AawazAI
  1. Home
  2. Blog
  3. Scan a purchase bill, checked

Using AawazAI

Scan a purchase bill: what gets checked before it is booked

Reading a bill is the easy part. The value is in what happens after the read: is the GSTIN real, does the tax add up, is it addressed to you, have you already entered it. This is the full list of checks between a photo and a purchase entry, and what each one shows you.

General information, not professional advice. GST rules, rates and due dates change by notification. Check the current position on the GST Portal or with your CA before you act on anything here.

The short version

  • Photos, PDFs and multi-page PDFs go in. Each page is read on its own, and a PDF holding ten suppliers' bills becomes ten cards, not one merged bill.
  • A GSTIN is checked against its own check digit. A misread character is repaired only when exactly one swap fixes it, and the repair is shown, never hidden.
  • Arithmetic is checked in code: quantity times rate, tax from taxable value and rate, and the total. A printed figure the arithmetic contradicts is named, with both numbers.
  • A bill not addressed to your GSTIN can be booked as a cost, but its tax credit waits until you settle it.
  • Duplicates are caught before the read: the same file, the same bill photographed again, or a second copy inside one PDF. Nothing is saved until you confirm.

Photos, PDFs and many pages

You can photograph a bill, pick one from the gallery, or upload a PDF. A digital PDF is read as text before any image model is used. A bill that runs across several photographs can be sent as up to 10 pages of one bill, and a month of bills can go in together: up to 100 files, or one zip, each reviewed on its own.

The rule that matters for PDFs: pages are never flattened into one blob. Each page is read independently, and pages are grouped into one bill only when the document's own identifiers or page markers support it. So a PDF in which every page is a different supplier's bill becomes separate cards, and a bill with a missing page says which page is missing instead of quietly totalling what is there.

A long file is read in the background, with progress ("page 3 of 10"), so a ten-bill PDF does not time out in front of you.

The GSTIN checks

A GSTIN carries a check digit, and a misread character still matches the pattern, so the pattern is not enough. Three things happen.

  • Check digit. A GSTIN that fails it is flagged on the card.
  • One-character repair. Readers confuse I and 1, O and 0, S and 5, B and 8, Z and 2, G and 6. If exactly one such swap restores a valid check digit, that is taken as the reading and marked as a correction, with the printed reading kept as a tap-to-restore alternative. If two or more swaps work, nothing is changed and you are asked.
  • Sides swapped. On a two-column Seller and Buyer layout a reader can put the buyer's GSTIN in the supplier's slot. When the registered names of the two GSTINs match the printed names the other way round, the sides are swapped back. The names are looked up in your own parties first, then the stored registry, and the GST portal only when neither has it.

The PAN inside a GSTIN is cross-checked against the PAN printed on the bill, and a one-character difference is raised on the card.

The arithmetic

This part is pure code, never the model. A bill states the same money several times, in each line, the sum, the taxable value, the tax columns and the grand total, so one misread digit is usually outvoted by the rest. The checks:

  1. Each line: quantity times rate, less any discount, against its printed amount. A line that fails is named, with both figures.
  2. The rate: a line read as 9% on a bill that prints CGST 9% and SGST 9% is an 18% line, corrected only when the bill's own totals agree.
  3. The tax heads are worked out again from taxable value and rate, in paise, rounded half up. A printed head that arithmetic contradicts (CGST 30.15 where 9% of 3,350 is 301.50) is corrected, and the card says so.
  4. The total: when two readings of the total are possible, the one that taxable value plus tax reaches within a rupee of round-off wins.
  5. What is left. If a real gap remains, you get one plain sentence and your choices, not a silent guess. A figure you typed yourself is never overwritten.

Tax, place of supply and HSN

Line rates are checked against the slabs that applied when the bill was issued, so a bill from before 22 September 2025 at 12% or 28% is not called wrong, and the 40% slab is understood. A supplier in your own state that charged IGST is raised as a warning, because the supply is intra-state and the tax should have been CGST and SGST (IGST Act section 8, CGST Act section 9); credit on that IGST is at risk until the supplier reissues the bill. More on that in place of supply.

