into the ERP. It takes three to five minutes per invoice and leaves plenty of room for transcription errors.
OCR invoice processing removes most of that manual typing. You photograph the invoice with your phone or scan it on a flatbed scanner. Optical character recognition reads the document, identifies the vendor name, invoice number, date, line items, VAT amount, and total, and populates a purchase entry form with those values. You review, correct anything OCR missed, and confirm. The physical document is attached to the entry as supporting evidence. The whole cycle takes under a minute for a clean, printed invoice.
This article walks through the full OCR workflow for Nepali supplier invoices - what the technology does well, where it needs human oversight, and how to build it into your day-to-day accounts payable process.
Capture - Photograph or Scan the Invoice
The capture step happens on a phone camera or a flatbed scanner. For most Nepal SMEs, the phone camera is the practical tool - the accountant is at the counter when goods arrive, the delivery comes with a challan, and photographing it immediately is faster than queuing it for later scanning. The OCR engine is designed to handle common capture conditions: slight angle, uneven lighting from a desk lamp, slightly creased paper. For documents printed on standard paper with a laser printer - the majority of Nepal supplier invoices - recognition is reliable without any special preparation. Older thermal receipts or very faint carbonated copies benefit from a flatbed scanner with better light control.
Extraction - What OCR Reads from a Nepal Invoice
The OCR engine reads the captured image and extracts structured fields. For a standard Nepal supplier invoice, the target fields are: vendor name, vendor PAN number, invoice number, invoice date (in either BS or AD - the system recognizes both calendar formats), individual line items with quantity and unit rate, subtotal, VAT amount at 13%, and grand total. If the invoice contains TDS information - which some vendors include as a note - that field is captured too. The extraction runs in seconds and returns a draft purchase entry form populated with those values.
Nepal supplier invoices often carry dates in Bikram Sambat format - "2081 Kartik 15" or in the short form "2081/07/15". OCR systems trained on Nepali documents recognize BS date formats natively and convert them to both BS and AD in the ERP entry. Invoices printed in Devanagari script for vendor names and item descriptions require an OCR model trained specifically on Devanagari - standard Latin-script OCR engines miss these fields. Accuracy on Devanagari text has improved significantly but still benefits from human review for unusual characters or handwritten elements.
Review - The Human Accuracy Check
Every OCR output is a draft. The accountant reviews each extracted field against the physical document before confirming. Common review points: vendor name matched correctly (OCR sometimes reads partial names from stamps), invoice number read as intended (confusion between 0 and O, or 1 and l, is common in printed fonts), VAT amount matches 13% of the subtotal, and line items are complete if the invoice spans multiple pages. The review step takes thirty to sixty seconds for a clean invoice. For a messy or handwritten invoice, it may take two minutes. In both cases, that is faster than typing the entire entry from scratch.
Multi-page invoices need each page captured in sequence before OCR processes the document as a whole. Construction and import invoices sometimes run to three or four pages for a single delivery. The mobile app supports a multi-page scan session - capture page one, capture page two, capture page three, then submit the bundle for extraction. The OCR engine combines the pages and returns a single structured entry.
Posting - From OCR Draft to Confirmed Journal
After the accountant reviews and corrects the draft entry, they confirm the posting. The ERP creates the purchase voucher with the attached invoice image as supporting documentation, posts the double-entry journal to the general ledger, credits the input VAT account for the 13% amount, updates the accounts payable balance for the vendor, and adds the invoice to the VAT register in IRD format. The scanned image is stored with the transaction - if an IRD auditor requests the source document for any posted entry, the accountant can produce the invoice image from within the ERP without searching physical files. The entire sequence from photograph to posted entry takes under two minutes for a typical Nepal supplier invoice.
Exception Handling - When OCR Needs Help
No OCR system achieves 100% accuracy across all document types. Handwritten amounts on a partly-filled form, very small printed text on a thermal receipt, smudged ink on a carbonated copy - these reduce confidence. A well-designed OCR workflow flags low-confidence fields visually so the accountant knows exactly which cells to check rather than reviewing the entire form. Fields where the engine is confident show normally; fields where confidence is below the threshold are highlighted for attention. This triage approach means the accountant invests review time where it is actually needed, not uniformly across every field on every invoice.
OCR invoice processing works as a speed layer on top of human review, not as a replacement for it. The technology handles the mechanical reading and typing; the accountant handles accuracy verification and judgment calls. The combination processes a standard Nepal supplier invoice in under two minutes, with the source document stored and attached to the posted entry for audit access.
Frequently Asked Questions
Handwritten documents are the most challenging input for OCR. Printed invoices from a laser or inkjet printer achieve high accuracy. Handwritten amounts and names in Devanagari script are recognized partially - usually the structured fields like numbers, dates, and PAN numbers are captured reliably, while free-form handwritten text in Devanagari benefits from human review. For businesses that receive a high proportion of handwritten invoices - small local suppliers who write bills by hand - OCR still saves time on the numeric fields even if vendor name and item descriptions need manual correction.
IRD requires businesses to retain source documents for seven years. The digital scan stored in the ERP does not replace the legal obligation to keep the physical document during the retention period. Best practice is to scan and attach the document to the ERP entry for day-to-day access and audit convenience, while storing physical invoices in dated monthly files. For IRD e-billing registered businesses, the electronic invoice itself serves as the primary record. Check with your CA whether your specific business type and transaction volume qualifies for fully paperless records under the current IRD guidelines.
Many Nepali business documents mix Devanagari and Latin text - the company name in Devanagari, the amounts in standard numerals, and a PAN number in Latin digits. OCR engines with auto-language detection handle these mixed documents by identifying which script each region of the document uses and applying the appropriate recognition model. The fields extracted are language-neutral where possible (PAN numbers, amounts, dates) and script-specific where needed (vendor names, item descriptions). For fully bilingual invoices printed in both Nepali and English, the system picks up whichever script yields higher confidence for each field.
Scan-to-Entry for Nepali Documents, Built into the Mobile App
MISAC's scan-to-entry feature is part of the AI-First architecture and works directly from the mobile app. The accountant opens a new purchase entry, taps the scan button, photographs the supplier invoice, and within seconds the form fields populate with the extracted values. Vendor name is matched against the existing vendor master using fuzzy matching - so "Rajesh Hardwares Pvt Ltd" on the invoice matches "Rajesh Hardware" in the system without requiring an exact string match. The scanned image is attached to the entry automatically, serving as the supporting document that travels with the voucher through the approval workflow and into the audit trail.
The Mobile ERP architecture means scan-to-entry works on Android and iOS, in the field as well as in the office. A site engineer receiving construction materials at a Pokhara site can photograph the delivery challan on the spot, create the GRN entry, and the system captures the delivery record with a GPS timestamp confirming the location and time. No paper trail to lose, no delay in the accounts office waiting for the physical document to arrive from the site. The OCR engine handles both English and Devanagari text with automatic language detection.
MISAC Intelligence Pvt. Ltd. has designed the scan-to-entry flow around Nepal's actual document formats - including BS date recognition, Devanagari vendor names, and IRD-format VAT invoices. Businesses that have switched from full manual entry to OCR-assisted entry report processing the same invoice volume in roughly half the time, with errors concentrated on documents that genuinely require manual judgment rather than spread uniformly across all entries. Reach us at mis.ac to see the scan flow with your own supplier invoices.
Ready to See MISAC in Action?
See how MISAC's scan-to-entry turns your phone camera into an invoice processing tool for your Nepal accounting team.