Receipt Pro uses two different kinds of intelligence, and they do different jobs. Optical character recognition (OCR) turns pixels into text. The categorisation layer then decides what that text probably means for your budget. Both run on the device.
Step one: OCR reads the paper
Apple's Vision framework finds text regions in the receipt image and recognises characters. Receipt Pro maps the result onto fields it can use: merchant, date, items, subtotal, VAT and total.
This is not a language model guessing at a story. It is text recognition followed by parsing rules, which is why a clear photo matters more than a clever prompt.
Step two: a category is suggested
Once the merchant and line items are known, Receipt Pro suggests a category. A supermarket receipt usually lands in groceries; a hardware store in home or repairs; a fuel receipt in transport. The suggestion is based on the merchant name and the words it read on the receipt.
You always see the suggestion before it is saved. If it is wrong, pick another category. The app does not silently rewrite your history.
What "on-device" means here
The receipt image, the recognised text and the category suggestion are processed locally. Receipt Pro's documentation states that receipts are not uploaded to external servers and that the App Store privacy label reports no data collection. That is the practical difference between on-device categorisation and a cloud service that receives a copy of every purchase.
Where the AI stops
Receipt Pro does not file your taxes, give financial advice or connect to your bank. It organises what you scanned and turns it into summaries, category breakdowns and trends. You remain responsible for checking the numbers that go into a report.
Why this matters for accuracy
A categorisation system is only as good as its input. A sharp, complete receipt produces better fields, and better fields produce a better category. If a receipt is faded, correct the merchant and total before saving; the category will make more sense and your monthly summary will be more honest.
