Receipt and expense report processing
Tavnit reads receipts — including crumpled ones photographed in bad light — and returns merchant, date, amount, tax and category as typed fields. Expenses arrive categorised and checkable, so approving a report is a review rather than a transcription job.
Why this is painful
Expense processing fails at both ends. Employees put off submitting because filling the form is tedious, and finance then re-types what was submitted because the form was filled inconsistently. Reimbursements slip, and nobody enjoys any part of it.
Receipts are also the worst-quality documents most companies process: thermal paper that has faded, photographed at an angle, folded, sometimes partly in another language.
What to extract
| Field | Why it needs care |
|---|---|
| Merchant name | Often a stylised logo rather than text — one of the harder fields on a receipt. |
| Transaction date and time | Time matters more than people expect for per-diem and duplicate detection. |
| Total, tax and tip | Separating tax matters for reclaim; separating tip matters for policy. |
| Payment method and last four digits | What lets you reconcile against a card statement automatically. |
| Expense category | Best inferred by an AI category Cleaner from merchant and line items rather than asked of the employee. |
| Currency | Travel expenses arrive in whatever currency was spent; conversion belongs in the pipeline. |
What makes expense reports hard
Receipt quality is genuinely bad
Faded thermal paper, glare, folds and angles are the norm rather than the exception. Extraction handles most of it, but this is a document type where conditional review earns its place: flag low-confidence or out-of-policy amounts for a human and let the clean ones pass.
Categorisation is the actual work
Reading the total is easy. Deciding whether a restaurant charge is client entertainment or a team meal is the part that consumes finance time. An AI category Cleaner assigns a category from the merchant and line items, so the employee is not guessing and finance is correcting rather than classifying.
Duplicates are common and expensive
The same receipt submitted twice — once photographed, once as a PDF from the merchant — is a routine source of over-reimbursement. Extracting merchant, exact amount and timestamp gives you enough to detect it in the Bucket rather than discovering it in an audit.
How the pipeline handles it
- Email Integration — Employees forward receipts to an address instead of learning an app. Digital receipts can be auto-forwarded on arrival.
- Cleaners — Assigns a category by AI, converts foreign currency and standardises merchant names.
- Human in the Loop — Conditional review — only expenses over a threshold or outside policy reach an approver.
- Buckets — Expenses as a table you can group, chart and check for duplicates before paying.
Common questions
- Can it read a photo of a crumpled receipt?
- Usually. Photos, faded thermal paper and folded receipts all go through OCR, though quality sets the ceiling — which is why conditional review exists for the ones the pipeline is least sure about.
- Can expenses be categorised automatically?
- Yes. An AI category Cleaner assigns a category from the merchant and line items, so finance corrects the occasional edge case rather than classifying everything.
- What about foreign currency on travel expenses?
- Currency is its own extracted field, and a conversion Cleaner can restate amounts in your reporting currency during processing.
Other document types
Try it on one document
Build a flow, send a real expense report through it, and see what comes back. Free credits to start.
Start free