Sections09
Reading time8 min
EvidencePractical OCR guide
Employee expense capture
Bookkeeping from paper receipts
Item-level purchase analysis

Practice with receipt ocr files

How Receipt OCR works

Dynamite Docs logoDynamite Docs
The handoff
Receipt OCR: Receipt detailsNormalize the photo. Read merchant and transaction data. Separate item rows. Check the final charge. Illustrative example, not extracted customer data.FROM SOURCE FILE TO A RESULT YOU CAN USEReceipt OCRReview workspaceReceiptdetails1Normalize thephoto2Read merchant andtransaction data3Separate itemrows
  1. 01Normalize the photo
  2. 02Read merchant and transaction data
  3. 03Separate item rows
  4. 04Check the final charge
Small print. Important numbers. Illustrative example.
01

Normalize the photo

Correct perspective, orientation, contrast, and background clutter without cropping the edges.

02

Read merchant and transaction data

Capture store name, location, date, time, receipt number, and payment details.

03

Separate item rows

Group abbreviations, quantities, discounts, and prices into one row per purchase.

04

Check the final charge

Resolve subtotal, tax, service charge, tip, and total, including handwritten additions.

Structured output

Example output

Dynamite Docs logoDynamite Docs
Inside the output
Receipt OCR: Receipt detailsMerchant: Harbor Street Market. Date and time: 2026-08-28 18:42. Items: Item, quantity, unit price, amount. Illustrative example, not extracted customer data.Source-linked field inspectorReceipt OCRHarbor Street Market2026-08-28 18:42Item, quantity, unit price…MerchantHarbor Street MarketDate and time2026-08-28 18:42ItemsItem, quantity, unit price,amount
Merchant
Harbor Street Market
Date and time
2026-08-28 18:42
Items
Item, quantity, unit price, amount
Small print. Important numbers. Illustrative example.
Field or structureExample
MerchantHarbor Street Market
Date and time2026-08-28 18:42
ItemsItem, quantity, unit price, amount
Tax and tip3.86 / 8.00
Total60.06 USD
Payment methodVisa ending 1042
Review rules

Check the digits that change the booked expense

A single faded decimal or handwritten tip can change the booked expense.

Check handwritten tips and totals against the card charge.
Confirm the date, currency, and merchant before categorization.
Watch for returns, coupons, discounts, and negative line items.
Keep personal card details masked in stored output.
Use the full-resolution image instead of a compressed chat screenshot.

Implementation guide

Test receipt OCR on the images people actually submit

A clean flatbed scan is a weak test if employees usually upload phone photos. Include glare, curled paper, long receipts, faded thermal print and images taken against a cluttered background. Keep the original resolution and note which fields are unreadable to a person before judging the OCR.

Separate transaction fields from item rows in the scorecard. Merchant, date, currency and total may be enough for an expense claim, while item-level analysis needs descriptions, quantities, discounts and prices. Decide which output you need before asking the model to capture every abbreviation on the paper.

Check the charged total against the source and, where available, the payment record. Handwritten tips, returns and coupons deserve their own review rule. If the image does not support a confident reading, the safest structured value is an unresolved field, not a polished guess.

OCR API

Design the response for review.

Document extraction API guide
01

Accept image and PDF inputs, and return both transaction-level fields and an optional items array.

02

Record image-quality warnings so low-confidence photos can route to manual review instead of silently entering an expense system.

Expense reports

Pre-fill merchant, date, category inputs, tax, tip, and total from a phone photo.

Bookkeeping

Turn a folder of scanned receipts into rows that can be matched with [card transactions](/accounting/process-credit-card-statements-using-ai).

Purchase detail

Capture item rows when the expense category depends on what was bought.

Free OCR tools

Test the document you have.

Upload one supported file and review the extraction before creating an account. Pick the tool that matches the document and output.

Receipt OCR questions

Can receipt OCR read faded thermal paper?

Sometimes. Even lighting, a dark background, and the original high-resolution photo help. Very faint digits may still need manual correction.

Can it read handwritten tips?

Clear handwriting may be recognized, but the tip and final total should always be compared with the source and payment record.

Does receipt OCR identify expense categories?

OCR extracts evidence such as merchant and items. Categorization is a separate decision and should follow your chart of accounts and policy.

How does local Ollama receipt OCR work?

A compatible vision model reads the receipt image through the [local companion](/guides/local-ai). Review the merchant, date, currency, tax, tip and final charge in the workspace. Local inference cannot recover missing pixels, so keep faded amounts and unclear handwritten tips for manual checking.

Can we save optional cookies?

Essential cookies keep you signed in. With your permission, optional analytics cookies and masked session replays help us understand usage and diagnose errors. Privacy policy. Change your choice anytime in Cookie settings.