Skip to content

AI bill scanning: how it reads a purchase bill, and what to check

Purchase bills arrive in every form except a convenient one. A photo on WhatsApp, a PDF by email, a carbon copy stapled to a challan, a thermal print already fading. Somebody then types all of it into the accounts, and every typing mistake becomes an input tax credit problem two months later.

AI bill scanning is the attempt to remove that typing step. You point a camera at the bill, or upload the PDF, and the software returns a draft entry with the vendor, the GSTIN, the taxable value, the tax split and the line items already filled in.

It works well on most bills and badly on some. This post explains what happens between the photo and the ledger entry, which fields to check hardest, and a review routine that catches errors before they reach your returns.

What happens between the photo and the ledger entry

  1. Capture. The image or PDF is taken in. A PDF generated by the supplier’s billing software already has a text layer, which is the easiest case. A photo does not.
  2. Clean up. The image is straightened, cropped to the page edges and sharpened so faint print becomes readable.
  3. Text recognition. Optical character recognition converts the pixels into characters and records where on the page each character sits.
  4. Field extraction. This is the part people call AI. A model reads the text with its position on the page and decides which number is the invoice number, which date is the invoice date, which block is the tax split, and which rows are line items. Indian bills have no standard layout, so the model works from the words around each number, not from fixed coordinates.
  5. Matching against your masters. The GSTIN and name are matched to a supplier in your books, or flagged as new. Items are matched to your item list.
  6. Arithmetic checks. Line values are added, tax is recomputed from the rate, and the total is compared with the printed one.
  7. Draft for review. A person confirms or corrects, and only then does it become a voucher.

Step seven is not optional. Everything before it is a good guess.

The fields that matter on an Indian purchase bill

Rule 46 of the CGST Rules sets out what a tax invoice must carry. Check those fields first, because they decide your credit.

Field Why it matters Where extraction slips
Supplier GSTIN Decides whether the credit exists at all and which state the supply came from 0 read as O, 5 read as S, stamps printed over the number
Invoice number Rule 46 allows up to sixteen characters; it is how the bill matches in GSTR-2B Prefixes and slashes dropped, leading zeros lost
Invoice date Fixes the tax period and the credit deadline Date of despatch or e-way bill date picked up instead
Place of supply Rule 46 requires it with the state name for inter-state supplies; it drives IGST or CGST and SGST Shipping address and billing address confused
Taxable value The base for the tax Discount rows and freight rows missed
Tax split IGST, or CGST and SGST, must be posted to the right heads Column headings misread on narrow bills
HSN or SAC Rule 46 requires the code, and it carries into your own returns Codes truncated, or a six digit code read as four
Quantity and unit Rule 46 requires quantity and the unique quantity code for goods Bags, boxes and pieces used loosely by the supplier
Invoice total The figure your payment must eventually match Round off rows of a rupee or two

If your suppliers use codes inconsistently, read our explainer on HSN codes and SAC codes before standardising your item master.

Checking the GSTIN

The GSTIN is the field to be strict about. A wrong character does not just misname a vendor, it detaches the bill from the credit.

Compare the GSTIN printed on the bill with the one saved against that supplier in your books. For a new supplier, confirm the number on the GST portal’s Search Taxpayer service before you save the master. If the supplier has more than one registration, make sure the branch that billed you is the branch you recorded.

Then check the tax heads follow from it. A supplier in another state should have charged IGST, a supplier in your own state CGST and SGST. When the tax split and the GSTIN disagree, ask for a corrected bill rather than adjust it yourself.

Where AI bill scanning gets it wrong

Handwriting and rubber stamps

Handwritten quantities, a stamped GSTIN and a signature across a figure are all difficult. Anything handwritten deserves a human look every time.

Carbon copies and thermal prints

The second and third copies of a carbon book are pale and smudged, and thermal paper fades within weeks. Recognition accuracy falls sharply. Photograph these the day they arrive, in daylight, flat on a table.

Rounding and totals

Suppliers round differently. Some round each line, some round only the tax, some add a round off row. A scan that recomputes tax from the rate often lands a rupee away from the printed total. Decide one rule for your books, apply it to every vendor, and treat a difference larger than a rupee or two as an error, not rounding.

Multi page and multi item bills

Long bills that run onto a second page lose rows when only the first page is photographed. Check the number of line items captured matches the number printed.

The same bill twice

A bill photographed by the storekeeper and emailed by the supplier becomes two entries. Matching on supplier GSTIN, invoice number and date catches almost all duplicates, so keep that check on.

A review routine before you post

  1. Open the scanned image beside the draft entry. Do not review from the extracted fields alone.
  2. Read the supplier GSTIN character by character against the master.
  3. Confirm the invoice number and date match the printed bill exactly.
  4. Check the tax heads agree with the supplier’s state.
  5. Add the line values yourself for any bill above your review threshold.
  6. Compare the computed total with the printed total, and explain any difference beyond a rupee.
  7. Confirm HSN or SAC codes are present and have the right number of digits.
  8. Look for an IRN and QR code. An e-invoice is cleaner data than a scan.
  9. Save the image with the voucher so the evidence stays attached to the entry.

The bill is only half the story

A perfectly scanned bill does not by itself give you credit. Section 16(2) of the CGST Act requires that you hold the invoice, that the details have been furnished by the supplier in his outward statement, that you have received the goods or services, that the tax has actually been paid to the government, and that you have filed your return. It also requires payment to the supplier within 180 days.

So the scan feeds a reconciliation, not a conclusion. GSTR-2B is generated on the 14th of the following month for monthly filers, covering documents filed by suppliers from the 12th of the relevant month to the 11th of the next. Since October 2024 the Invoice Management System lets you accept, reject or keep a supplier document pending, and no action counts as accepted when GSTR-2B is generated.

Watch the outer limit too. Section 16(4) bars credit on an invoice after 30 November following the end of the financial year, or the filing of the annual return, whichever is earlier. A bill scanned in April and never reconciled is a bill you may quietly lose. The way these returns lock together is covered in our post on GSTR-1, GSTR-2B and GSTR-3B.

Frequently asked questions

How accurate is AI bill scanning on a phone photo?

It depends almost entirely on the photo. A flat, well lit, in focus shot of an original print reads very well. A crooked photo of a faded carbon copy does not. Fix the photo and most accuracy complaints disappear.

Do I still need the paper bill?

Keep the original until your retention period is over. Store the scan against the voucher so the entry, the image and the reconciliation sit together when a query comes.

Can it read bills in regional languages?

Most Indian tax invoices print the GSTIN, HSN and amounts in English numerals even when the header is in another script, and those are the fields that matter. Confirm descriptive text in a regional script by hand.

What about e-invoices?

Where a supplier is covered by e-invoicing, the invoice is reported to the portal and carries an IRN and a QR code. That data is structured and need not be guessed from an image, so prefer it when available.

Where Quipu fits

Quipu AI Accounting includes AI bill scanning from camera, gallery or PDF, reading the vendor, GSTIN, amounts, tax and HSN and checking the GSTIN, then carrying the same bills through ITC reconciliation against GSTR-2A and GSTR-2B. The review routine above applies whatever software you use.

Sources

Keep reading

WhatsApp