OCR versus AI reading
Classic OCR, short for optical character recognition, does one job. It looks at an image and returns the characters it sees. For a receipt, that means a block of text: a store name, some numbers, a date in some format, words like SUBTOTAL and TOTAL, maybe a loyalty message at the bottom.
Text is not yet bookkeeping. Someone, or something, still has to decide which of the six numbers on the slip is the total, whether 08/09 means August 9 or September 8, which line is sales tax and which is a bottle deposit, and whether the name at the top is the vendor or the mall it sits in.
ReceiptQBO uses an AI model that reads the receipt as a document, the way a person does. It looks at layout and context together with the characters, so it can tell a total from a subtotal and a tip line from a tax line. The output is not a text dump. It is a set of named fields ready to post to QuickBooks Online, which you can see on the page about how we scan receipts into QuickBooks Online.
Which fields are read
Every document that arrives is read for the same set of fields, so the data is consistent across clients and channels.
- Vendor, matched to the client's existing vendor list when possible.
- Date of the purchase.
- Subtotal, sales tax, tip and total, read as separate amounts.
- Currency of the receipt.
- Payment method and the card last 4 digits, which help tie the receipt to the right card account.
- Line items, useful for split purchases and supplier bills.
- Document type: receipt, bill or credit note, because each one posts differently.
Reading the tip separately matters more than it seems. A restaurant slip often shows a total before tip and a total after tip written in by hand. Knowing which is which is the difference between a receipt that matches the card charge and one that does not. The same goes for sales tax, which many businesses need to see as its own amount.
Fwd: Your Lowe's receipt
Forwarded email, 2 min ago
IMG_4471.HEIC
Client link, 9 min ago
Invoice #88213 from Sunbelt Rentals
Gmail filter, 22 min ago
38 files from October folder
Batch upload, 1 hr ago
A confidence score on every key field
No reader is right every time, whether it is OCR, AI or a tired person on the last day of the month. What makes a scanner safe for bookkeeping is knowing how sure it is.
ReceiptQBO gives a confidence score to vendor, date, total, tax and the proposed account, separately. A receipt can be certain on the vendor and total and unsure about the tax. In review, the uncertain field is the one that stands out, so you check that field instead of rereading the whole receipt.
Each client has a threshold, 90 percent by default. Documents at or above it are approved and posted automatically. Documents below it wait in a single Review queue, lowest confidence first. For a client with clean, printed receipts you can keep the default. For a client with messy paper, you can raise it until vendor rules are in place.
Faded thermal paper, handwriting and phone photos
Real receipts are rarely clean. Thermal paper fades in a car on a hot day. Contractors write totals by hand on a supplier pad. Clients take photos at an angle, in a dim truck cab, with a thumb in the corner.
The AI model reads these cases better than character-by-character OCR, because it uses layout and context to fill in what is faint or unusual. A handwritten total next to a printed subtotal and tax, for example, can still be checked against the math on the slip. When a field is truly unclear, the confidence score drops, and the receipt goes to review instead of posting with a guess.
A practical tip for clients: a photo taken flat, in good light, with the full receipt in frame helps any reader. Encourage them to snap the receipt the day they get it, before the paper fades. Weekly missing receipt reminders help build that habit.
Receipt
Ridgeline Hardware
Oct 3, $232.59
Card ending 0981
Bank line in QuickBooks
RIDGELINE HARDWARE #12
Oct 4, $232.59
Matched, 2 day window
HEIC photos, PDFs and email bodies
File formats are a common source of friction. iPhones save photos as HEIC by default, and some tools reject them or ask the client to convert first. ReceiptQBO accepts JPG, PNG, HEIC and PDF directly, one at a time or in batches of up to 200 files.
Multi-page PDFs, such as supplier bills and utility statements, are read as one document. Many online purchases never produce an attachment at all, only an order confirmation in the email body. When such an email is forwarded to the client's receipt address, the receipt is read from the body text. Our page on how to email receipts to QuickBooks explains the forwarding options, including a Gmail filter.
Batch uploads help at the start of a new client relationship, when a box of old receipts arrives at once. Scan the stack to PDF or photograph it, upload up to 200 files in one go, and let the reading run while you work on something else. Anything uncertain lands in review with its weak field marked.
Why coding matters more than OCR
Most receipt readers today get the total right on a clean receipt. The harder and more valuable step comes next: putting the receipt in the right place in the books.
ReceiptQBO codes each receipt to the client's own QuickBooks Online lists, synced through the connection: expense account, class, location, customer or project, billable flag and memo. Vendor rules override the AI where you have a fixed answer, and any correction can be saved as a rule in one click. A duplicate check stops the same receipt from posting twice, and bank matching attaches the receipt to an existing transaction instead of creating a second expense.
That is where time is saved. Reading a receipt in two seconds instead of thirty is nice. Not having to choose the account, class and customer for it is what clears the pile. Try one receipt in the demo on the home page of our QuickBooks receipt scanner and look at the proposed account, not just the numbers.
Plain OCR compared with AI reading and coding
| Step | Plain OCR | ReceiptQBO |
|---|---|---|
| Output | Raw text from the image | Named fields such as vendor, date, tax, tip, total |
| Total versus subtotal | Left for a person to decide | Read from layout and context |
| Certainty | Usually none shown | Confidence score per key field |
| Faint or handwritten parts | Often dropped or misread | Read with context, sent to review when unclear |
| Account and class | Not part of OCR | Coded to the client's QuickBooks Online lists |
| Duplicates | Not checked | Flagged and not posted |