TechnologyApr 17, 20269 min read

23 items, 1.8 seconds — what our OCR actually captured from a real wholesaler invoice

A live demo of ShelfLifePro's OCR engine on a real distributor invoice — 23 line items, handwritten batches, bilingual product names, photocopied and skewed. Here's what came back.

ShelfLifePro Editorial Team

Inventory management insights for retail and pharmacy

This is a live product demo. We ran OCR on a real distributor invoice for a supermarket client in March 2026. Vendor name is redacted to "Vendor A." Everything else — the line items, the accuracy, the time — is real.

Why retailers ask about OCR before they ask about anything else

Ask any multi-store retailer what their single biggest operational pain is, and the answer is almost never "sales." It's receiving. Somebody has to take 23 line items off a crumpled distributor invoice — SKU, batch number, expiry date, quantity, rate, tax, landed cost — and type them into the system correctly. Every mistake becomes a phantom stock count three weeks later.

Typical time to enter a 23-line GRN manually: 40-50 minutes. Typical error rate: 10-15%. Typical staff morale after 30 of these a week: terrible.

This is exactly the problem good software should eat alive. We shipped global OCR in March 2026 and have been posting weekly benchmarks on real invoices since. Here's the walkthrough for one of them.

Free Tool

Not sure how much you're losing to expiry?

Run a free inventory waste audit — find your bleeding SKUs in 60 seconds. No sign-up required.

Run free audit

The invoice

Vendor A is a mid-sized FMCG distributor covering dairy, packaged food, and personal care. The invoice in question had:

23 line items across 4 categories
Handwritten batch numbers on 8 of them (the printed ones said "see carton")
Mix of Hindi product names ("Amul Masti Dahi") and English
Two different tax lines (GST 5% and GST 18%)
A freight adjustment at the bottom
Photocopied, slightly skewed on the tray, one corner torn

In other words: the exact kind of invoice any OCR trained on clean PDFs would choke on.

Step 1 — Photo

The retailer's staff took a single photo with an entry-level Android phone. No scanner, no PDF. Just drop it in the Receiving screen on the ShelfLifePro app. Elapsed time: 4 seconds (including framing and the tap).

Step 2 — Extract

Extraction took 1.8 seconds. Here's what came back, field by field.

Invoice header (100% correct):

Supplier name (Vendor A — normalised against supplier master)
GSTIN: matched against existing supplier record
Invoice number and date
Delivery address
Subtotal, tax breakdown, total

Line items (22 of 23 correct on first pass):

SKU normalisation: 19 matched to existing catalog, 4 auto-created with confidence score
Batch numbers: 21 read correctly, including 6 of the 8 handwritten
Expiry dates: 22 read correctly, one reversed DD/MM → flagged
Quantities: 23 read correctly
Per-unit rates: 23 read correctly
GST split per line: 23 read correctly

Calculated automatically:

Landed cost per unit (rate + allocated freight / taxes — within 1 paisa of invoice total)
GST input-credit lines grouped by HSN — ready for GSTR-2B reconciliation

The one line that needed review was a handwritten batch number on a ghee SKU — flagged for human check. Staff reviewed, corrected one character, approved. Total human time: 12 seconds.

Step 3 — Auto-GRN

On approval, the system:

Created the Goods Received Note
Posted inventory updates at the batch level (not just SKU level) so FEFO rotation works correctly
Matched the GRN to the open PO (97% line-item match, remaining 3% flagged for review)
Queued a Tally voucher for the supplier's ledger
Updated the supplier's on-time performance score (this delivery was 1 day late, factored in)
Generated an e-way bill reference if applicable

Elapsed time from photo to finished GRN: 2 minutes 20 seconds. The retailer's previous workflow took a staff member 47 minutes.

The benchmark numbers

Across the last 500 distributor invoices we've run through OCR in March-April 2026:

Average line-item accuracy: 96.2%
Average extraction time: 1.9 seconds per page
Average human review time: 15-30 seconds per invoice
Auto-match to PO: 94%
Typical staff time saved per invoice: 40 minutes

A three-store supermarket receiving 15-20 invoices a day is reclaiming ~10 hours of staff time per day. That's more than the cost of the entire ShelfLifePro subscription, just on this one feature.

Where OCR breaks (the honest version)

We're not going to pretend it's perfect.

Invoices with no line-item detail (just a "supply of goods" lump sum) — we can't invent detail that isn't there. These get flagged for manual entry.
Faded thermal printouts (especially from older dot-matrix printers) — if a human can't read it, we can't either.
Fresh produce invoices with loose units (kg / pieces mixed) — we read them but the quantity normalisation needs human review.
Very long invoices (50+ line items) — accuracy holds, but the review takes longer because there's more of it.

For everything else — standard distributor invoices, photocopies, phone photos, handwritten batches, bilingual product names — OCR is already the right answer.

Where this is going

The roadmap for Q2-Q3 2026:

Multi-page invoice stitching (you photograph page 1 and page 2 separately, we merge)
PO auto-negotiation — if the line-item price is 2% over the agreed PO, flag, don't silently accept
Predictive expiry inference — if a batch number is legible but the expiry is smudged, infer the expiry from the supplier's typical shelf life for that SKU
Voice-note supplement — staff can leave a voice note flagging issues and the system parses it into the GRN review queue

How to try it

The OCR engine is included in every plan from Pro Growth (INR 2,699/month) upwards. The 14-day free trial has it enabled by default. Start a trial, photograph your next supplier invoice, and see what comes back.

Frequently Asked Questions

What was the real vendor name?

Redacted. The invoice came from a real FMCG distributor for a real supermarket client in March 2026; we don't publish client or vendor names unless explicitly approved.

Is 96% accuracy enough for compliance?

The 4% flagged lines are reviewed by staff before the GRN is finalised — net accuracy after review is ~99.5%. You get the speed benefit without the compliance risk.

Does OCR work with photocopied and handwritten invoices?

Yes. Handwritten batch numbers, photocopies, phone photos, skewed frames, and bilingual product names are all in the training set. The invoice in this demo had all of the above.

ShelfLifePro Editorial Team

The ShelfLifePro editorial team covers inventory management, expiry tracking, and waste reduction for pharmacies, supermarkets, and retail businesses worldwide.