Turn a box of Arabic paper receipts into a spreadsheet you can hand to a donor

What happens to your information

What it costs

Before you start

• Synthetic receipts for the workshop and a donor spreadsheet template.
• A person authorised to classify receipts and approve any hosted route.
• A protected folder, camera backup controls and a spreadsheet that imports UTF-8.
• A second person or independent calculation for totals.

The steps

1. Step 1 / 8

   Sort every receipt before photography into public or synthetic, internal and identifiable lanes. Give each receipt a unique paper reference.

   Check it worked

   The receipt count equals the reference list, and identifiable receipts are physically separated for manual entry.

2. Step 2 / 8

   Turn off automatic photo backup and camera location recording. Photograph only the approved set, one receipt per image, and copy it to the protected folder.

   Check it worked

   No receipt photo appears in a cloud photo account, and a copied sample shows no location metadata.

3. Step 3 / 8

   Define the spreadsheet before extraction: reference, filename, date, vendor, description, currency, amount as printed, numeric amount, category, confidence and reviewer.

   Check it worked

   The template uses an explicit currency field and no formula mixes currencies.

4. Step 4 / 8

   Choose manual, hosted or verified local entry. Hosted use is limited to the approved lane and exact account. For Ollama, confirm the selected model is local, has no -cloud suffix and runs on synthetic images with the network disconnected.

   Check it worked

   The work log records route, product, model, account or device, settings, source, date and approver.

5. Step 5 / 8

   Process no more than ten images per batch. Require one row per named filename, preserve the printed amount exactly, convert digits separately and mark uncertainty instead of guessing.

   Copy this exactly

   CopyCopied

   Create CSV rows for the listed receipt images. Use only visible information. Return one row per filename in the same order. Preserve amount_as_printed exactly. Put UNKNOWN when currency or digits are unclear. Do not calculate a missing value.
   
   FILENAMES:
   [paste ordered list]
   
   COLUMNS:
   receipt_ref,image_file,date,vendor,description,currency,amount_as_printed,numeric_amount,category,confidence,review_note

   Check it worked

   Returned row count equals image count and every row carries the correct paper reference and filename.

6. Step 6 / 8

   Import the CSV as UTF-8 and compare every row with its paper receipt. Check date, vendor, currency, digit order, separators, amount, tax and category.

   Check it worked

   Every row is initialled by a reviewer, and unclear cells remain marked rather than silently corrected.

7. Step 7 / 8

   Recalculate totals independently by currency. Compare receipt count, row count, batch totals and donor-category totals. Inspect the largest and smallest values.

   Check it worked

   Independent totals match the verified rows. Differences are traced to a receipt before the batch is accepted.

8. Step 8 / 8

   Enter identifiable receipts manually, merge only verified batches, and add a paper-to-row audit log. Store the spreadsheet as a combined movement and spending record.

   Check it worked

   A reviewer can select any row and find the paper receipt, entry route, checker and correction history.