Scanning Paper Receipts With Your Phone: An Honest Guide to When It Works

Scanning Paper Receipts With Your Phone

Paper receipts are one of the reasons household budgets fall apart. They pile up in wallets, fade in the laundry, get tossed in the general direction of a drawer. By the time you sit down to log them, half are unreadable and the other half are missing.

Receipt scanning fixes most of this — but not magically. It's worth understanding how it actually works, so you know when to trust the result and when to double-check.

How Receipt OCR Works (Briefly, in Plain Language)

OCR stands for Optical Character Recognition. The app takes a photo, looks at the shapes of the characters, and converts them into text. Once it has text, it hunts for patterns that look like an amount (numbers with decimals or euro signs), patterns that look like a date, and patterns that look like a shop name (usually near the top).

Okidoky uses Tesseract.js — an open-source OCR engine that runs entirely in your browser. The photo is processed on your own device. It's not uploaded to a cloud service, not handed to a third party, and not kept as a training sample. This matters if your receipts contain things you'd rather not share: the pharmacy, the doctor, the bookshop with the embarrassing purchase.

When OCR Reads a Receipt Accurately

In the best case, you get the amount, date, and shop name all filled in correctly, and you just tap Save. This tends to work when:

  • The receipt is flat, not crumpled.
  • The photo is taken in good light — daylight or a bright overhead lamp.
  • The whole receipt fits in frame, with a bit of margin around it.
  • The text is still dark and readable — not faded thermal paper from six months ago.
  • The shop uses a clear font — most large chains do.

For typical Dutch supermarkets (Albert Heijn, Jumbo, Lidl, Dirk), receipts scanned in good light will usually come through clean.

When OCR Fails (And What It Actually Gets Wrong)

Honesty helps here. OCR is not a solved problem. Here are the failure modes you'll encounter:

Faded thermal paper. The receipts printed in small cafés, taxis, and hardware stores tend to fade fast. If you can barely read it yourself, OCR can't either. The longer a receipt sits in your wallet (heat, friction), the worse this gets.

Long receipts. Supermarket receipts that run 40 items long often don't fit in one photo. Either the top gets blurry or the bottom gets cut off. For these, you only need the total — so focus the camera on the bottom section.

Creased or folded receipts. Every fold is a line OCR has to guess through. Smooth the receipt flat before taking the photo. A kitchen counter works better than your lap.

Handwritten amounts. Some small businesses write amounts by hand on a partial printout. OCR is trained on printed text and will usually miss handwriting completely.

Multi-currency receipts. A receipt showing both euros and another currency (like on international flights) can confuse the amount detection — you may get the wrong number selected.

When one of these happens, you can edit the amount, date, or shop name by hand. It's not a failure of the workflow — it's just the part OCR couldn't do for you.

The Practical Workflow

A workable habit looks like this:

  1. Scan at the shop. Not later at home. Waiting means forgotten receipts and accumulated photos you'll eventually skip.
  2. Scan in good light. Step outside, or under a clear overhead lamp.
  3. Glance at the filled-in amount before saving. If it's wrong, it's one tap to fix. If you skip this step, small OCR errors compound silently.
  4. Throw the paper receipt out once saved. This is the actual win — you're done with it.

If you're doing a monthly catch-up instead, sit at a table in daylight, smooth the receipts, and go through them in one session. It's faster than processing them one-by-one over weeks.

Why Local OCR Matters for Privacy

A lot of receipt-scanning apps send the photo to a cloud service for processing. That's cheaper for the vendor (no local model to maintain) and it's sometimes more accurate, because the cloud models are bigger. But it also means:

  • Your receipt photo travels over the internet.
  • A third party holds a copy.
  • Health-related, pharmacy, or other sensitive purchases become part of someone else's log.

With local OCR, none of that happens. The trade-off is that local models are smaller and sometimes miss what a big cloud model would catch. For most household receipts, the difference is minor. For privacy, the difference is the whole point.

What We Don't Promise

A few things worth being clear about:

  • Receipt scanning doesn't replace bookkeeping. If you need VAT-exact records for a business, verify every amount by hand. OCR is a convenience, not an audit tool.
  • Some receipts will always need manual entry. That's not a bug. It's just that some paper receipts aren't machine-readable, no matter which app you use.
  • Quality improves with photo quality, not with more apps. If you're getting bad results with one scanner, another one probably won't help — the photo is usually the bottleneck.

The Bottom Line

Receipt scanning saves real time on most transactions and removes the "where did I put that receipt" problem. It won't read everything perfectly, and the apps that claim 99% accuracy are measuring in ideal conditions, not real wallets. Expect it to get most things right, glance at the result before saving, and accept that a minority of receipts will still need a quick manual tap.

That's the honest version. It's also the version that keeps working after the novelty wears off.

Try receipt scanning for yourself. Start free with Okidoky.