An OCR receipt scanner that dumps text into a folder is still a shoebox. Search finds dollar signs. It does not label which number is the total, which is tax, and which line item is sitting in the way. Done is date, vendor, total, and tax as confirmed rows in a sheet you own.
Paper Home Depot tape, $156.88, May 12, crumpled in the truck. Phone snap of Blue Bottle Coffee, $18.47, April 18, still in the camera roll. Emailed Amazon Web Services invoice, $428.17, April 3, sitting in that inbox folder. An OCR dump makes those three searchable. It does not make them three ledger rows the same day.
What people mean by OCR receipt scanner
People who type "ocr receipt scanner" or "receipt scanner ocr" usually do not want an open-source engine to babysit. They want the thermal out of the glovebox, the PDF out of the inbox, and the phone photo out of the camera roll, into columns they can sort next month. Spreadsheet-allergic freelancers and one-person shops still need that file. They just do not want to rebuild Excel hell by hand every Sunday.
OCR pulls the digits off the page. AI and vision models are how modern scanners guess fields. That is the mechanism, not the story. The story ends when the row is yours: CSV, Excel, or Google Sheets, with a glance at the totals before you keep the file.
Three slips, one table
Same columns, three sources.
| Date | Vendor | Total | Tax | Was sitting in |
|---|---|---|---|---|
| 2026-04-03 | Amazon Web Services | 428.17 | 0.00 | Inbox folder |
| 2026-04-18 | Blue Bottle Coffee | 18.47 | 1.52 | Phone camera roll |
| 2026-05-12 | Home Depot | 156.88 | 12.91 | Paper in the truck |
That table is the destination. Not a rented gallery of scans. Not a searchable folder of OCR text. A blank Excel workbook or a Sheets tab named Receipts is still empty until those four columns exist.
Measured: totals can look perfect while tax is wrong
We ran twenty synthetic Q2 2026 receipt PNGs through Tesseract 5.5.0 as a dumb OCR / Drive-like baseline, then a simple heuristic parse. Not a paid document-AI dunk. Clean renders, about 9.4 seconds for the pass.
Field accuracy: vendor 95% (19/20), date 100% (20/20), total 100% (20/20), tax 60% (12/20). All four fields correct on 11 of 20 (55%).
Ground-truth Q2 sum: $2,034.43. OCR-parsed sum of totals: $2,034.43. Absolute error on the sum: $0.00.
If you only check whether the month adds up, you can feel done and still be wrong on tax, vendor naming, or which line was the real total. Several tax misses were zero-tax SaaS invoices where the parser returned blank instead of $0.00. Bookkeeping does not treat blank and zero as the same thing.
A searchable OCR folder is still ambiguous
Treat the OCR text like a Drive dump and grep for currency-looking amounts. Every receipt (20/20) had two or more. Average: 6.05 matches per slip. Corpus-wide: 121 dollar-like hits for 20 true totals. Ambiguity rate: 100%.
That is the shoebox in another costume. Forwarding slips into Drive or a Gmail label feels organized. You still decide which number is the total, which is tax, and which vendor string survived the OCR noise.
The longer lab write-up of the same corpus is OCR text is not a ledger. This page is only the OCR receipt scanner job: text dump versus confirmed rows.
What done looks like
Photograph the Home Depot tape. Snap Blue Bottle. Forward the AWS PDF. Review the three rows. Export. Open the file in Google Sheets or Excel, or keep the CSV. You own the file. The service that extracted it is optional after that.
Versus a blank spreadsheet: you are not typing totals from crumpled paper. Versus a shoebox or an inbox folder named Receipts: you are not hoping next-month-you will sort it. Versus a gallery of OCR text: the columns are already labeled.
Slipsheet Starter is $5 a month for 250 receipts. Pro is $12 a month for 1,500. Fourteen-day trial, no card. Photo, PDF, or forward to receipts@slipsheet.app, then CSV, Excel, or Sheets. You still glance at the totals. The text dump is optional.
A few straight answers
What is an OCR receipt scanner supposed to finish?
Confirmed fields: date, vendor, total, tax, in a file you can download. Reading characters off the page is the intake. The scanner job ends at the row.
Is a Drive folder of OCR'd PDFs enough?
No. Searchable text is still a pile. On that synthetic set every slip had multiple dollar-like amounts, so search alone cannot pick the total.
Can plain OCR nail the grand total and still miss tax?
Yes. In the same run, totals matched on every slip and the Q2 sum error was $0.00, while tax was right on only 12 of 20. A perfect sum is not a finished ledger.
Do paper, phone photos, and email belong in the same sheet?
Yes. Spending does not care how the proof arrived. One table, three sources, same columns.