An HSN code or GST rate that the system suggested rather than read from the paper is held back: it cannot be written to the books as if the bill had printed it. You confirm it on the card first.

Where the bill carries an e-invoice QR code, the code's signature is verified first and the signed details outrank the printed text. A QR that does not verify is worth no more than the printed text, so a forged code pasted onto a bill gains nothing. See e-invoicing applicability for who must issue one.

Whose bill is it, and is the credit safe

Input tax credit needs the invoice in the recipient's name, and a valid supplier GSTIN (the conditions). So two more verdicts are stored on every read:

  • Addressed to you? A bill whose buyer GSTIN is somebody else's can be recorded as a cost. Its tax cannot be claimed, and the card says why.
  • Does the registry agree with the paper? A misread GSTIN that still passes its check digit belongs to somebody, possibly a different firm. When the registered name does not match the printed name, the printed name is kept, the conflict is shown, and credit waits until you settle it.

Duplicates

Re-sending a bill is normal: the WhatsApp upload failed, a family member photographed it again, the transporter's copy came too. Recording it twice doubles an expense and the credit, so duplicates are surfaced before the read.

  • The identical file is recognised by its fingerprint.
  • The same bill photographed again is recognised by a perceptual hash, which survives recompression, resizing and mild exposure change.
  • Inside one PDF, a second copy marked "Duplicate for Transporter" or "Extra Copy" is grouped with the original and nothing is booked for it.

A duplicate warning never blocks you: you decide, because sometimes a repeat is a genuinely repeated purchase.

A worked example

A Pune buyer scans Kaveri Industrial Supplies' bill (sample data): taxable ₹6,20,000, IGST at 18% ₹1,11,600. The two readers disagree on the total, 7,31,600 and 7,31,800. Taxable value plus tax is 7,31,600, so that reading wins and the card notes that it chose it. The supplier's state code 29 differs from the buyer's 27, so IGST is the right head. The GSTIN passes its check digit and is addressed to the buyer. The HSN codes were printed, not suggested. Result: a clean card, one tap to confirm, and a balanced entry.

Change one fact, say the buyer's GSTIN printed is another firm's, and the same bill comes back with the cost booked and the ₹1,11,600 credit held, with the reason in plain words.

What a scan costs

Scans use credits, charged by what the AI reads, and the Credits screen shows what each one used. Sending the exact same file again is free; a new photo of the same bill is a new read. A page that could not be read is not charged. Every business starts with 50 credits (terms apply), and typing a bill in by hand costs nothing. In our own test on 33 real supplier bills (50 scanned pages) on 13 and 14 October 2026, every bill total was read right, and you still check each line before it is saved. Details are on the pricing page.

Photograph one bill and read the card

Take the next supplier bill, photograph it, and see what the checks found before you confirm.

Frequently asked questions

Does it read multi-page PDF invoices?

Yes. Each page is read on its own and pages are grouped into one bill only when the document's own identifiers or page markers support it. A PDF of ten different suppliers' bills becomes ten cards, and a missing page is named.

What if the GSTIN is read wrongly?

The check digit catches most misreads. When exactly one confusable swap, such as O for 0 or S for 5, restores a valid GSTIN, it is applied and marked as a correction with the printed reading kept. If several swaps would work, nothing is changed and you are asked.

Will it book the same bill twice?

Duplicates are flagged before the read: the same file, the same bill photographed again, and a second copy inside one PDF. The warning never blocks you, so you can still record a genuine repeat purchase.

Is the HSN code checked?

An HSN code or GST rate that was suggested rather than printed on the bill is held back until you confirm it, so it is never written to the books as if the bill had said so.

What if the bill is addressed to someone else?

It can be recorded as a cost, but the input tax credit on it waits, and the card tells you the buyer GSTIN is not yours. Credit needs the invoice in the recipient's name.

Can it save a bill without my confirming?

No. Extraction produces a card. The purchase, stock and supplier balance are written only after you confirm.

Written by the AawazAI team. AawazAI is built by GenCrafter Technologies Private Limited: the AI accountant for Indian businesses of every size, and a filing workspace for chartered accountants. We write about the parts of GST and bookkeeping our users ask us about, and we say plainly what the product does and does not do. Corrections are welcome at admin@aawazai.com.