Receipt processing that survives real receipts

Receipts are the messiest documents in any business: crumpled paper, overexposed phone photos, item lines that outnumber the totals, gratuity hiding under the subtotal. Dynamite Docs reads them anyway, hands you confidence-scored rows, and remembers your fixes for the next pile.

Try this workflow free or view pricing.

Receipts are high-volume, low-structure, and always urgent

A week of employee spending is dozens of receipts in different currencies, states, and legibility.
Merchants, dates, and amounts are the only fields that matter, but they are the exact fields smudged, faded, or truncated on real paper.
Itemized receipts mix product lines with the totals, and tips break naive "sum to total" checks.
Manually keying receipts into an expense system is the same retyping work, just smaller and more frequent.

How Dynamite Docs handles receipt processing

01 — Snap it, scan it, or drag it in

Paper, phone photos, email attachments, and PDFs all work. Receipts without a text layer are read by the vision model you choose.

02 — The schema is inferred

Merchant, date, time, item lines, subtotal, tax, gratuity, total, and payment method are pulled from each layout without a template.

03 — Verify the amount, skip the rest

Confidence scores flag exactly the fields worth a human glance, usually the total and the category. Fix them once and the pattern sticks.

04 — Export or push

Send clean rows to Excel, CSV, JSON, or Sheets, or push them into your expense or accounting system with the API and webhooks.

Document types this workflow handles

Thermal paper receipts — including faded ink and curled paper
Receipt photos — phone photos at any angle or light level
Receipt PDFs & email attachments — e-mailed and scanned receipts
Itemized receipts — line items, subtotal, tax, and gratuity

What gets extracted from a receipt

Inferred per layout. A typical receipt run yields:

Merchant — Joe's Diner #14
Date — 2026-08-04
Time — 12:41 PM
Item lines — 2 × Breakfast burrito
Subtotal — $31.80
Tax — $2.54
Gratuity — $6.00
Total — $40.34
Payment method — Visa ····2104
Currency — USD

A realistic example

A mixed batch of business receipts extracts to rows like these, ready for the expense report:

Merchant	Date	Category	Total	Currency	Payment
Joe's Diner #14	2026-08-04	Meals & entertainment	$40.34	USD	Visa ····2104
Uber BV	2026-08-04	Travel	$18.20	EUR	Corporate card
STAPLES	2026-08-05	Office supplies	$63.97	USD	Amex ····8012

What good receipt processing looks like in practice

Receipts arrive in every medium: thermal paper that has faded in a glovebox, a photo taken in bad light at a client dinner, an email attachment from a conference hotel. The merchants repeat, coffee shops, ride-share, airlines, office supply, and the amounts are small, which is exactly why the pile is so easy to defer and so costly to process late. Deferred receipts become missed reimbursements, angry employees, and an expense close that starts from a shoebox.

The realistic fix is a vision model that reads the receipt wherever it sits, plus a review pass that only looks at the totals and categories. Once a merchant is categorized, that assignment is remembered, so the third coffee-shop receipt in a month needs no review at all. The output is a clean expense row set that exports to your expense system, your books, or the reimbursement run. A single team member can clear a week of receipts in a lunch break once the patterns are set.

Begin with the receipts you process most, meals, travel, office supplies, and let the remembered patterns cover the many small suppliers. Teams that do this get reimbursement cycles down from weeks to days, because the slow part stops being transcription and becomes only the approval it should be.

Why bring-your-own-AI matters for receipts

Receipt reading is a vision-model tradeoff. Better models read worse ink but cost more per image. Dynamite Docs lets you pick the model for the batch, your cheap key for the clean scans, your best vision key for the crumpled ones.

Receipts can be the most sensitive documents in the pile, corporate cards, per-diems, private spending. With your own key on a no-training provider, or fully local via Ollama, the image never leaves your control.

Your categorization and merchant fixes are remembered, so the model gets better at your company’s categories specifically, intelligence you keep, not a vendor’s.

Questions, answered

Can it read a photo taken at a bad angle?

Vision models handle real-world photos well, including angle and lighting problems. You pick the model. Bring a strong vision key for the worst receipts or use the hosted models on your plan.

Does it handle itemized receipts with tips?

Yes. Item lines, subtotal, tax, and gratuity are extracted as separate fields so the total checks out against the sum.

Multi-currency receipts?

Currency is extracted per receipt. Export rows as-is or convert in your spreadsheet before importing.

Is this different from your invoice and expense-report pages?

Receipts are single, low-structure documents, no PO, no GL coding, no report. For the full employee expense-report cycle, see expense report processing.